Privacy Issues and Pitfalls in VANET Standards
Sebastian Bittl and Arturo A. Gonzalez
Fraunhofer ESK, Munich, Germany
Keywords:
Privacy, Tracking, VANET, ETSI ITS, WAVE.
Abstract:
Wireless vehicular networks are about to enter the deployment stage in the next years with important progress
being made in Europe and the USA. Thereby, one of the core concerns is privacy of vehicles and their drivers,
especially in Europe. Prior work has regarded only a small sub-set of the information exposed by current
standards to an attacker for vehicle tracking. Thus, we take a close look on the data contained on different
protocol layers of an ETSI ITS system. We find that much data is very distinctive and can be used to identify
static vehicle parameters such as manufacturer or even model. This greatly reduces the usability of formerly
proposed cooperative pseudonym switching strategies. Many more constraints have to be applied for selecting
cooperation partners significantly reducing their availability. Therefore, current techniques cannot provide the
level of privacy defined in VANET standards. Suggestions for improving the security entity and facility layer
of ETSI ITS are given to limit the impact of the found issues.
1 INTRODUCTION
Wireless intelligent transport systems (ITS) are about
to enter the mass marking in upcoming years. Impor-
tant examples being ETSI ITS in Europe (MoU, 2011)
and WAVE in the USA (J. Harding et. al., 2014).
Thus, these systems’ security and privacy aspects are
gaining increased attention. Thereby, the possibility
to track vehicles is a core point of concern, espe-
cially in Europe (Sch
¨
utze, 2011). Many approaches
for realizing such tracking exist. Typically, such at-
tacks use the temporarily fixed pseudonym certificate,
used by vehicles to authenticate their broadcast mes-
sages. However, higher level protocol information,
e.g., identifiers or current position and velocity of a
vehicle, is regarded for this purpose as well (Gerlach
and G
¨
uttler, 2007).
Many studies have shown the possibility to track
vehicles in ITS systems based on the mentioned data
sets, e.g. (Tomandl et al., 2012). Therefore, a num-
ber of countermeasures has been published. These
include context aware pseudonym changes (Gerlach
and G
¨
uttler, 2007) and time synchronized pseudonym
switching (Wiedersheim et al., 2010). Unfortunately,
these mechanisms require the exchange of further
messages between vehicles cooperating during the
pseudonym change. This clearly increases the already
significant overhead introduced by security mecha-
nisms (see e.g., (Bittl et al., 2014)). Moreover, none
of these works studies the influence of metadata con-
tained in the security envelope of current ETSI ITS
and WAVE systems on the privacy of vehicles. Ad-
ditionally, only a fraction of the vast number of data
fields from higher level applications is taken into con-
sideration in prior work.
Our contribution focuses on the influence on the
privacy of information broadcast by vehicles in ETSI
ITS conforming VANETs. Thereby, we especially
study the metadata contained in the security en-
velope of broadcast messages apart from the used
pseudonym. Furthermore, we take a close look on
the high number of data sets used by higher level pro-
tocols regarding their possibility to ruin the privacy
efforts taken elsewhere. For the sake of compactness
we focus the study of the ETSI ITS facility layer on
Cooperative Awareness Messages (CAMs). However,
much similarity between ETSI ITS and WAVE exists
on the different protocol layers. Thus, we especially
point out the cases which also apply for WAVE based
systems.
The further outline is as follows. Section 2 re-
views related work. Afterwards, Section 3 provides
the in detail study of the impact of individual data
fields on privacy of broadcasting vehicles. In Section
4, the achieved results are used to determine a metric
for vehicle uniqueness within its vehicular environ-
ment. Finally, a conclusion is provided in Section 5
alongside with possible topics of future work.
144
Bittl S. and A. Gonzalez A..
Privacy Issues and Pitfalls in VANET Standards.
DOI: 10.5220/0005459101440151
In Proceedings of the 1st International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS-2015), pages 144-151
ISBN: 978-989-758-109-0
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
2 RELATED WORK AND
ATTACKER MODEL
Recent work on privacy in vehicular area networks
(VANETs) or intelligent transport systems (ITSs)
includes (Gerlach and G
¨
uttler, 2007; Wiedersheim
et al., 2010; Eichler, 2007; Scheuer et al., 2008;
Buttyan et al., 2009; Tomandl et al., 2012). Basi-
cally, privacy in such networks relies on a pseudonym
scheme which changes the identifiers (IDs) of a vehi-
cle (or ITS-station (ITS-S)) on all protocol layers on
a regular basis to avoid tracking (Sch
¨
utze, 2011).
There are mainly two kinds of attacks on privacy
in VANETs. Simple attacks just use identifiers like
the station ID and a very limited set of additional in-
formation about the ITS-S, typically only the vehicle
position. Advanced attacks include more context in-
formation for tracking, e.g., behavior of other vehicles
(Gerlach and G
¨
uttler, 2007; Tomandl et al., 2012).
Thereby, it has been shown that simple pseudonym
change, like in ETSI ITS and WAVE, cannot avoid
tracking. The probability of two (or even more) vehi-
cles changing their pseudonym in close vicinity just
by chance, confusing an attacker, is just to small.
Many approaches to confuse an attacker trying
to track vehicles have been proposed. These ap-
ply concepts like MixZones (Scheuer et al., 2008),
silent periods, SLOW (Buttyan et al., 2009), con-
text aware pseudonym changes (Gerlach and G
¨
uttler,
2007) and time synchronized pseudonym switching
(Wiedersheim et al., 2010) (see also (Tomandl et al.,
2012) and references within). A common requirement
of all these concepts is that vehicles must find indis-
tinguishable partners in their vicinity with whom they
cooperate to perform a secure pseudonym change. All
vehicles should change all of their identification pa-
rameters together to confuse the attacker (Sch
¨
utze,
2011). However, we show that finding such partners
is quite unlikely to happen in VANETs using current
ETSI ITS and WAVE standards.
A commonly assumed attacker model is the global
passive adversary (Tomandl et al., 2012). This passive
attacker can monitor all messages in the whole ITS
system. This model is also assumed in the following.
Even advanced attacks from prior work (e.g.,
(Wiedersheim et al., 2010; Tomandl et al., 2012))
have so far not included usage of the biggest share of
metadata from the security envelope and higher pro-
tocol level data from cyclic messages in VANETs fol-
lowing current standards. Thus, we study the usability
of these data for more advanced attacks.
Properties of the studied standards from the ETSI
ITS and WAVE families are explained in the next sec-
tion alongside with their impact on privacy aspects.
3 DATA FOR VEHICLE
IDENTIFICATION
ETSI ITS uses Cooperative Awareness Messages
(CAMs) while WAVE utilizes Basic Safety Messages
(BSMs) for the main data exchange in their VANET
systems. Therefore, our focus is on the contents of
theses messages on the different protocol layers.
When looking for possible privacy issues, the core
focus is on data which differs for different groups of
vehicles but is also constant for the individual vehicle
over a long time. One example is vehicle dimensions
which are identical for all vehicles of the same model
but different for other models with high probability.
We call that kind of data volatile constant data. In
typical traffic scenarios many different vehicle types
and models are present in the vicinity of a vehicle in-
tending to perform a secure pseudonym change. To
do so, the above described cooperative pseudonym
strategies select partners whose broadcast information
is as similar as possible to confuse the attacker. The
presence of volatile constant data clearly makes it less
probable to find such proper partners leading to possi-
bly insecure pseudonym changes. Thus, the presence
of such data should be avoided as far as possible.
Current standards bind the lifetime of MAC ad-
dress, network layer address and station ID of the fa-
cility layer to the one of the pseudonym. This means,
once the pseudonym gets changed the other identifiers
get changed, too. Therefore, an attacker cannot profit
from looking on more than one of these identifiers at
once as they all provide the same temporarily valid
information. Moreover, for the simple case of sin-
gle hop broadcast, like it is used for CAM and BSM,
the network and access layer do not add any informa-
tion to the transmitted messages which can be used to
track their senders.
In the next section metadata from the security en-
velope will be studied. Afterwards, the content of
CAMs at the facility layer will be discussed.
3.1 Metadata in Security Envelope
The security envelope is used to secure content from
the facility and network layer protocols by embedding
them into a dedicated header and trailer, each consist-
ing of different sub-parts. Thereby, content handed
over to the security entity is treated in dependency of
a so called security profile. These profiles determine
the required header fields as well as the used cryp-
tographic techniques, which can be digitally signing
and/or encryption. The definition of the security en-
velope is quite similar in ETSI ITS (103, 2013) and
WAVE (WAV, 2012).
PrivacyIssuesandPitfallsinVANETStandards
145
In ETSI ITS the sets of mandatory header fields
for security profiles CAM and Generic are subsets of
the one for DENM (used for Decentralized Environ-
ment Notification Messages (DENMs)). Thereby, the
location stamp in profiles DENM and Generic carries
the same information as the vehicle position inside a
CAM (103, 2013; 102, 2013b). Thus, this field is not
discussed separately and the reader is referred to Sec-
tion 3.2.1 for details.
We focus the further discussion on mandatory
header fields from the CAM security profile. Pri-
vacy issues resulting from such fields are more severe
than those from optional ones, as these can be simply
skipped in practical implementations. In contrast, to
fix issues regarding mandatory fields the standard has
to be changed. Furthermore, (Nowdehi and Olovsson,
2014) suggests to remove the possible inclusion of op-
tional fields. We support this proposal, as differing
combinations of data sets in the envelope by different
implementations clearly give an attacker a possibil-
ity to easily distinguish vehicles independently from
their pseudonyms.
The following sections discuss the different
header fields’ privacy implications in detail.
3.1.1 Protocol Version
The used protocol version will be constant for all
vehicles at the beginning of the deployment phase.
However, over time it is very likely that multiple ver-
sions will be present in VANETs. As this value is
constant for an individual vehicle over a long time, it
is clearly volatile constant data. Thus, the presence
of many different versions should be avoided even if
they are otherwise compatible.
3.1.2 Security Profile
The content of this data field identifies the message
type. In case all vehicles monitored by an attacker
only send the same type of message, e.g., CAMs, he
cannot discriminate the sender based on this data.
3.1.3 Signer Info
The signer information of a message may hold differ-
ent contents. Thus the available information for the
attacker differs. However, one can always uniquely
determine the signer (and sender) of the message.
Therefore, this is often called the pseudonym ID of
the sender. In both CAM and BSM the field’s content
can either be the hash of the used pseudonym certifi-
cate (PSC) or the full certificate (103, 2013; WAV,
2012). Both systems use cyclic inclusion of the full
certificate every 0.5 or 1 second, respectively. In case
of security profiles DENM and Generic the full cer-
tificate is always present (103, 2013).
In case of an included PSC the following data is
available to the receiver (103, 2013; WAV, 2012).
Signer Info of Pseudonym Certificates. A signer
info field in a PSC identifies its signer which is an
authorization authority (AA). This can be done either
by a hash digest or by the full AA certificate. Both
uniquely identify the AA. Current standards allow for
a possible multitude of such entities to exist. In prac-
tice this will be probably done by the car manufactur-
ers (OEMs). However, this leads to a privacy issue as
the signer information is volatile constant data. An at-
tacker can directly determine the OEM and use this to
distinguish PSCs and thereby vehicles. PSCs signed
by different AAs are very unlikely to be used by the
same vehicle and a vehicle will very likely use only
PSCs issued by the same AA. Obviously, vehicles of
low volume OEMs will be particularly vulnerable.
To limit the usability of the AAs identity for an at-
tacker one can think of mainly two countermeasures.
Firstly, one could increase the number of AA certifi-
cates and make a single AA use a multitude of them.
Thereby, the effort for an attacker to keep track of all
certificates would increase. However, this would sig-
nificantly increase the effort for AA certificate distri-
bution to all ITS-S for a small security gain.
Secondly, one could limit the number of AAs. An
ideal choice would be to have only one AA. This
would clearly resolve the above described privacy is-
sue completely, as an attacker cannot distinguish ve-
hicles based on their used AA anymore. To imple-
ment this, OEMs would have to cooperate and use
a common AA. As they plan to establish a common
root certificate authority (CA) for Europe, this seems
to be a usable approach. In order to limit the number
of PSCs signed by a single AA certificate, one could
significantly limit its lifetime. New ones can be de-
ployed together with PSC updates.
Additionally, one should coordinate the lifetime
of an AAs certificate with the lifetime of its issued
PSCs. Thereby, any possibility to distinguish PSCs
based on their signing AA should be ruled out. More-
over, the number of AA certificates to be stored se-
curely inside the vehicles is kept (very) low.
Validity Restriction. The mandatory validity re-
striction of PSCs is a limited validity period. It is
determined by a start and end time stamp. Both are
used with an accuracy of one second. The PSC dis-
tribution scheme described in (WAV, 2012) and (102,
2010) defines that PSCs are delivered from an AA to
an ITS-S upon request of the ITS-S. The remaining
VEHITS2015-InternationalConferenceonVehicleTechnologyandIntelligentTransportSystems
146
details are implementation specific and not covered
by the standard. However, a possible pitfall for pri-
vacy of pseudonym users exists which is caused by
the mentioned time stamps.
This issue arises from the planned way of (re-
)using PSCs in Europe. Thereby, each vehicle uses
a pool of PSCs which are (re-)used until the full pool
gets updated (Tomandl et al., 2012). The update pe-
riod will probably be in the order of months.
A different approach is described in (J. Harding
et. al., 2014) for the USA. Thereby, each PSC is only
used once and the validity period is the order of min-
utes. However, this approach introduces significant
overhead in the ITS system for PCS distribution. Ei-
ther vehicles require frequent, reliable connections to
the AA (or pseudonym certificate authority (PCA) (J.
Harding et. al., 2014)) or a huge buffer filled with
PSCs for future use. Even if the initially proposed va-
lidity period of five minutes gets doubled, this would
still require a maximum amount of 144 PSCs per day.
To protect the buffered PSCs, these have to be stored
in secure memory, e.g., inside a Hardware Security
Module (HSM). However, adding more memory to
an HSM significantly increases its price. Moreover,
many issued PSCs will stay unused as their validity
period elapses while the vehicle is not in use. One
would have to know the usage times of each vehi-
cle in advance to avoid that, which is hardly practi-
cable. Thus, the approach from (J. Harding et. al.,
2014), while providing good privacy, probably bears
too much overhead for large scale deployment.
An alternative approach for securing re-usage of
PSCs is discussed in the following.
There are mainly two approaches for PSC gen-
eration inside the AA. Either the AA generates the
PSCs upon request or the AA keeps track of the expi-
ration of its users’ PSCs to generate new ones in ad-
vance. In both cases a straight forward implementa-
tion would take the same time stamp (e.g., the current
time at the AA) and use it as the common start va-
lidity time stamp of the signed PSCs. However, this
means that all PSCs of a set delivered to an ITS-S
have a very similar (or even the same) start validity
time stamp. Thereby, making this information volatile
constant data. Furthermore, this time stamp will be
different with a very high probability for most cars as
there is no timed synchronization of PSC requests.
The PSC users have no possibility to protect them-
selves against an attacker using validity time stamps
for tracking them, as they cannot change the content
of a PSC without invalidating its signature. Therefore,
countermeasures have to be taken within AAs.
A straight forward solution would be to dis-
cretized the time stamps defining the validity period
of PSCs. For example, all PSCs issued in one month
could receive the start of this month as their start va-
lidity time stamp. The longer the discretization steps,
the more vehicles will receive a set of PSCs with the
same validity period. Thus the probability that multi-
ple vehicles with common values in these data fields
meet on the street increases removing the possibility
to distinguish them.
Subject Attribute. The subject attribute field holds
the subject type and public key of the PSC. This key is
randomly generated and the subject type is fixed for
all PSCs. Thus, there is no possibility to link PSCs
based on this data set.
Subject Info. The subject info field holds a fixed
value for all PSCs. Thus, it provides no possibility to
track vehicles.
3.1.4 Generation Time
The generation time is individual for each message.
However, the time difference between two sequential
messages is clearly defined by the standard. Neither
ETSI ITS nor WAVE define any change to the sending
interval before or after a pseudonym change.
A common assumption is that clocks of ITS-S are
well synchronized using GPS (Wiedersheim et al.,
2010). Thus, time intervals between message gener-
ation of individual cars should be quite stable. Addi-
tionally, inside a group of cars the generation times of
messages should be randomly distributed leading to
an even distribution of used time stamps. Moreover,
these time stamps are generated and transmitted with
microsecond resolution (103, 2013). Thus, collisions
in this data field confusing an attacker are unlikely.
Hence, an attacker can track vehicles just on the gen-
eration time of their messages with high probability.
In case of BSMs the sending interval is fixed. For
CAMs, it is determined by multiple parameters and
can be in the range from 1 to 10 Hz. However, the
current interval can be found in the transmitted CAM
itself (102, 2013b). Thus, the attacker can easily use
this information to avoid being confused by the vari-
able sending interval of CAMs.
Furthermore, the time step is set at the network
layer. Thus, the actual sending time being somehow
randomized by the lower layer CSMA-CA scheme
does not confuse the attacker.
We propose two solutions to overcome the de-
scribed vulnerability. Both require the cooperating
vehicles to use the same sending frequency before
and after the pseudonym change for a minimum time
span, e.g., one second. Firstly, one could reduce
PrivacyIssuesandPitfallsinVANETStandards
147
the accuracy of the generation time to the maximum
transmission interval being 100 ms for BSMs and 1s
for CAMs. The security entity does not need to deter-
mine the sequence of received messages according to
standards. Moreover, the validity time spans of PSCs
are also given with full second resolution. Therefore,
currently there is no need to use a high precision time
stamp for the generation time of type Time64 and it
should be substituted by the lower resolution Time32
type. A side effect would be to reduce the size of the
security envelope by four bytes (103, 2013).
For the second solution, immediately after the
pseudonym change the next sending must be delayed
by a random waiting time. Its length should be in
the order of the normal time difference between two
successive transmissions. For example, for BSMs it
would be between zero and 100 ms. Thereby, the at-
tacker cannot determine the next generation time and
gets confused. The impact on higher level applica-
tions should be low. From their perspective a maxi-
mum delay looks just like one missed message from
the other vehicle.
3.1.5 Message Type
The message type field holds the same information as
the security profile does. Moreover, the security entity
does not need to distinguish different message types
sharing the same security profile. Therefore, this field
should be removed as it only adds overhead to the se-
curity envelope.
3.1.6 PSC Request List
An ITS-S requests up to six unknown PSCs by using
the least three bytes of their hash values. Standard are
unclear about when to remove entries from the request
list. It should be flushed after a pseudonym change.
3.1.7 Trailer Field
There is only one type of trailer field in the standards.
It holds metadata for interpreting the digital signature
as well as the signature itself. Most parts of the trailer
are fixed and the signature of multiple messages can
only be linked together with the help of the respective
public key. Therefore, the signature does not carry
any additional privacy related information compared
to the public key in the corresponding PSC (see Sec-
tion 3.1.3 above).
Moreover, the encoding of the used ECC (elliptic
curve cryptography) point may vary in general, but
is probably constant for a particular vehicle. There
are four options for the ECC point type field in the
standard, with the core difference being enabled or
disabled ECC point compression. With both choices
used, this information is volatile constant data. In the
worst case, with only two cars in a group and both
using a different ECC point type, this information is
already enough to render any pseudonym change use-
less. Thus, the standard should only allow only one
option to be used. For other reasons to do so see
(Nowdehi and Olovsson, 2014).
3.2 Data from Facility Layer
The CAM is defined as a deeply nested data struc-
ture (302, 2013). Thereby, an ItsPduHeader and a
CoopAwareness field are present on the top level. The
simple ItsPduHeader only holds basic information
like the protocol version, message id and station id.
These fields hold the same information as their re-
spective counterparts in the security envelope. There-
fore, their impact on privacy aspects is the same as for
those data sets already described in Section 3.1.3.
The CoopAwareness field has two parts being the
current generation interval (usable by an attacker as
described in Section 3.1.4) and the CamParameters
field consisting of several different containers. These
are described in detail in the following.
3.2.1 Basic Container
The always present basic container holds the compo-
nents station type and reference position.
Station Type. The station type associates the vehi-
cle to some generic class, e.g., passenger car or light
truck. This unchanging information is clearly volatile
constant data.
Reference Position. The current position of the
ITS-S measured at the vehicle’s reference point (see
(102, 2013a)) is available in each CAM. Prior work
already showed that this information can be used
to bypass simple pseudonym changes (Gerlach and
G
¨
uttler, 2007; Wiedersheim et al., 2010). There-
fore, the advanced pseudonym switching strategies
suggested in these references should be used.
3.2.2 High Frequency Container
The high frequency container is part of every CAM.
In case of an ITS-S being a vehicle the only used
sub-part is a basic vehicle container. Parameters of
the vehicle’s current movement are given in this data
set. These include heading, speed and driving direc-
tion. All these values can be used for advanced vehi-
cle tracking (Gerlach and G
¨
uttler, 2007; Wiedersheim
VEHITS2015-InternationalConferenceonVehicleTechnologyandIntelligentTransportSystems
148
et al., 2010). However, the remaining data inside this
container has not been regarded in prior work.
Dimensions. The vehicle’s dimensions length and
width are given. According to (102, 2013b) the reso-
lution is set to 0.1 meters. This value stays constant
during one journey of a vehicle and thus it has to be
regarded as volatile constant data. It is possible that
the length of a vehicle changes from one journey to
another, e.g., by extending it with a trailer. However,
this is rare in practice especially for passenger cars.
To evaluate privacy aspects of broadcasting a ve-
hicle’s dimensions, we determined the number of
different currently sold vehicle models in Germany.
Then, we assigned them to the individual discretiza-
tion steps of vehicle length and width. We took
publicly available data from the German Kraftfahrt-
Bundesamt (Kraftfahrt-Bundesamt, 2014) to obtain
the share of different vehicle types, separated into
OEMs and their models, on the overall traffic in Ger-
many caused by new cars. Foreign cars traveling on
German roads are excluded from this data set. How-
ever, it should still give a reasonable estimate about
the distribution of models’ dimensions. Moreover, we
used public information from the 45 different OEMs
present in (Kraftfahrt-Bundesamt, 2014) to obtain the
individual dimensions of models.
We find that 73% of all vehicle models share a
common combination of width and length with at
least one other model. These cars have a market share
of 75%. Thus, for a share of 25% one can determine
the model directly given its discretized dimensions.
Even the most populated set of vehicles with length
4.3 m and width 2.0 m includes only 17% of all cars.
Thus, distribution of vehicle dimensions clearly
decreases the probability to find proper (i.e., indis-
tinguishable) partners for a cooperative pseudonym
change. Further discretization of the values to, e.g.,
0.3 m would significantly improve the situation for
many vehicles but can still not help the ones with out-
standing dimensions and/or low penetration rates.
Dynamics. The parameters longitudinal accelera-
tion, curvature (consists of curvature value and con-
fidence), curvature calculation mode and yaw rate are
included in the high frequency container. Thereby, the
curvature calculation mode is again a value which is
unlikely to change for an individual vehicle and may
differ for different vehicles. Therefore, it should be
regarded as volatile constant data.
The remaining values model a vehicle’s trajectory.
Many approaches for modeling and predicting such
trajectories exists, e.g., (Ammoun and Nashashibi,
2009; Houenou et al., 2013). In case of pure track-
ing, i.e., no realtime interaction between attacker and
vehicles, the attacker does not need to process the in-
formation in realtime. Thus, he can use computation-
ally expensive but accurate and complex movement
models. As we have seen above, the attacker can de-
termine either the vehicle type directly or a group of
possible vehicle types. This information can be used
to tune the parameters of a movement model making
it very accurate. Moreover, the prediction must only
work well for a short time span as the CAM genera-
tion rate is at most one second.
To evaluate the impact of using an advanced
movement model on the attackers ability to track ve-
hicles one should use data obtained from real test
drives instead of pure simulator output. This is be-
cause simulators like the well known SUMO use a
predefined vehicle model. Therefore, tracking these
simulated vehicles with a model which fits the one
used to generate their movement will probably yield
unrealistically high success rates. Further analysis of
this issue is beyond the scope of this work and is a
subject to future work.
Optional Data. Six more data sets may be option-
ally present in the container. Three of them (steering
wheel angle, lateral, vertical acceleration) can be used
to improve the movement model described above.
The remaining three values (acceleration control,
lane position, performance class) each describe a ve-
hicle’s feature. These can be expected to change quite
slowly, i.e., they should be regarded as volatile con-
stant data. As all these fields are optional and can be
added or removed individually, also the combination
of sent data sets may differ between vehicles. Thus,
usage of each extra value will increase the change that
a particular vehicle uses a unique set of data inside its
current vicinity. Thereby, it will strip itself from find-
ing proper partners for a secure pseudonym change.
3.2.3 Optional Containers
In addition to the basic and high frequency container,
the low frequency container is distributed cyclically,
but not in every single CAM. It contains the vehicle
role, exterior lights and path history fields. See (302,
2013) for details about inclusion rules.
In case of an uncommon vehicle role, e.g., res-
cue vehicle, the corresponding additional container is
present in the CAM. The density of such vehicles in
ordinary traffic is usually low. Thus, an attacker can
easily track them just based on the presence of their
dedicated containers in their CAMs.
PrivacyIssuesandPitfallsinVANETStandards
149
Typically, the status of exterior lights changes
slowly. Thus, it is volatile constant data.
The path history field should obviously be erased
when a pseudonym change occurs or the inclusion
rate of the container has to be such low that sequen-
tially sent values of this field cannot be linked. Oth-
erwise the attacker can simply link the pseudonyms
based on this data. However, the current standards do
not ensure such behavior.
4 VEHICLE UNIQUENESS
Secure pseudonym switching schemes from prior
work are based on the assumption that broadcast data
cannot be mapped to an individual vehicle except of
the changed identifiers. We have shown in Section 3
that this is clearly not the case due to the presence of
volatile constant data. To evaluate the impact of our
findings on vehicle privacy we introduce the metric
of vehicle uniqueness (VU). It measures how much
a particular vehicle differs from its vehicular environ-
ment regarding data observable by an attacker.
Prior work showed that tracking of vehicles be-
comes more difficult alongside with higher traffic
density and longer distances traveled during a cooper-
ative pseudonym switching maneuver (Tomandl et al.,
2012). However, this only holds in case the attacker
has no extra information for re-identification of vehi-
cles after a pseudonym change. VU is a metric for the
availability of such extra information. In case a vehi-
cle is unique inside the area of pseudonym switching
the attacker can always track it, independently of the
used pseudonym switching algorithm.
To calculate VU an exposed feature vector e
i
hold-
ing all available volatile constant data is assigned to
each vehicle. Thereby, i I relates to a particular ve-
hicle within a group of vehicles I (|I| 1) cooperating
during a pseudonym change. VU is defined by
VU = Pr{
|
{x|e
x
= e
y
;x 6= y;x, y I}
|
= 0} .
This means that VU [0;1] is the probability that
there is just one car within I having one particular ex-
posed feature vector. Such vehicles are indistinguish-
able for an attacker in regard to volatile constant data.
In the following, we take three different
pseudonym switching schemes into regard. These are
1. uncoordinated pseudonym switching (ETSI ITS
and WAVE) with |I| = 1 with high probability,
2. mix zones with |I| depending on traffic flow and
size of the mix zone and
3. silent periods with |I| depending on traffic flow
and length of silent periods.
Moreover, we include the following data into e
i
:
AA of PSCs, we assume one AA per OEM and
vehicle dimensions (see Section 3.2.2).
We assume that all cars from the same OEM use the
same encoding of ECC points (see Section 3.1.7).
Thus, this data does not influence VU in our case and
is not regarded further. The rest of the volatile con-
stant data sets from Section 3 are assumed to be iden-
tical for all cars. This leads to a best case assumption
for privacy of vehicles. Moreover, we assume that
the probability of two cars within I sharing a com-
mon value of e
i
(|e
i
| = 3) only depends on the share
of their particular model within the set of all vehicles.
We use the vehicle distribution from (Kraftfahrt-
Bundesamt, 2014) to estimate VU. Moreover, an
analysis of vehicle dimensions for the models of dif-
ferent OEMs (see also Section 3.2.2) shows that the
data included in e
i
allows to uniquely identify the
model of a vehicle, e.g., as VW Golf VII, from a sin-
gle CAM including the PSC. Thus, one can calculate
the probability to encounter a vehicle with a particular
e
i
from the mentioned vehicle distribution data set.
The number of vehicles encountered during a
pseudonym switching maneuver |I| is varied by vary-
ing the traffic flow (given in
vehicles
kilometer
) and size of mix
zones or length of silent periods, respectively. The
traffic density is varied from 16 to 45
vehicles
kilometer
per
lane following (Gerlach and G
¨
uttler, 2007) to repre-
sent low volume traffic as well as a jammed setup. We
use the parameter set from (Tomandl et al., 2012) for
the size of mix zones (25m - 400m), length of silent
periods (1250ms - 20s) and velocity range (0 - 250
km
h
). The obtained results are given in Figure 1.
1e-06
1e-05
0.0001
0.001
0.01
0.1
1
5 10 15 20 25 30 35 40 45 50 55 60
probability of being
non-unique (1 - VU)
number vehicles encountered during pseudonym change (|I|)
average
best
worst
Figure 1: Vehicle uniqueness during pseudonym change.
Thereby, the best case relates to the most common
car model. It is obviously the least unique one within
the set of all vehicles. However, only about 7.7% of
all vehicles can profit from the good results for this
model having a high chance to find indistinguishable
partners for a cooperative pseudonym change. More-
over, the worst case relates to the least common car.
One can see from Figure 1 that the value of 1VU
increases alongside with |I|. However, for an aver-
age vehicle it is very low for all regarded values of
|I|. However , combinations of high velocity and high
VEHITS2015-InternationalConferenceonVehicleTechnologyandIntelligentTransportSystems
150
traffic flow, leading to high values of |I|, rarely occur
in practice. Thus, VU will exceed 99.9% in most real
world scenarios with moderate traffic flow.
Higher values of |I| than the ones used above
would relate to unrealistically high traffic flow or ex-
tending the size of mix zones and length of silent peri-
ods to values rendering higher level ITS-applications
unusable (Tomandl et al., 2012). Calculation of VU is
independent of the pseudonym switching strategy, but
the achievable size of |I| differs. While cooperative
PSC switching strategies can adjust it, uncoordinated
ones, e.g., from ETSI ITS or WAVE, cannot do so.
The obtained results show that even without other
tracking mechanism an attacker can track a vehicle
with high probability using just a small set of con-
stant volatile data, even though the vehicle performed
a pseudonym change. This shows that the presence of
volatile constant data is able to render PSC changes
useless, as an attacker can re-identify vehicles us-
ing this data after the pseudonym change. Combin-
ing this attack with further tracking mechanisms, e.g.,
from (Tomandl et al., 2012), promises to achieve very
high tracking probabilities. Thus, the mechanisms for
avoiding volatile constant data in VANET messages
suggested in Section 3 should be used to limit the
trackability of vehicles.
5 CONCLUSIONS AND FUTURE
WORK
With upcoming deployment also privacy aspects of
VANETs gain increased attention. Therefore, we
studied the influence of information currently present
in ETSI ITS and WAVE standards on proposed pri-
vacy protecting pseudonym usage strategies.
Thereby, we find that the main requirement of
pseudonym change strategies, the cooperation of mul-
tiple indistinguishable vehicles, is unlikely to be
found in practice with current standards being in use.
Multiple suggestions have been made to improve this
situation, which require to adjust the standards.
Future work can implement the outlined tracking
mechanisms in a simulation environment to study the
influence of parameters like CAM generation rules.
REFERENCES
(2010). Intelligent Transport Systems (ITS); Security; Se-
curity Services and Architecture. V1.1.1.
(2011). Memorandum of Understanding for OEMs within
the CAR 2 CAR Communication Consortium on De-
ployment Strategy for cooperative ITS in Europe.
(2012). Draft Standard for Wireless Access in Vehicular
Environments - Security Services for Applications and
Management Messages. P1609.2, D12.
(2013a). Intelligent Transport Systems (ITS); Facilities
layer function; Facility Position and time manage-
ment. V0.0.2.
(2013). Intelligent Transport Systems (ITS); Security; Se-
curity header and certificate formats. V1.1.1.
(2013b). Intelligent Transport Systems (ITS); Users and
applications requirements; Part 2: Applications and
facilities layer common data dictionary. V1.1.1.
(2013). Intelligent Transport Systems (ITS); Vehicular
Communications; Basic Set of Applications; Part 2:
Specification of Cooperative Awareness Basic Ser-
vice. ES 302637-2, V1.3.0.
Ammoun, S. and Nashashibi, F. (2009). Real time trajec-
tory prediction for collision risk estimation between
vehicles. In IEEE ICCP.
Bittl, S., Gonzalez, A. A., and Heidrich, W. (2014). Perfor-
mance Comparision of Encoding Schemes for ETSI
ITS C2X Communication Systems. In VEHICULAR,
pages 58–63.
Buttyan, L., Holczer, T., Weimerskirch, A., and Whyte,
W. (2009). SLOW: A Practical pseudonym changing
scheme for location privacy in VANETs. In IEEE Ve-
hicular Networking Conference, pages 1–8.
Eichler, S. (2007). Strategies for Pseudonym Changes in
Vehicular Ad Hoc Networks depending on Node Mo-
bility. In IEEE Intelligent Vehicles Symposium.
Gerlach, M. and G
¨
uttler, F. (2007). Privacy in VANETs us-
ing Changing Pseudonyms - Ideal and Real. In IEEE
VTC, pages 2521–2525.
Houenou, A., Bonnifait, P., Cherfaoui, V., and Yao, W.
(2013). Vehicle Trajectory Prediction based on Mo-
tion Model and Maneuver Recognition. In IEEE
IROS, pages 4363–4369.
J. Harding et. al. (2014). Vehicle-to-Vehicle Communica-
tions: Readiness of V2V Technology for Application.
Technical Report DOT HS 812 014, Washington, DC:
National Highway Traffic Safty Administration.
Kraftfahrt-Bundesamt (2014). Neuzulassungen von Person-
enkraftwagen im August 2014 nach Marken und Mod-
ellreihen. online: 10.01.2015.
Nowdehi, N. and Olovsson, T. (2014). Experiences from
Implementing the ETSI ITS SecuredMessage Service.
In IEEE IVS, pages 1055–1060.
Scheuer, F., Pl
¨
oßl, K., and Fedrrath, H. (2008). Preventing
Profile Generation in Vehicular Networks. In IEEE
WiMob, pages 520–525.
Sch
¨
utze, T. (2011). Automotive Security: Cryptography for
Car2X Communication. In Embedded World Confer-
ence.
Tomandl, A., Scheuer, F., and Federrath, H. (2012).
Simulation-based Evaluation of Techniques for Pri-
vacy Protection in VANETs. In IEEE WiMob, pages
165–172.
Wiedersheim, B., Ma, Z., Kargl, F., and Papadimitratos, P.
(2010). Privacy in Inter-Vehicular Networks: Why
simple pseudonym change is not enough. In WONS,
pages 176–183.
PrivacyIssuesandPitfallsinVANETStandards
151