Packet hiding methods for preventing selective

Packet-Hiding Methods for Preventing Selective
Jamming Attacks
Alejandro Proaño and Loukas Lazos
Dept. of Electrical and Computer Engineering, University of Arizona, Tucson, AZ, USA
E-mail:{aaproano, llazos}@ece.arizona.edu
Abstract—The open nature of the wireless medium leaves it vulnerable to intentional interference attacks, typically referred to as jamming.
This intentional interference with wireless transmissions can be used as a launchpad for mounting Denial-of-Service attacks on wireless
networks. Typically, jamming has been addressed under an external threat model. However, adversaries with internal knowledge of protocol
specifications and network secrets can launch low-effort jamming attacks that are difficult to detect and counter. In this work, we address the
problem of selective jamming attacks in wireless networks. In these attacks, the adversary is active only for a short period of time, selectively
targeting messages of high importance. We illustrate the advantages of selective jamming in terms of network performance degradation and
adversary effort by presenting two case studies; a selective attack on TCP and one on routing. We show that selective jamming attacks can be
launched by performing real-time packet classification at the physical layer. To mitigate these attacks, we develop three schemes that prevent
real-time packet classification by combining cryptographic primitives with physical-layer attributes. We analyze the security of our methods
and evaluate their computational and communication overhead.
Index Terms—Selective Jamming, Denial-of-Service, Wireless Networks, Packet Classification.
1 INTRODUCTION
Wireless networks rely on the uninterrupted availability of
the wireless medium to interconnect participating nodes.
However, the open nature of this medium leaves it vulner-
able to multiple security threats. Anyone with a transceiver
can eavesdrop on wireless transmissions, inject spurious
messages, or jam legitimate ones. While eavesdropping and
message injection can be prevented using cryptographic
methods, jamming attacks are much harder to counter.
They have been shown to actualize severe Denial-of-Service
(DoS) attacks against wireless networks [12], [17], [36], [37].
In the simplest form of jamming, the adversary interferes
with the reception of messages by transmitting a continuous
jamming signal [25], or several short jamming pulses [17].
Typically, jamming attacks have been considered under
an external threat model, in which the jammer is not
part of the network. Under this model, jamming strategies
include the continuous or random transmission of high-
power interference signals [25], [36]. However, adopting an
“always-on” strategy has several disadvantages. First, the
adversary has to expend a significant amount of energy
to jam frequency bands of interest. Second, the continuous
presence of unusually high interference levels makes this
type of attacks easy to detect [17], [36], [37].
Conventional anti-jamming techniques rely extensively
on spread-spectrum (SS) communications [25], or some
form of jamming evasion (e.g., slow frequency hopping,
or spatial retreats [37]). SS techniques provide bit-level pro-
tection by spreading bits according to a secret pseudo-noise
(PN) code, known only to the communicating parties. These
methods can only protect wireless transmissions under the
external threat model. Potential disclosure of secrets due
A preliminary version of this paper was presented at IEEE ICC 2010 Conference.
This research was supported in part by NSF (CNS-0844111, CNS-1016943). Any
opinions, findings, conclusions, or recommendations expressed in this paper are
those of the author(s) and do not necessarily reflect the views of NSF.
to node compromise, neutralizes the gains of SS. Broadcast
communications are particularly vulnerable under an in-
ternal threat model because all intended receivers must be
aware of the secrets used to protect transmissions. Hence,
the compromise of a single receiver is sufficient to reveal
relevant cryptographic information.
In this paper, we address the problem of jamming un-
der an internal threat model. We consider a sophisticated
adversary who is aware of network secrets and the imple-
mentation details of network protocols at any layer in the
network stack. The adversary exploits his internal knowl-
edge for launching selective jamming attacks in which specific
messages of “high importance” are targeted. For example,
a jammer can target route-request/route-reply messages at
the routing layer to prevent route discovery, or target TCP
acknowledgments in a TCP session to severely degrade the
throughput of an end-to-end flow.
To launch selective jamming attacks, the adversary must
be capable of implementing a “classify-then-jam” strategy
before the completion of a wireless transmission. Such
strategy can be actualized either by classifying transmitted
packets using protocol semantics [1], [33], or by decoding
packets on the fly [34]. In the latter method, the jammer
may decode the first few bits of a packet for recovering
useful packet identifiers such as packet type, source and
destination address. After classification, the adversary must
induce a sufficient number of bit errors so that the packet
cannot be recovered at the receiver [34]. Selective jamming
requires an intimate knowledge of the physical (PHY) layer,
as well as of the specifics of upper layers.
Our Contributions–We investigate the feasibility of real-
time packet classification for launching selective jamming
attacks, under an internal threat model. We show that
such attacks are relatively easy to actualize by exploiting
knowledge of network protocols and cryptographic primi-
tives extracted from compromised nodes. We investigate the
IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING ,VOL. 9, NO. 1, JAN-FEB 2012

Preamble PHY hdr Payload
PHY
trailer
Frame
control
Source
adr.
Dest.
adr.
Seq.
number
MAC hdr
MAC
CRC
Addl.
param.
(a) (b)
Fig. 1. (a) Realization of a selective jamming attack, (b) a generic frame format for a wireless network.
impact of selective jamming on critical network functions.
Our findings indicate that selective jamming attacks lead
to a DoS with very low effort on behalf of the jammer.
To mitigate such attacks, we develop three schemes that
prevent classification of transmitted packets in real time.
Our schemes rely on the joint consideration of crypto-
graphic mechanisms with PHY-layer attributes. We analyze
the security of our schemes and show that they achieve
strong security properties, with minimal impact on the
network performance.
The remainder of the paper is organized as follows. In
Section 2, we describe the problem addressed, and state the
system and adversarial model. In Section 3, we show the
feasibility of selective jamming attacks. Section 4 illustrates
the impact of selective jamming. In Sections 5, 6, and 7,
we develop methods for preventing selective jamming. In
Section 8, we evaluate the impact of our attack mitigation
methods on the network performance. Section 9, presents
related work. In Section 10, we conclude.
2 PROBLEM STATEMENT AND ASSUMPTIONS
2.1 Problem Statement
Consider the scenario depicted in Fig. 1(a). Nodes A and B
communicate via a wireless link. Within the communication
range of both A and B there is a jamming node J. When A
transmits a packet m to B, node J classifies m by receiving
only the first few bytes of m. J then corrupts m beyond
recovery by interfering with its reception at B. We address
the problem of preventing the jamming node from classifying
m in real time, thus mitigating J’s ability to perform selective
jamming. Our goal is to transform a selective jammer to
a random one. Note that in the present work, we do not
address packet classification methods based on protocol
semantics, as described in [1], [4], [11], [33].
2.2 System and Adversary Model
Network model–The network consists of a collection of
nodes connected via wireless links. Nodes may commu-
nicate directly if they are within communication range, or
indirectly via multiple hops. Nodes communicate both in
unicast mode and broadcast mode. Communications can be
either unencrypted or encrypted. For encrypted broadcast
communications, symmetric keys are shared among all
intended receivers. These keys are established using pre-
shared pairwise keys or asymmetric cryptography.
Communication Model–Packets are transmitted at a rate
of R bauds. Each PHY-layer symbol corresponds to q bits,
where the value of q is defined by the underlying digital
modulation scheme. Every symbol carries α
β q data bits,
where α/β is the rate of the PHY-layer encoder. Here, the
transmission bit rate is equal to qR bps and the information
bit rate is α
β qR bps. Spread spectrum techniques such
as frequency hopping spread spectrum (FHSS), or direct
sequence spread spectrum (DSSS) may be used at the PHY
layer to protect wireless transmissions from jamming. SS
provides immunity to interference to some extent (typically
20 to 30 dB gain), but a powerful jammer is still capable of
jamming data packets of his choosing.
Transmitted packets have the generic format depicted
in Fig. 1(b). The preamble is used for synchronizing the
sampling process at the receiver. The PHY layer header
contains information regarding the length of the frame,
and the transmission rate. The MAC header determines
the MAC protocol version, the source and destination ad-
dresses, sequence numbers plus some additional fields. The
MAC header is followed by the frame body that typically
contains an ARP packet or an IP datagram. Finally, the
MAC frame is protected by a cyclic redundancy check
(CRC) code. At the PHY layer, a trailer may be appended
for synchronizing the sender and receiver.
Adversary Model–We assume the adversary is in control
of the communication medium and can jam messages at any
part of the network of his choosing (similar to the Dolev-
Yao model). The adversary can operate in full-duplex mode,
thus being able to receive and transmit simultaneously. This
can be achieved, for example, with the use of multi-radio
transceivers. In addition, the adversary is equipped with
directional antennas that enable the reception of a signal
from one node and jamming of the same signal at another.
For analysis purposes, we assume that the adversary can
pro-actively jam a number of bits just below the ECC
capability early in the transmission. He can then decide to
irrecoverably corrupt a transmitted packet by jamming the
last symbol. In reality, it has been demonstrated that selective
jamming can be achieved with far less resources [32], [34].
A jammer equipped with a single half-duplex transceiver is
sufficient to classify and jam transmitted packets. However,
our model captures a more potent adversary that can be
effective even at high transmission speeds.
The adversary is assumed to be computationally and
storage bounded, although he can be far superior to normal
nodes. In particular, he can be equipped with special pur-
pose hardware for performing cryptanalysis or any other
required computation. Solving well-known hard crypto-
graphic problems is assumed to be time-consuming. For the
purposes of analysis, given a ciphertext, the most efficient
method for deriving the corresponding plaintext is assumed

to be an exhaustive search on the key space.
The implementation details of every layer of the network
stack are assumed to be public. Furthermore, the adversary
is capable of physically compromising network devices
and recovering stored information including cryptographic
keys, PN codes, etc. This internal adversary model is re-
alistic for network architectures such as mobile ad-hoc,
mesh, cognitive radio, and wireless sensor networks, where
network devices may operate unattended , thus being
susceptible to physical compromise.
3 REAL-TIME PACKET CLASSIFICATION
In this section, we describe how the adversary can classify
packets in real time, before the packet transmission is
completed. Once a packet is classified, the adversary may
choose to jam it depending on his strategy.
Consider the generic communication system depicted in
Fig. 2. At the PHY layer, a packet m is encoded, interleaved,
and modulated before it is transmitted over the wireless
channel. At the receiver, the signal is demodulated, de-
interleaved, and decoded, to recover the original packet m.
Fig. 2. A generic communication system diagram.
The adversary’s ability in classifying a packet m depends
on the implementation of the blocks in Fig. 2. The channel
encoding block expands the original bit sequence m, adding
necessary redundancy for protecting m against channel
errors. For example, an α/β-block code may protect m
from up to e errors per block. Alternatively, an α/β-rate
convolutional encoder with a constraint length of Lmax, and
a free distance of e bits provides similar protection. For our
purposes, we assume that the rate of the encoder is α/β.
At the next block, interleaving is applied to protect m from
burst errors. For simplicity, we consider a block interleaver
that is defined by a matrix Ad×β
1
. The de-interleaver is
simply the transpose of A. Finally, the digital modulator
maps the received bit stream to symbols of length q, and
modulates them into suitable waveforms for transmission
over the wireless channel. Typical modulation techniques
include OFDM, BPSK, 16(64)-QAM, and CCK.
In order to recover any bit of m, the receiver must collect
d·β bits for de-interleaving. The d·β de-interleaved bits are
then passed through the decoder. Ignoring any propagation
and decoding delays, the delay until decoding the first
block of data is ⌈dβ
q ⌉ symbol durations. As an example, in
the 802.11a standard, operating at the lowest rate of 6 Mbps,
1. Without loss of generality we assume that the number of columns of
the interleaving matrix is equal to the length β of the codewords.
data is passed via a 1/2-rate encoder before it is mapped
to an OFDM symbol of q = 48 bits. In this case, decoding
of one symbol provides 24 bits of data. At the highest data
rate of 54 Mbps, 216 bits of data are recovered per symbol.
From our analysis, it is evident that intercepting the first
few symbols of a packet is sufficient for obtaining relevant
header information. For example, consider the transmission
of a TCP-SYN packet used for establishing a TCP connec-
tion at the transport layer. Assume an 802.11a PHY layer
with a transmission rate of 6 Mbps. At the PHY layer, a 40-
bit header and a 6-bit tail are appended to the MAC packet
carrying the TCP-SYN packet. At the next stage, the 1/2-
rate convolutional encoder maps the packet to a sequence
of 1,180 bits. In turn, the output of the encoder is split into
25 blocks of 48 bits each and interleaved on a per-symbol
basis. Finally, each of the blocks is modulated as an OFDM
symbol for transmission. The information contained in each
of the 25 OFDM symbols is as follows:
- Symbols 1-2 contain the PHY-layer header and the
first byte of the MAC header. The PHY header reveals
the length of the packet, the transmission rate, and
synchronization information. The first byte of the MAC
header reveals the protocol version and the type and
subtype of the MAC frame (e.g., DATA, ACK).
- Symbols 3-10 contain the source and destination MAC
addresses, and the length of the IP packet header.
- Symbols 11-17 contain the source and destination IP
addresses, the size of the TCP datagram carried by the
IP packet, and other IP layer information. The first two
bytes of the TCP datagram reveal the source port.
- Symbols 18-23 contain the TCP destination port, se-
quence number, acknowledgment number, TCP flags,
window size, and the header checksum.
- Symbols 24-25 contain the MAC CRC code.
Our example illustrates that a packet can be classified at
different layers and in various ways. MAC layer classifica-
tion is achieved by receiving the first 10 symbols. IP layer
classification is achieved by receiving symbols 10 and 11,
while TCP layer classification is achieved by symbols 12-19.
An intuitive solution to selective jamming would be
the encryption of transmitted packets (including headers)
with a static key. However, for broadcast communications,
this static decryption key must be known to all intended
receivers and hence, is susceptible to compromise. An
adversary in possession of the decryption key can start
decrypting as early as the reception of the first ciphertext
block. For example, consider the cipher-block chaining
(CBC) mode of encryption [27]. To encrypt a message m
with a key k and an initialization vector IV, message m is
split into x blocks m1, m2, . . . mx, and each ciphertext block
ci, is generated as:
c1 = IV, ci+1 = Ek(ci ⊕ mi), i = 1, 2, . . . , x, (1)
where Ek(m) denotes the encryption of m with key k. The
plaintext mi is recovered by:
mi = ci ⊕ Dk(ci+1), i = 1, 2, . . . , x. (2)

0 0.2 0.4 0.6 0.8
10
0
10
1
10
2
10
3
Jamming Probability p
E[D](sec)
TCP-ACK
RTS/CTS
Data
Random
0 0.2 0.4 0.6 0.8
10
2
10
3
10
4
10
5
E[T](bps)
TCP-ACK
RTS/CTS
Data
Random
0 0.2 0.4 0.6 0.8
10
0
10
1
10
2
10
3
NumberofPackets
TCP-ACK
RTS/CTS
Data
Random
(a) (b) (c)
0 0.2 0.4 0.6 0.8
10
−8
10
−6
10
−4
10
−2
10
0
t(normalized)
TCP-ACK
RTS/CTS
Data
Random
R. 0.3 R. 0.5 R. 0.7 R. 0.9 Sel. Con.
0
0.2
0.4
0.6
0.8
1
Routes(normalized)
R 0.3 R 0.5 R 0.7 R 0.9 Sel. Con.
10
−5
10
−4
10
−3
10
−2
10
−1
10
0
t(normalized)
(d) (e) (f)
Fig. 3. (a) Average application delay E[D], (b) average effective throughput E[T ], (c) number of packets jammed, (d) fraction
of time the jammer is active, (e) number of connections established in the network, (f) fraction of time the jammer is active.
R p: random jammer with probability p; Con.: constant jammer; Sel.: selective jammer.
Note from (2) that reception of ci+1 is sufficient to recover
mi if k is known (c1 = IV is also known). Therefore, real-
time packet classification is still possible.
One solution to the key compromise problem would
be to update the static key whenever it is compromised.
However, such a solution is not useful if the compromised
node obtains the new key. This can only be avoided if there
is a mechanism by which the set of compromised nodes can
be identified. Such a task is non-trivial when the leaked key
is shared by multiple nodes. Any node that possesses the
shared key is a candidate malicious node.
Moreover, even if the encryption key of a hiding scheme
were to remain secret, the static portions of a transmitted
packet could potentially lead to packet classification. This
is because for computationally-efficient encryption meth-
ods such as block encryption, the encryption of a prefix
plaintext with the same key yields a static ciphertext pre-
fix. Hence, an adversary who is aware of the underlying
protocol specifics (structure of the frame) can use the static
ciphertext portions of a transmitted packet to classify it.
4 IMPACT OF SELECTIVE JAMMING
In this section, we illustrate the impact of selective jamming
attacks on the network performance. We used OPNETTM
Modeler 14.5 [18] to implement selective jamming attacks
in two multi-hop wireless network scenarios. In the first
scenario, the attacker targeted a TCP connection established
over a multi-hop wireless route. In the second scenario, the
jammer targeted network-layer control messages transmit-
ted during the route establishment process.
Selective Jamming at the Transport Layer–In the first
set of experiments, we setup a file transfer of a 3 MB file
between two users A and B connected via a multi-hop
route. The TCP protocol was used to reliably transport the
requested file. At the MAC layer, the RTS/CTS mechanism
was enabled. The transmission rate was set to 11 Mbps at
each link. The jammer was placed within the proximity
of one of the intermediate hops of the TCP connection.
Four jamming strategies were considered: (a) selective jam-
ming of cumulative TCP-ACKs, (b) selective jamming of
RTS/CTS messages, (c) selective jamming of data packets,
and (d) random jamming of any packet. In each of the
strategies, a fraction p of the targeted packets is jammed.
In Fig. 3(a), we show the average delay E[D] for complet-
ing the file transfer, as a function of the jamming probability
p (averaged over repeated experiments). In Fig. 3(b), we
show the average throughput E[T ] as a function of p. It
can be observed that all jamming attacks have significant
impact on E[D] which grows several orders of magnitude
larger compared to the delay in the absence of a jammer.
Similarly, the effective throughput drops drastically under
both random and selective jamming attacks. TCP perfor-
mance under jamming of TCP-ACKs can be interpreted
by the congestion control mechanism of the TCP protocol.
When cumulative ACKs are lost (in our case jammed),
the sender has to retransmit all unacknowledged data
packets, thus increasing the incurred delay while reducing
the effective throughput. At the same time, the sender
interprets the loss of ACKs as congestion and throttles
its packet transmission rate by reducing the size of the

transmission window. This leads to a further slow down
of the application. Note that, for values of p > 0.4, the TCP
connection is aborted for the case of random and TCP-ACK
jamming, due to the repeated timeouts at the sender.
Fig. 3(c) depicts the number of packets that were jammed
by the adversary for each value of p. Finally, Fig. 3(d) shows
the fraction of time that the jammer remained active. Here,
for selective jamming attacks, we assumed that 13% of the
packet has to be corrupted in order to be dropped [17]. In
the case of random jamming, the adversary is not aware of
the type of packets transmitted (by means of processing the
header of these packets). Hence, he is assumed to jam the
entire packet in order to drop it. We observe that selective
jamming requires the jamming of approximately one order
of magnitude less packets than random jamming. This is
because, as the packet transmission rate of the sender drops
fewer packets need to be selectively targeted. Moreover,
in selective jamming, the fraction of time the adversary
remains active is several orders of magnitude less compared to
random jamming. From Fig. 3(d), we observe that targeting
control packets such as RTS/CTS messages and TCP-ACKs
yields the lowest jamming activity, because control packets
are significantly smaller compared to data packets. Such
low-effort jamming attacks are not only efficient in terms of
energy expenditure, but also challenging in localizing and
physically removing the jamming devices. Typical methods
of transmitter localization such as received signal strength
and angle of arrival measurements require that the jamming
device remains active for extended periods of time.
Selective Jamming at the Network Layer–In this sce-
nario, we simulated a multi-hop wireless network of 35
nodes, randomly placed within a square area. The AODV
routing protocol was used to discover and establish routing
paths [19]. Connection requests were initiated between ran-
dom source/destination pairs. Three jammers were strate-
gically placed to selectively jam non-overlapping areas
of the network. Three types of jamming strategies were
considered: (a) a continuous jammer, (b) a random jammer
blocking only a fraction p of the transmitted packets, and (c)
a selective jammer targeting route request (RREQ) packets.
In Fig. 3(e), we show the number of connections es-
tablished, normalized over the number of connections in
the absence of the jammers. Fig. 3(f) shows the fraction of
time that the jammer was active during our simulation, for
each jamming strategy. We observe that a selective jamming
attack against RREQ messages is equally effective to a
constant jamming attack. However, selective jamming is
several orders of magnitude more efficient as it is illustrated
in Fig. 3(f). On the other hand, random jamming fails to
disrupt the route discovery process due to the flooding
mechanism of AODV.
5 HIDING BASED ON COMMITTMENTS
In this section, we show that the problem of real-time packet
classification can be mapped to the hiding property of
commitment schemes, and propose a packet-hiding scheme
based on commitments.
5.1 Mapping to Commitment Schemes
Commitment schemes are cryptographic primitives that
allow an entity A, to commit to a value m, to an entity
V while keeping m hidden. Commitment schemes are
formally defined as follows [7].
Commitment Scheme: A commitment scheme is a two-
phase interactive protocol defined as a triple {X, M, E}.
Set X = {A, V } denotes two probabilistic polynomial-time
interactive parties, where A is known as the committer and
V as the verifier; set M denotes the message space, and set
E = {(ti, fi)} denotes the events occurring at protocol stages
ti (i = 1, 2), as per functions fi (i = 1, 2). During commit-
ment stage t1, A uses a commitment function f1 = commit()
to generate a pair (C, d) = commit(m), where (C, d) is
called the commitment/decommitment pair. At the end of
stage t1, A releases the commitment C to V . In the open
stage t2, A releases the opening value d. Upon reception
of d, V opens the commitment C, by applying function
f2 = open(), thus obtaining a value of m′
= open(C, d). This
stage culminates in either acceptance (m′
= m) or rejection
(m′
= m) of the commitment by V . Commitment schemes
satisfy the following two fundamental properties:
- Hiding: For every polynomial-time party V interacting
with A, there is no (probabilistic) polynomially-efficient
algorithm that would allow V to associate C with m
and C′
with m′
, without access to the decommitment
values d or d′
respectively, and with non-negligible
probability.
- Binding: For every polynomial-time party A interact-
ing with V , there is no (probabilistic) polynomially-
efficient algorithm that would allow A to generate a
triple (C, d, d′
), such that V accepts the commitments
(C, d) and (C, d′
), with non-negligible probability.
In our context, the role of the committer is assumed by
the transmitting node S. The role of the verifier is assumed
by any receiver R, including the jammer J. The committed
value m is the packet that S wants to communicate to
R. To transmit m, the sender computes the corresponding
commitment/decommitment pair (C, d), and broadcasts C.
The hiding property ensures that m is not revealed during
the transmission of C. To reveal m, the sender releases the
decommitment value d, in which case m is obtained by
all receivers, including J. Note that the hiding property,
as defined in commitment schemes, does not consider the
partial release of d and its implications on the partial reveal
of m. In fact, a common way of opening commitments is
by releasing the committed value itself [7].
For most applications, partial reveal of m with the partial
release of d does not constitute a security risk. After all, the
committer intends to reveal m by exposing d. However,
in our context, a partial reveal of m while d is being
transmitted can lead to the classification of m before the
transmission of d is completed. Thus, the jammer has
the opportunity to jam d instead of C once m has been
classified. To prevent this scenario, we introduce the strong
hiding property:

- Strong Hiding: For every polynomial-time party V in-
teracting with A and possessing pairs (C, dpart) and
(C′
, d′
part), there is no (probabilistic) polynomially-
efficient algorithm that would allow V associate C
with m and C′
with m′
, with non-negligble probability.
Here, dpart and d′
part are partial releases of d and d′
,
respectively, and the remaining parts of d and d′
are
assumed to be secret.
In the above definition, it is easily seen that the release
of dpart must be limited to a fraction of d, in order for
m to remain hidden. If a significant part of d becomes
known to the verifier, trivial attacks, such as brute forcing
the unknown bits of d, become possible.
5.2 A Strong Hiding Commitment Scheme (SHCS)
We propose a strong hiding commitment scheme (SHCS),
which is based on symmetric cryptography. Our main
motivation is to satisfy the strong hiding property while
keeping the computation and communication overhead to
a minimum. Assume that the sender S has a packet m for
R. First, S constructs (C, d) = commit(m), where,
C = Ek(π1(m)), d = k.
Here, the commitment function Ek() is an off-the-shelf
symmetric encryption algorithm (e.g., DES or AES [27]),
π1 is a publicly known permutation, and k ∈ {0, 1}s
is a
randomly selected key of some desired key length s (the
length of k is a security parameter). The sender broadcasts
(C||d), where “||” denotes the concatenation operation.
Upon reception of d, any receiver R computes
m = π−1
1 (Dk(C)) ,
where π−1
1 denotes the inverse permutation of π1. To satisfy
the strong hiding property, the packet carrying d is format-
ted so that all bits of d are modulated in the last few PHY layer
symbols of the packet. To recover d, any receiver must receive
and decode the last symbols of the transmitted packet,
thus preventing early disclosure of d. We now present the
implementation details of SHCS.
5.3 Implementation Details of SHCS
The proposed SHCS requires the joint consideration of the
MAC and PHY layers. To reduce the overhead of SHCS, the
decommitment value d (i.e., the decryption key k) is carried
in the same packet as the committed value C. This saves the
extra packet header needed for transmitting d individually.
To achieve the strong hiding property, a sublayer called the
“hiding sublayer” is inserted between the MAC and the
PHY layer. This sublayer is responsible for formatting m
before it is processed by the PHY layer. The functions of
the hiding sublayer are outlined in Fig. 4.
Consider a frame m at the MAC layer delivered to the
hiding sublayer. Frame m consists of a MAC header and
the payload, followed by the trailer containing the CRC
code. Initially, m is permuted by applying a publicly known
permutation π1. The purpose of π1 is to randomize the
Fig. 4. Processing at the hiding sublayer.
input to the encryption algorithm and delay the reception
of critical packet identifiers such as headers. After the
permutation, π1(m) is encrypted using a random key k to
produce the commitment value C = Ek(π1(m)). Although
the random permutation of m and its encryption with a
random key k seemingly achieve the same goal (i.e., the
randomization of the ciphertext), in Section 5.4 we show
that both are necessary to achieve packet hiding.
In the next step, a padding function pad() appends
pad(C) bits to C, making it a multiple of the symbol size.
Finally, C||pad(C)||k is permuted by applying a publicly
known permutation π2. The purpose of π2 is to ensure that
the interleaving function applied at the PHY layer does not
disperse the bits of k to other symbols. We now present the
padding and permutation functions in detail.
Padding–The purpose of padding is to ensure that k is
modulated in the minimum number of symbols needed for
its transmission. This is necessary for minimizing the time
for which parts of k become exposed. Let ℓ1 denote the
number of bits padded to C. For simplicity, assume that
the length of C is a multiple of the block length of the
symmetric encryption algorithm and hence, has the same
length ℓ as the original message m. Let also ℓ2 denote the
length of the header added at the PHY layer The frame
carrying (C, d) before the encoder has a length of (ℓ + ℓ1 +
ℓ2 +s) bits. Assuming that the rate of the encoder is α/β the
output of the encoder will be of length, β
α (ℓ + ℓ1 + ℓ2 + s).
For the last symbol of transmission to include α
β q bits of
the key k, it must hold that,
ℓ1 =
α
β
q − (ℓ + ℓ2)
β
α
mod q) . (3)
Permutation–The hiding layer applies two publicly
known permutations π1 and π2 at different processing
stages. Permutation π1 is applied to m before it is encrypted.
The purpose of π1 is twofold. First, it distributes critical
frame fields which can be used for packet classification
across multiple plaintext blocks. Hence, to reconstruct these
fields, all corresponding ciphertext blocks must be received
and decrypted. Moreover, header information is pushed
at the end of π1(m). This prevents early reception of the
corresponding ciphertext blocks.
For example, consider the transmission of a MAC frame
of length 2,336 bytes which carries a TCP data packet. The

m . . .
}
}
}
}
. . . . . . . . . . . . . . .
Field 1 Field 2 Field 3 Field n Random Payload
}
. . .. . . . . .1π (m)
1
π (m)E ( )k
. . .
Fig. 5. Application of permutation π1 on packet m.
MAC header is 28 bytes long and has a total of 18 distinct
fields. TCP header is 20 bytes long (assuming no optional
fields) and has 17 distinct fields. Assume the encryption
of a fixed block of 128 bits. Packet π1(m) is partitioned to
146 plaintext blocks {p1, p2, . . . , p146}, and is encrypted to
produce 146 ciphertext blocks C = c1||c2|| . . . ||c146. Each
field of the TCP and MAC headers is distributed bit-by-bit
from the most significant bit (MSB) to the least significant
bit (LSB) to each of the plaintext blocks in the reverse block
order. This process is depicted in Fig. 5.
For fields longer than one bit, bits are numbered from
the LSB to the MSB and are placed in reverse order to
each plaintext block. To recover any field i that is ℓi bits
long, the last ℓi ciphertext blocks must be received and
decrypted. If ℓi > ℓb, where ℓb denotes the ciphertext block
length, the bit placement process continues in a round-
robin fashion. The second goal of the permutation π1 is
to randomize the plaintext blocks. Assuming a random
payload, the permutation distributes the payload bits to all
plaintext blocks processed by the encryption function, thus
randomizing each ciphertext block.
Permutation π2 is applied to reverse the effects of in-
terleaving on the bits of k, so that k is contained at the
packet trailer. Interleaving can be applied across multiple
frequencies on the same symbol (e.g., in OFDM), or it may
span multiple symbols [9]. For example, consider a d × β
block interleaver. Without loss of generality, assume that
β = q, and let the last n rows of the last block passed
via the interleaver correspond to the encoded version of
the random key k. Permutation π2 rearranges the bits of
k at the interleaver matrix Ad×β in such a way that all
bits of k appear in the last n columns. Therefore, the
bits of k will be modulated as the last n symbols of the
transmitted packet. Note that this operation affects only
the interleaver block(s) that carries k. For the rest of the
packet, the interleaving function is performed normally,
thus preserving the benefits of interleaving. For PHY layer
implementations in which interleaving is applied on a per-
symbol basis (e.g, 802.11a and 802.11g), the application of
permutation π2 is not necessary.
5.4 Security Analysis
In this section, we analyze the security of SHCS by evalu-
ating the ability of J in classifying a transmitted packet at
different stages of the packet transmission.
Release of C–We first examine if J can classify m by
observing the commitment value C. Though C and k are
part of the same packet, symbols corresponding to C are
received first. The jammer can attempt to classify m by
launching a ciphertext-only attack on C as early as the re-
ception of the first ciphertext block. Because the encryption
key is refreshed at every transmission, a very small number
of ciphertext blocks are available for cryptanalysis. Appro-
priate selection of the key length s can prevent this type
of attack. Note that s can be well below the cryptographic
standards, due to the limited time available to the adversary
(until the transmission is completed). For instance, a 56-bit
long DES key is more than adequate for our purposes, since
the fastest known brute force attack on DES takes almost a
day [24]. Other types of known attacks such as differential
and linear cryptanalysis are not applicable, because they
require the collection of a large number of chosen or known
plaintext/ciphertext pairs [27].
Even if the key for a particular packet is revealed to
the adversary, packet classification is delayed until the end
of C’s transmission. The application of the permutation
function π1 distributes frame fields to ciphertext blocks
in the reverse order of transmission, with the MSBs from
each field appearing on the last ciphertext block. Hence,
reception of all blocks of C is required for the complete
recovery of headers. To minimize the communication over-
head, k must be selected to be of the smallest length
adequate for the protection of C, for the time required
to transmit one packet. However, special care must be
taken to withstand codebooks attacks on k. In such attacks,
the adversary can encrypt a particular message of inter-
est with all possible keys and construct a look-up table
(codebook) of all possible ciphertexts. If the encryption
of all possible messages with all possible keys results in
unique ciphertexts, there is a 1-1 correspondence between
a ciphertext and the generating plaintext/key pair. This
property is guaranteed with high probability when the
plaintext space M and the key space K are much smaller
than the ciphertext space C. Assuming the encryption of
a plaintext block mi with a key ki randomly maps to a
ciphertext ci = Eki (mi), every ciphertext ci ∈ C occurs with
probability pc = 1
|C| . The problem of finding the probability
that all |M||K| ciphertexts produced by the encryption of
all plaintexts with all keys are unique, can be formulated
as a “birthday problem” [27]:
Pr[ciphertexts unique] ≈ e
−|M|·|K|(|M|·|K|−1)
2|C| .
As an example, consider the encryption of a message m =
{m1, m2, . . . mx} with a key k of length 56 bits, using blocks
of 128 bits. For a fairly small plaintext space (e.g., |M| = 16),
the probability of ciphertext uniqueness is equal to 99.8%.
Thus, the adversary can recover k, by launching a codebook
attack on m1. The remaining ci’s are decrypted in real-time,
using the known value of k. Here, the plaintext space for m1
is considered to be small because of the structure imposed
by the static header of a packet (all fields of the header are
known to the adversary). Randomization of the plaintext,
ensures that all plaintexts are possible, thus equating the
plaintext space with the ciphertext space.

Partial release of d–Depending on the PHY layer im-
plementation, d = k requires n ≥ 1 symbols for its
transmission. Hence, part of k may become known before
the completion of the transmission of the packet at hand.
This release reduces the search space for a brute force attack
on k. Assume that the adversary pro-actively jams a few
symbols below the ECC capability of the receiver during
the transmission of C. In the best case, he can postpone
his decision to jam a transmitted packet until the trans-
mission of the last symbol (jam one more symbol to drop
the packet). He must therefore complete the classification
process before the last symbol is transmitted. Assuming that
the adversary waits until the maximum number of bits of
k are released, the key search space before the transmission
of the last two symbols is equal to 22 α
β q
keys. The adversary
must be capable of performing on average N = 2(2 α
β q−1)
R
decryptions per second in order to find k before the last
symbol is transmitted2
. Here, we have assumed that, on
average, half the key space must be searched.
For example, assume an 802.11a PHY layer operating at
6 Mbps, with every symbol carrying 24 bits of information.
Consider k to be a 56-bit DES key, fitting in three symbols.
The computational capability of the adversary must be
equal to N ≈ 3.52×1019
decryptions/sec in order to recover
k before the completion of the packet transmission. The
fastest known hardware implementation of a DES cracker
achieves a throughput of 2.92 × 1011
keys per second [24].
For an operating rate of 54 Mbps, all 56 bits of the key k
fit in one symbol (symbol size is 216 bits), thus preventing
the partial release of the decommitment value d.
A brute force attack on k may be successful if q ≤
β
2α log2 N + 1 . For instance, when N = 2.92 × 1011
, the
adversary can find k if q ≤ 19 bits. In fact, for small values
of q (e.g., 4), the adversary can launch the brute force attack
on k, several symbols before the end of k’s transmission.
Therefore, SHCS is suitable for PHY layer implementations
where the number of bits per symbol q is sufficiently large.
Note that our security analysis has excluded all processing
delays from the time that symbols are received to the time
that they become available for cryptanalysis.
Binding property–The binding property is not a security
requirement of SHCS under our adversary model. Since
the primary goal of any sender S in the network is to
communicate m, S has no interest in modifying m after
he has committed to it. However, under a more general
adversary model, the jammer may launch denial of service
attacks by making the receiver R to accept a k′
= k such
that m′
= D′
k(C) is a meaningful message. Even though
SHCS is not designed to ensure the binding property of
commitment schemes, generating a k′
= k that opens a
valid value of m′
= m is a computationally hard task.
In order to find such a k′
, the jammer has to launch a
brute force attack on C. Here, not only the attack must be
performed in a timely manner, but m′
has to be in the right
2. A more accurate calculation of N would assume an adversary trying
a brute force attack on k, with the reception of the first ciphertext block,
and adjusting the searching space according to the partial release of k.
format (source/destination address must be in the context
of the communications, CRC code must be valid, etc). Given
that k is transmitted right after C, the jammer has no time
to find an appropriate k′
that would lead to the decryption
of an acceptable m′
, assuming that such m′
exists. If m′
is
not meaningful, substituting k with k′
is equivalent to a
jamming attack on m without classification (no selectivity).
The binding property can be theoretically achieved if a
random string r is appended to m [23]. In this case, the
commitment/decommitment pair (C, d) is,
C = (γ, δ) = (Ek(m||r), r), d = k.
Provided that r is sufficiently long, a computationally
bounded jammer cannot find a k′
such that Dk′ (C) = m′
||r.
In this case, r preserves the integrity of message m. Since
the addition of r is not necessary for preventing real-time
classification of m, we leave the implementation of the
binding property to the discretion of the system designer.
5.5 Resource Overhead of SHCS
In this section we analyze the per-packet communication
and computational overhead of SHCS.
Communication Overhead–For every packet m, a ran-
dom key k of length s is appended. Also, (ℓb − (ℓ mod ℓb))
bits of overhead are added by the encryption algorithm,
to convert a plaintext of length ℓ to a multiple of the
encryption block. Thus, the communication overhead of
SHCS is equal to s + (ℓb − (ℓ mod ℓb)), per packet. Here,
we do not account for the padding string pad(C), because
the addition of pad(C) does not increase the number of
transmitted symbols.
Computation Overhead–The computation overhead of
SHCS is one symmetric encryption at the sender and one
symmetric decryption at the receiver. Because the header
information is permuted as a trailer and encrypted, all
receivers in the vicinity of a sender must receive the entire
packet and decrypt it, before the packet type and destina-
tion can be determined. However, in wireless protocols such
as 802.11, the complete packet is received at the MAC layer
before it is decided if the packet must be discarded or be
further processed [9]. If some parts of the MAC header are
deemed not to be useful information to the jammer, they
can remain unencrypted in the header of the packet, thus
avoiding the decryption operation at the receiver.
6 HIDING BASED ON CRYPTOGRAPHIC PUZZLES
In this section, we present a packet hiding scheme based on
cryptographic puzzles. The main idea behind such puzzles
is to force the recipient of a puzzle execute a pre-defined
set of computations before he is able to extract a secret of
interest. The time required for obtaining the solution of
a puzzle depends on its hardness and the computational
ability of the solver [10]. The advantage of the puzzle-
based scheme is that its security does not rely on the PHY
layer parameters. However, it has higher computation and
communication overhead.

Sender S
generate: k, tp
compute
P = puzzle(k, tp)
C = Ek(π1(m))
Receiver R
C, P C', P'
k' = solve(P)
compute: m' = π1
-1
(Dk'(C'))
verify: m' is meaningful
if not: discard m'
Fig. 6. The cryptographic puzzle-based hiding scheme.
In our context, we use cryptographic puzzles to tempo-
rary hide transmitted packets. A packet m is encrypted
with a randomly selected symmetric key k of a desirable
length s. The key k is blinded using a cryptographic puzzle
and sent to the receiver. For a computationally bounded
adversary, the puzzle carrying k cannot be solved before
the transmission of the encrypted version of m is completed
and the puzzle is received. Hence, the adversary cannot
classify m for the purpose of selective jamming.
6.1 Cryptographic Puzzle Hiding Scheme (CPHS)
Let a sender S have a packet m for transmission. The
sender selects a random key k ∈ {0, 1}s
, of a desired
length. S generates a puzzle P = puzzle(k, tp), where
puzzle() denotes the puzzle generator function, and tp
denotes the time required for the solution of the puzzle.
Parameter tp is measured in units of time, and it is directly
dependent on the assumed computational capability of the
adversary, denoted by N and measured in computational
operations per second. After generating the puzzle P, the
sender broadcasts (C, P), where C = Ek(π1(m)). At the
receiver side, any receiver R solves the received puzzle P′
to recover key k′
and then computes m′
= π−1
(Dk′ (C′
)). If
the decrypted packet m′
is meaningful (i.e., is in the proper
format, has a valid CRC code, and is within the context
of the receiver’s communication), the receiver accepts that
m′
= m. Else, the receiver discards m′
. Fig. 6 shows the
details of CPHS.
6.2 Implementation Details of CPHS
In this section, we consider several puzzle schemes as the
basis for CPHS. For each scheme, we analyze the imple-
mentation details which impact security and performance.
Cryptographic puzzles are primitives originally suggested
by Merkle as a method for establishing a secret over an
insecure channel [16]. They find a wide range of applica-
tions from preventing DoS attacks to providing broadcast
authentication and key escrow schemes.
Time-lock Puzzles–Rivest et al. proposed a construction
called time-lock puzzles, which is based on the iterative
application of a precisely controlled number of modulo
operations [22]. Time-lock puzzles have several attractive
features such as the fine granularity in controlling tp and
the sequential nature of the computation. Moreover, the
puzzle generation requires significantly less computation
compared to puzzle solving.
In a time-lock puzzle, the puzzle constructor generates
a composite modulus g = u · v, where u and v are two
large random prime numbers. Then, he picks a random
a, 1 < a < g and hides the encryption key in Kh =
k + a2t
mod g, where t = tp · N, is the amount of time
required to solve for k. Here, it is assumed that the solver
can perform N squarings modulo g per second. Note that
Kh can be computed efficiently if φ(g) = (u − 1)(v − 1) or
the factorization of g are known, otherwise a solver would
have to perform all t squarings to recover k. The puzzle
consists of the values P = (g, Kh, t, a).
In our setup, the value of the modulus g is known a priori
and need not be communicated (may change periodically).
The sender reveals the rest of the puzzle information in the
order (Kh, t, a). Note that if any of t, a are unknown, any
value of k is possible [22].
Puzzles based on hashing–Computationally limited re-
ceivers can incur significant delay and energy consumption
when dealing with modulo arithmetic. In this case, CPHS
can be implemented from cryptographic puzzles which
employ computationally efficient cryptographic primitives.
Client puzzles proposed in [10], use one-way hash func-
tions with partially disclosed inputs to force puzzle solvers
search through a space of a precisely controlled size. In our
context, the sender picks a random key k with k = k1||k2.
The lengths of k1 and k2 are s1, and s2, respectively. He then
computes C = Ek(π1(m)) and transmits (C, k1, h(k)) in this
particular order. To obtain k, any receiver has to perform
on average 2s2−1
hash operations (assuming perfect hash
functions). Because the puzzle cannot be solved before h(k)
has been received, the adversary cannot classify m before
the completion of m’s transmission.
6.3 Security Analysis of CPHS
With the completion of the transmission of P, any receiver
can recover m. Therefore, a selective jammer must attempt
to classify m before the transmission of P has been com-
pleted. We analyze the security of CPHS at different stages
of its execution.
Reception of C–The jammer can attempt to classify m
by cryptanalyzing ciphertext C = Ek(π1(m)). This attack is
identical to the effort of classifying m with the transmission
of C at the SHCS. The same analysis presented in Section
5.4 holds for the case of CPHS. The selection of a key of
adequate length (e.g., 56-bit DES key) is sufficient to prevent
both ciphertext-only and codebook attacks.
Solving P–The transmission of k in the form of a puzzle
P prevents any receiver from recovering k for at least time
tp, after the puzzle has been received. A jammer may try
to guess and solve P before its transmission is completed.
In the best case, the adversary must finish the classification
of m before the transmission of the last symbol of P. The
number of possible puzzle values at the beginning of the
second to last symbol are 22 α
β q
. Assuming a brute force
attack on the missing bits of the puzzle, the computational
load of the adversary increases on average to 22 α
β q−1
tp.

The value of tp has already been selected to prevent the
puzzle solution until its transmission is completed. Hence,
early solution of P before all its bits are received cannot be
achieved. Note that the security of CPHS is not dependent
on the PHY layer parameter q, but on the selection of
tp. Therefore, this method is applicable even to wireless
systems where q obtains relatively small values.
6.4 Resource Overhead of CPHS
Communication Overhead–The per-packet communication
overhead of CPHS is equal to the length of P, in addition to
the padding added by the encryption function. If the puzzle
is realized using time-locks, the length of P is equal to the
lengths of Kh, a, and t. The value Kh is computed modulo g
and has the same length as g. Similarly, a has a length equal
to the length of g. The size of t is potentially smaller than
a, g, and Kh, and depends on the computational capability
of the adversary. The security of time locks depends on the
difficulty in factoring g or finding φ(g), where φ() denotes
the Euler φ−function. Typical values of g are in the order
of 1,024 bits [27]. Since messages need to remain hidden for
only a short period of time, the modulo can be chosen to be
of much smaller size and be periodically refreshed. In the
case of hash-based puzzles, the communication overhead is
equal to the transmission of the key k1 which is of length s1
and the hash value h(k). The typical length of hash function
is 160 bits [27].
Computation Overhead–In time-lock puzzles, the sender
has to apply one permutation on m, perform one symmetric
encryption, and one modulo squaring operation to hide k.
On the receiver side, the receiver has to perform t modulo
squaring operations to recover k, one symmetric decryption
to recover π1(m), and apply the inverse permutation. In the
case of hash-based puzzles, the modulo squaring operation
is substituted by, on average, 2s2−1
hashing operations.
7 HIDING BASED ON ALL-OR-NOTHING TRANS-
FORMATIONS
In this section, we propose a solution based on All-Or-
Nothing Transformations (AONT) that introduces a modest
communication and computation overhead. Such transfor-
mations were originally proposed by Rivest to slow down
brute force attacks against block encryption algorithms [21].
An AONT serves as a publicly known and completely
invertible pre-processing step to a plaintext before it is
passed to an ordinary block encryption algorithm. A trans-
formation f, mapping message m = {m1, · · · , mx} to a
sequence of pseudo-messages m′
= {m′
1, · · · , m′
x′ }, is an
AONT if [21]: (a) f is a bijection, (b) it is computationally
infeasible to obtain any part of the original plaintext, if
one of the pseudo-messages is unknown, and (c) f and its
inverse f−1
are efficiently computable.
When a plaintext is pre-processed by an AONT before
encryption, all ciphertext blocks must be received to obtain
any part of the plaintext. Therefore, brute force attacks are
slowed down by a factor equal to the number of ciphertext
blocks, without any change on the size of the secret key.
Note that the original AONT proposed in [21] is computa-
tionally secure. Several AONT schemes have been proposed
that extend the definition of AONT to undeniable security
[26]. Under this model, all plaintexts are equiprobable in
the absence of at least one pseudo-message.
7.1 An AONT-based Hiding Scheme (AONT-HS)
In our context, packets are pre-processed by an AONT be-
fore transmission but remain unencrypted. The jammer can-
not perform packet classification until all pseudo-messages
corresponding to the original packet have been received
and the inverse transformation has been applied. Packet m
is partitioned to a set of x input blocks m = {m1, . . . , mx},
which serve as an input to an AONT f : {Fu}x
→ {Fu}x′
.
Here, Fu denotes the alphabet of blocks mi and x′
denotes
the number of output pseudo-messages with x′
≥ x. The
set of pseudo-messages m′
= {m′
1, . . . , m′
x′ } is transmitted
over the wireless medium. At the receiver, the inverse
transformation f−1
is applied after all x′
pseudo-messages
are received, in order to recover m.
7.2 Implementation details of the AONT-HS
In this section, we describe two AONTs which can be
employed in AONT-HS; a linear transformation [26], and
the original package transformation [21].
Linear AONT–In [26], Stinson showed how to construct
a linear AONT when the alphabet of the input blocks is
a finite field Fu, with the order u being a prime power.
He showed that if an invertible matrix M = {mij|mij ∈
Fu, mij = 0}x×x exists, then the transformation f(m) =
mM−1
is a linear AONT. He also provided a method for
constructing such M which is as follows.
Let u = vi
, where v is prime and i is a positive integer.
Choose λ ∈ Fu such that λ /∈ {n−1 (mod v), n−2 (mod v)}
and define the linear AONT LT to be,
LT =





1 0 · · · 0 1
...
...
...
...
...
0 0 · · · 1 1
1 1 · · · 1 λ





(4)
Given m = {m1, . . . , mx},
m′
x = λmx +
x−1
j=1
mj, m′
i = mi + m′
x, 1 ≤ i ≤ (x − 1). (5)
Conversely, given m′
= {m′
1, . . . , m′
x}, the original input
m = {m1, . . . , mx} is recovered as follows:
mi = m′
i − m′
x, 1 ≤ i ≤ (x − 1), (6)
mx = γ(m′
1 + . . . m′
x−1 − m′
x), γ =
1
n − λ − 1
. (7)
Note from (6), (7) that if any of the {m′
i} is missing, all
values of mi are possible, for every i. Thus, the linear AONT
provides undeniable security.

Sender S
compute:
m || pad(m)
transform:
m' = f (m || pad(m))
Receiver R
receive m'
compute
m || pad(m)= f -1
(m')
recover m
m'
Fig. 7. The AONT-based Hiding Scheme (AONT-HS).
The Package Transform–In the package transform [21],
given a message m, and a random key k′
, the output
pseudo-messages are computed as follows:
m′
i = mi ⊕ Ek′ (i), for i = 1, 2, . . ., x (8)
m′
x+1 = k′
⊕ e1 ⊕ e2 ⊕ · · · ⊕ ex, (9)
where ei = Ek0 (m′
i ⊕ i), for i = 1, 2, . . ., x, and k0 is a fixed
publicly-known encryption key. With the reception of all
pseudo-messages message m is recovered as follows:
k′
= m′
x+1 ⊕ e1 ⊕ e2 ⊕ · · · ⊕ ex, (10)
mi = m′
i ⊕ Ek′ (i), for i = 1, 2, . . . , x. (11)
Note that if any m′
i is unknown, any value of k′
is
possible, because the corresponding ei is not known. Hence,
Ek′ (i) cannot be recovered for any i, making it infeasible
to obtain any of the mi.
Hiding Sublayer Details–AONT-HS is implemented at
the hiding sublayer residing between the MAC and the
PHY layers. In the first step, m is padded by applying
function pad() to adjust the frame length so that no padding
is needed at the PHY layer, and the length of m becomes
a multiple of the length of the pseudo-messages m′
i. This
will ensure that all bits of the transmitted packet are part of
the AONT. In the next step, m||pad(m) is partitioned to x
blocks, and the AONT f is applied. Message m′
is delivered
to the PHY layer. At the receiver, the inverse transformation
f−1
is applied to obtain m||pad(m). The padded bits are
removed and the original message m is recovered. The steps
of AONT-HS are shown in Fig. 7.
7.3 Security Analysis of the AONT-HS
Partial reception of m′
i, i < x′
–In the AONT-HS, the
jammer may attempt to classify m without receiving all m′
i
(1 ≤ i ≤ x′
). By definition, AONTs prevent the computation
of any part of m without the reception of all the pseudo-
messages. In fact, for the linear AONT, undeniable security
is achieved. The jammer can launch a brute force attack on
m as early as the reception of m′
1. However, the system of
equations formed by m′
i’s when at least one is missing, has
a number of solutions equal to the message space. All these
solutions are equiprobable.
Partial release of m′
x–With the partial release of the
last pseudo-message m′
x, the space of the possible original
messages m is reduced. As stated by our adversarial model,
the classification of m must be completed before the last
symbol of m′
x, is transmitted. The search space for m′
x is
reduced to its smallest value before the transmission of the
last two symbols, in which case the possible values of m are
equal to 22 α
β q
. The adversary must be capable of solving on
average 22 α
β q−1
systems of linear equations in time equal
to the length of one symbol ( 1
R sec), in the case of the
linear AONT, or perform the same number of decryptions
for the case of the package transform. For instance when
q = 48 and α/β = 1/2 (802.11a), the search space is equal
to 1.4 × 1014
. As in the case of SHCS, when the value of q
becomes small (q ≤ β
2α log2 N +1), a brute force attack on m
is possible. Therefore, AONT-HS is suitable for PHY layer
implementations where q is sufficiently large.
7.4 Resource Overhead of the AONT-HS
Communication Overhead–In AONT-HS, the original set
of x messages is transformed to a set of x′
pseudo-messages,
with x′
≥ x. Additionally, the function pad() appends
(ℓb − (ℓ mod ℓb)) bits in order to make the length of m a
multiple of the length ℓb of the pseudo-messages m′
. Hence,
the communication overhead introduced is (ℓb(x′
−x)+ℓb −
(ℓ mod ℓb)) bits. For the linear AONT, x = x′
, and therefore,
only the padding communication overhead is introduced.
For the package transform, the overhead is equal to the
length of one pseudo-message (x′
= x + 1).
Computation Overhead–The linear AONT requires only
elementary arithmetic operations such as string addition
and multiplication, making it particularly fast due to its
linear nature. The package transform requires x′
symmetric
encryptions at the sender and an equal amount of sym-
metric decryptions at the receiver. Note that the length of
the plaintext for the x′
encryptions is relatively small com-
pared to the length of message m (indexes 1, . . . , x are en-
crypted). Therefore, only one ciphertext block is produced
per pseudo-message. Assuming a pseudo-message block
size equal to the ciphertext block size ℓb, the computational
overhead of the x′
encryptions required by the package
transform is equivalent to the overhead of one encryption
of a message of length ℓ + ℓb.
8 EVALUATION OF PACKET-HIDING TECHNIQUES
In this section, we evaluate the impact of our packet-hiding
techniques on the network performance via extensive sim-
ulations. We used the OPNETTM
Modeler 14.5 [18] to im-
plement the hiding sublayer and measure its impact on the
effective throughput of end-to-end connections and on the
route discovery process in wireless ad-hoc networks. We
chose a set of nodes running 802.11b at the PHY and MAC
layers, AODV for route discovery, and TCP at the transport
layer. Aside from our methods, we also implemented a
simple MAC layer encryption with a static key.
Impact on real-time systems–Our packet-hiding methods
require the processing of each individual packet by the
hiding sublayer. We emphasize that the incurred processing
delay is acceptable, even for real-time applications. The
SCHS requires the application of two permutations and one

N.H. M.E. C.S. T.P. H.P. L.T. P.T.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
E[T](Mbps)
0
0.1
0.2
0.3
0.4
0.5
RouteDiscoveryTime(sec)
0
0.2
0.4
0.6
0.8
1
E[T](Mbps)
(a) (b) (c)
Fig. 8. (a) Average effective throughput (line network topology), (b) average route discovery time (non-congested network),
(c) average effective throughput (congested network). N.H: No packet hiding, M.E: MAC-layer encryption with a static key,
C.S.: SHCS, T.P.: Time-lock CPHS, H.P.: Hash-based CPHS, L.T.: Linear AONT-HS, P.T.: Package transform.
symmetric encryption at the sender, while the inverse oper-
ations have to be performed at the receiver. Such operations
can be implemented in hardware very efficiently. Symmetric
encryption such as AES can be implemented at speeds of
tens of Gbps/s when realized with Application Specific
Integrated Circuits (ASICs) or Field Programmable Gate
Arrays (FPGAs) [6]. These processing speeds are orders
of magnitude higher than the transmission speeds of most
current wireless technologies, and hence, do not impose a
significant delay.
Similarly, the AONT-HS performs linear operations on
the packet that can be efficiently implemented in hardware.
We note that a non-negligible processing delay is incurred
by the CPHS. This is due to the cryptographic puzzle that
must be solved at the receiver. As suggested in Section 6,
CPHS should only be employed when the symbol size at
the PHY layer is too small to support the SHCS and AONT-
HS solutions. The processing delays of the various schemes
are taken into account in our experimental evaluations.
Experimental evaluation–In the first set of experiments,
we setup a single file transfer between a client and server,
connected via a multi-hop route. The client requested a 1
MB file from the server. We evaluated the effects of packet
hiding by measuring the effective throughput of the TCP
connection in the following scenarios: (a) No packet hiding
(N.H.), (b) MAC-layer encryption with a static key (M.E.),
(c) SHCS (C.S.), (d) Time-lock CPHS (T.P.), (e) Hash-based
CPHS (H.P.), (f) Linear AONT-HS (L.T.), and (g) AONT-HS
based on the package transform (P.T.).
In Fig. 8(a), we show the effective throughput aver-
aged over 100 different traces. We observe that MAC-layer
encryption, SHCS and the linear AONT-HS achieve an
effective throughput close to the throughput in the absence
of packet hiding. This is justified by the the relatively
small communication overhead of each hiding method and
the small queuing delay at intermediate routers due to
the absence of any cross traffic. The AONT-HS based on
the package transform achieved slightly lower throughput,
because it occurs a per-packet overhead of 128 bits as
opposed to 56 bits for SHCS. We also observe that hiding
techniques based on cryptographic puzzles decrease the ef-
fective throughput of the TCP connection to half, compared
to the no hiding case. This performance is anticipated since
the time required to solve a puzzle after a packet has been
received at the MAC layer is equal to the transmission
time of each packet. While this constitutes a significant
performance reduction, we emphasize that cryptographic
puzzles were suggested as a candidate solution only when
the symbol size is so small that more efficient hiding
methods do not provide adequate levels of security.
In the second set of experiments, we studied the impact
of packet hiding on the route discovery process in an
ad-hoc network. We generated a random topology of 54
nodes placed in an area of 500 × 400 m2
. Nodes discovered
routes using the AODV routing protocol. A total of twenty
client/server pairs exchanged messages of size 1 KB using
the TCP protocol, at randomly chosen starting times. The
size of the message exchanged between pairs of nodes was
kept small in order to avoid skewing of the route discovery
performance due to network congestion.
The average route discovery delay is shown in Fig. 8(b).
This delay is defined as the time difference between the
transmission of the first RREQ from a source and the
reception of the corresponding RREP from the destination.
We observe that the impact of packet hiding on the route
discovery delay is minimal compared to the case where
no packet hiding is employed. This similarity in perfor-
mance is due to the expanding ring search technique of
AODV, which is used to prevent unnecessary network-
wide dissemination of RREQs [19]. In order to discover
a route, the originating node sends a RREQ with a time-
to-live (TTL) value equal to one hop, and waits for the
corresponding RREP. If the RREP is not received before a
timeout value (to) expires, the originating node increases
the TTL and the timeout to, and re-broadcasts the RREQ.
This process is repeated until a valid RREP is received,
or the TTL value exceeds the maximum diameter of the
network. The expanding ring search technique introduces a
dominant delay in comparison to the delay introduced by
the packet-hiding techniques. For example, in the case of

time-lock CPHS, the per-packet delay overhead is tp = 448
µsec. Using the default values specified by RFC 3561 [19],
the value of the first timeout is to = 240 msec, which is 535
times higher than tp, making the total delay introduced by
time-lock CPHS insignificant.
In the third set of experiments, we evaluated the per-
formance of TCP in a congested ad-hoc network. We con-
sidered the same network topology used in the second set
of experiments. Twenty source/destination pairs simultane-
ously exchanged 2 MB files using TCP. In Fig. 8(d), we show
the effective throughput averaged over all 20 TCP connec-
tions. We observe that efficient packet-hiding techniques
such as SHCS, and AONT-HS have a relatively small impact
on the overall throughput. This is because in a congested
network, the performance is primarily dependent on the
queueing delays at the relay nodes. The communication
overhead introduced by the transmission of the packet-
hiding parameters is small and hence, does not significantly
impact the throughput. On the other hand, for CPHS, we
observe a performance reduction of 25% − 30% compared
to the case of no packet-hiding. This reduction is attributed
to the delay introduced by CPHS for the reception of each
packet. Note that in the congested network scenario, the
throughput reduction of CPHS is smaller compared to the
non-congested one because nodes can take advantage of the
queuing delays to solve puzzles.
9 RELATED WORK
Jamming attacks on voice communications have been
launched since the 1940s [25]. In the context of digital
communications, the jamming problem has been addressed
under various threat models. We present a classification
based on the selective nature of the adversary.
9.1 Prior work on Selective Jamming
In [33], Thuente studied the impact of an external selec-
tive jammer who targets various control packets at the
MAC layer. To perform packet classification, the adversary
exploits inter-packet timing information to infer eminent
packet transmissions. In [11], Law et al. proposed the
estimation of the probability distribution of inter-packet
transmission times for different packet types based on
network traffic analysis. Future transmissions at various
layers were predicted using estimated timing information.
Using their model, the authors proposed selective jamming
strategies for well known sensor network MAC protocols.
In [1], Brown et al. illustrated the feasibility of selective
jamming based on protocol semantics. They considered
several packet identifiers for encrypted packets such as
packet size, precise timing information of different pro-
tocols, and physical signal sensing. To prevent selectivity,
the unification of packet characteristics such as the mini-
mum length and inter-packet timing was proposed. Similar
packet classification techniques were investigated in [4].
Liu et al. considered a smart jammer that takes into
account protocol specifics to optimize its jamming strategy
[14]. The adversary was assumed to target control messages
at different layers of the network stack. To mitigate smart
jamming, the authors proposed the SPREAD system, which
is based on the idea of stochastic selection between a
collection of parallel protocols at each layer. The uncer-
tainty introduced by this stochastic selection, mitigated the
selective ability of the jammer. Greenstein et al. presented
a 802.11-like wireless protocol called Slyfi that prevents the
classification of packets by external observers. This protocol
hides all explicit identifiers from the transmitted packets
(e.g. MAC layer header and payload), by encrypting them
with keys only known to the intended receivers [8].
Selective jamming attacks have been experimentally im-
plemented using software-defined radio engines [32], [34].
Wilhelm et al. implemented a USRP2-based jamming plat-
form called RFReact that enables selective and reactive jam-
ming [34]. RFReact was shown to be agnostic to technology
standards and readily adaptable to any desired jamming
strategy. The success rate of a selective jamming attack
against a 802.15.4 network was measured to be 99.96%.
Blapa et al. studied selective jamming attacks against the
rate-adaptation mechanism of 802.11 [32]. They showed that
a selective jammer targeting specific packets in a point-to-
point 802.11 communication was able to reduce the rate
of the communication to the minimum value of 1 Mbps,
with relatively little effort (jamming of 5-8 packets per
second). The results were experimentally verified using the
USRP2/GNU radio platform.
Several researchers have suggested channel-selective jam-
ming attacks, in which the jammer targets the broadcast
control channel. It was shown that such attacks reduce the
required power for performing a DoS attack by several or-
ders of magnitude [3]. To protect control-channel traffic, the
replication of control transmission in multiple channels was
suggested in [3], [30], [31]. The “locations” of the control
channels where cryptographically protected. In [12], Lazos
et al. proposed a randomized frequency hopping algorithm
to protect the control channel from inside jammers. Strasser
et al. proposed a frequency hopping anti-jamming tech-
nique that does not require the existence of a secret hopping
sequence, shared between the communicating parties [28].
9.2 Non-Selective Jamming Attacks
Conventional methods for mitigating jamming employ
some form of SS communications [5], [25]. The transmitted
signal is spread to a larger bandwidth following a PN
sequence. Without the knowledge of this sequence, a large
amount of energy (typically 20-30 dB gain) is required to in-
terfere with an ongoing transmission. However, in the case
of broadcast communications, compromise of commonly
shared PN codes neutralizes the advantages of SS.
Pöpper et al. proposed a jamming-resistant communi-
cation model for pairwise communications that does not
rely on shared secrets. Communicating nodes use a physi-
cal layer modulation method called Uncoordinated Direct-
Sequence Spread Spectrum (UDSSS) [20]. They also pro-
posed, a jamming-resistant broadcast method in which

transmissions are spread according to PN codes randomly
selected from a public codebook [20]. Several other schemes
eliminate overall the need for secret PN codes [15], [29].
Lin et al. showed that jamming 13% of a packet is suffi-
cient to overcome the ECC capabilities of the receiver [13].
Xu et al. categorized jammers into four models: (a) a con-
stant jammer, (b) a deceptive jammer that broadcasts fab-
ricated messages, (c) a random jammer, and (d) a reactive
jammer that jams only if activity is sensed [37]. They further
studied the problem of detecting the presence of jammers
by measuring performance metrics such as packet delivery
ratio [35]–[37]. Cagalj et al. proposed wormhole-based anti-
jamming techniques for wireless sensor networks (WSNs)
[2]. Using a wormhole link, sensors within the jammed
region establish communications with outside nodes, and
notify them regarding ongoing jamming attacks.
10 CONCLUSION
We addressed the problem of selective jamming attacks
in wireless networks. We considered an internal adversary
model in which the jammer is part of the network under
attack, thus being aware of the protocol specifications and
shared network secrets. We showed that the jammer can
classify transmitted packets in real time by decoding the
first few symbols of an ongoing transmission. We evaluated
the impact of selective jamming attacks on network proto-
cols such as TCP and routing. Our findings show that a
selective jammer can significantly impact performance with
very low effort. We developed three schemes that transform
a selective jammer to a random one by preventing real-time
packet classification. Our schemes combine cryptographic
primitives such as commitment schemes, cryptographic
puzzles, and all-or-nothing transformations (AONTs) with
physical layer characteristics. We analyzed the security
of our schemes and quantified their computational and
communication overhead.
REFERENCES
[1] T. X. Brown, J. E. James, and A. Sethi. Jamming and sensing of
encrypted wireless ad hoc networks. In Proceedings of MobiHoc, pages
120–130, 2006.
[2] M. Cagalj, S. Capkun, and J.-P. Hubaux. Wormhole-based anti-
jamming techniques in sensor networks. IEEE Transactions on Mobile
Computing, 6(1):100–114, 2007.
[3] A. Chan, X. Liu, G. Noubir, and B. Thapa. Control channel jamming:
Resilience and identification of traitors. In Proceedings of ISIT, 2007.
[4] T. Dempsey, G. Sahin, Y. Morton, and C. Hopper. Intelligent sensing
and classification in ad hoc networks: a case study. Aerospace and
Electronic Systems Magazine, IEEE, 24(8):23–30, August 2009.
[5] Y. Desmedt. Broadcast anti-jamming systems. Computer Networks,
35(2-3):223–236, February 2001.
[6] K. Gaj and P. Chodowiec. FPGA and ASIC implementations of AES.
Cryptographic Engineering, pages 235–294, 2009.
[7] O. Goldreich. Foundations of cryptography: Basic applications. Cam-
bridge University Press, 2004.
[8] B. Greenstein, D. Mccoy, J. Pang, T. Kohno, S. Seshan, and D. Wether-
all. Improving wireless privacy with an identifier-free link layer
protocol. In Proceedings of MobiSys, 2008.
[9] IEEE. IEEE 802.11 standard. http://standards.ieee.org/getieee802/
download/802.11-2007.pdf, 2007.
[10] A. Juels and J. Brainard. Client puzzles: A cryptographic counter-
measure against connection depletion attacks. In Proceedings of NDSS,
pages 151–165, 1999.
[11] Y. W. Law, M. Palaniswami, L. V. Hoesel, J. Doumen, P. Hartel, and
P. Havinga. Energy-efficient link-layer jamming attacks against WSN
MAC protocols. ACM Transactions on Sensors Networks, 5(1):1–38, 2009.
[12] L. Lazos, S. Liu, and M. Krunz. Mitigating control-channel jamming
attacks in multi-channel ad hoc networks. In Proceedings of the 2nd
ACM conference on wireless network security, pages 169–180, 2009.
[13] G. Lin and G. Noubir. On link layer denial of service in data wireless
LANs. Wireless Communications and Mobile Computing, 5(3):273–284,
May 2004.
[14] X. Liu, G. Noubir, and R. Sundaram. Spread: Foiling smart jammers
using multi-layer agility. In Proceedings of INFOCOM, pages 2536–
2540, 2007.
[15] Y. Liu, P. Ning, H. Dai, and A. Liu. Randomized differential DSSS:
Jamming-resistant wireless broadcast communication. In Proceedings
of INFOCOM, San Diego, 2010.
[16] R. C. Merkle. Secure communications over insecure channels. Com-
munications of the ACM, 21(4):294–299, 1978.
[17] G. Noubir and G. Lin. Low-power DoS attacks in data wireless lans
and countermeasures. Mobile Computing and Communications Review,
7(3):29–30, 2003.
[18] OPNET. OPNETtm modeler 14.5. http://www.opnet.com/.
[19] C. Perkins, E. Belding-Royer, and S. Das. RFC 3561: Ad hoc on-
demand distance vector (AODV) routing. Internet RFCs, 2003.
[20] C. Pöpper, M. Strasser, and S. ˇCapkun. Jamming-resistant broadcast
communication without shared keys. In Proceedings of the USENIX
Security Symposium, 2009.
[21] R. Rivest. All-or-nothing encryption and the package transform.
Lecture Notes in Computer Science, pages 210–218, 1997.
[22] R. Rivest, A. Shamir, and D. Wagner. Time-lock puzzles and timed-
release crypto. Massachusetts Institute of Technology, 1996.
[23] B. Schneier. Applied cryptography: protocols, algorithms, and source code
in C. John Wiley & Sons, 2007.
[24] SciEngines. Break DES in less than a single day. http://www.
sciengines.com, 2010.
[25] M. K. Simon, J. K. Omura, R. A. Scholtz, and B. K. Levitt. Spread
Spectrum Communications Handbook. McGraw-Hill, 2001.
[26] D. Stinson. Something about all or nothing (transforms). Designs,
Codes and Cryptography, 22(2):133–138, 2001.
[27] D. Stinson. Cryptography: theory and practice. CRC press, 2006.
[28] M. Strasser, C. Pöpper, and S. ˇCapkun. Efficient uncoordinated fhss
anti-jamming communication. In Proceedings of MobiHoc, pages 207–
218, 2009.
[29] M. Strasser, C. Pöpper, S. ˇCapkun, and M. Cagalj. Jamming-resistant
key establishment using uncoordinated frequency hopping. In Pro-
ceedings of IEEE Symposium on Security and Privacy, 2008.
[30] P. Tague, M. Li, and R. Poovendran. Probabilistic mitigation of control
channel jamming via random key distribution. In Proceedings of
PIMRC, 2007.
[31] P. Tague, M. Li, and R. Poovendran. Mitigation of control channel
jamming under node capture attacks. IEEE Transactions on Mobile
Computing, 8(9):1221–1234, 2009.
[32] B. Thapa, G. Noubir, R. Rajaramanand, and B. Sheng. On the
robustness of IEEE802.11 rate adaptation algorithms against smart
jamming. In Proceedings of WiSec, 2011.
[33] D. Thuente and M. Acharya. Intelligent jamming in wireless networks
with applications to 802.11 b and other networks. In Proceedings of
the IEEE Military Communications Conference MILCOM, 2006.
[34] M. Wilhelm, I. Martinovic, J. Schmitt, and V. Lenders. Reactive
jamming in wireless networks: How realistic is the threat? In
Proceedings of WiSec, 2011.
[35] W. Xu, W. Trappe, and Y. Zhang. Anti-jamming timing channels for
wireless networks. In Proceedings of WiSec, pages 203–213, 2008.
[36] W. Xu, W. Trappe, Y. Zhang, and T. Wood. The feasibility of launching
and detecting jamming attacks in wireless networks. In Proceedings
of MobiHoc, pages 46–57, 2005.
[37] W. Xu, T. Wood, W. Trappe, and Y. Zhang. Channel surfing and spatial
retreats: defenses against wireless denial of service. In Proceedings of
the 3rd ACM workshop on Wireless security, pages 80–89, 2004.

Packet hiding methods for preventing selective

More Related Content

What's hot (20)

Viewers also liked (9)

Similar to Packet hiding methods for preventing selective (20)

Recently uploaded (20)

Packet hiding methods for preventing selective