Aerospace and Electronic Systems Magazine February 2018 - 7

Xu et al.

Figure 2.

Watermark-based data validation.

manipulation. In this way, avionics data can be securely stored
and verified for integrity.
There are two important modules within the watermark framework: (1) mark generation at the transmitter and (2) mark recognition at the receiver. The watermark system architecture consists
of two major components: the encoding process and the decoding
process, as shown in Figure 2.
In the watermark encoding process, the encoder initially generates the L-chip watermark code ωt,k at time t for user k's data
sequence. A PN code PN t , k [16] is also selected to spread the signal over a wider bandwidth (greater than its original signal bandwidth) with chip duration tpn. A modulation operation is defined
as Mo(). The transmitted baseband signal that is denoted as xk can
be expressed as xk = Mo[Qk(ωt,k)], where Qk() represents the signal
sequence. The values of the chip series are sequences of −1's and
+1's. Watermark values have a predefined amplitude Ak. Thus, at
any instant of time t, the watermark value is Akωt,k, where ωt,k ∈
(−1,+1). Then collected traffic flow data are embedded with the
previously generated watermarks. There are numerous ways to embed marks into data flow. Here, we choose a modulation procedure
that demonstrates the case of embedding the marks into the data
flow rate changes. On the basis of previous assumptions, the encoding process can be formalized as xk = Wt , k + Mo[ PN t , k , Qk ( Akωt , k )]
, where Wt,k represents an average data packet transmission rate at
time t for user k. When a chip in the watermark sequence is −1, a
strong interference is applied against the data flow so that the flow
has a lower rate with a reduction by Ak. Likewise, when a chip is
+1, a weak interference is applied against the data flow so that the
flow has a higher rate with an increase by Ak. Hence, the encoded
high value is Wt,k + Ak, and the low value is Wt,k − Ak. The value of
the target sensor data flow should be large enough for the defense
system to introduce watermarks. When the modulated avionics
data flow is transmitted in a wireless communication network, the
adversary may manipulate the data and disrupt watermarks, which
can be detected in the decoding process.
FEBRUARY 2018

In the watermark decoding process, a data center at the receiver captures the data flow and then divides it into segments.
Each segment lasts for a chip duration, and the average data
flow rate is calculated during each segment. Assume that the sequence has n continuous segments, which corresponds to a full
period of the encoding sequence. With the knowledge of the
watermark sequence before sending, the data sequence of the
original packet can be recovered by roughly representing as
Rt′, k = Wt , k + Mo[ PN t , k , Qk ( Akωt , k + Bt )] + ϕt, where all the items
are 1 × n vectors, R′t , k represents the received data sequence after
transmission, Bt is denoted as the attack sequence, and φt refers to
a random variable of noise. The data go next to the decoding process. A high-pass filter is applied against the received signal R′t , k to
remove the direct component Wt,k from the received signal. After
filtering, the filtered received signal R′t , k can be roughly expressed
as Rt′, k ≈ Mo[ PN t , k , Qk ( Akωt , k + Bt )] + ϕt, where PN r , k is a locally
generated PN code at the receiver, which is identical to the code
PN t , k at the transmitter. It is used to decode the received signal
after modulation, and the similarity degree can be formalized as
SD t , k = Rt′, k ·PN r , k = (Wt , k + Mo[ PN t , k , Qk ( Akωt , k + Bt )] + ϕt )·Mo( PN r , k )
, where · is the dot product operation. When PN r , k = PN t , k, we have
Mo[ PN t , k , Qk ( Akωt , k + Bt )]·Mo( PN r , k ) = 1 and ϕt ·Mo( PN r , k ) = 0, if
Bt = 0, which means there is no cyberattack during the data transmission. Hence, the original signal with the embedded watermark
Qk(Akωt,k) can be exactly recovered intact. Here, the decoding data
are correlated with the original data to determine the presence or
absence of watermarks. If the watermark is absent, we conclude
that the transmitted avionics data are manipulated by an adversary
during the transmission.

NOISE-MODULATED MULTIUSER SUPERPOSITION
COMMUNICATION
As shown in Figure 3, the watermark-encoded avionics data are
transferred to the transmitter. To effectively improve the covertness
of communication and serve multiple users simultaneously, noise
modulation is applied to conceal data sent over wireless channels,
and NOMA is used to exploit the power domain for superposition transmission. Each transmitter is equipped with one vertically
polarized (v-polarization) antenna and one horizontally polarized
(h-polarization) antenna, respectively.
We denote the number of users (receivers) as N, and each is
also equipped with a pair of orthogonally polarized antennas. The
watermark-encoded data intended for user k are denoted as xk, k
= 1, 2, · · ·, N. Then, it goes through the modulation and coding
scheme (MCS) block, wherein each data stream is encoded and
modulated separately. Herein, xk's MCS is determined by user k's
channel condition. For example, if the channel condition is good,
the transmitter will use a high-order MCS (e.g., 16 quadrature
amplitude modulation) to encode the data stream. Otherwise, the
transmitter will choose a low-order MCS (e.g., quadrature phase
shift keying) for robust data transmission. The output signal of
MCS block is denoted as sk, which represents the modulated signal for user k. Then, the transmitter will superpose N-modulated
signals with N distinct transmission power levels. In general, a

IEEE A&E SYSTEMS MAGAZINE

Table of Contents for the Digital Edition of Aerospace and Electronic Systems Magazine February 2018