Secure Space-Time Block Coding without

Transmitter CSI
T. Allen, Senior Member, IEEE, J. Cheng, Senior Member, IEEE, N. Al-Dhahir, Fellow, IEEE
AbstractA new technique is presented for securing a spacetime block code in the presence of an eavesdropper without
channel state information at the transmitter. Assuming channel
reciprocity between the transmitter and the intended receiver, we
use the receive signal strength indicator to randomly rotate the
information symbols at each transmit antenna without increasing
the transmit signals peak-to-average ratio. It is shown that the
proposed scheme guarantees full diversity to an intended
recipient while reducing the diversity order of an eavesdropper
to zero even in the presence of perfect channel state information.

I.
INTRODUCTION
Ensuring the confidentiality of communications is
fundamental to securing any network. This requirement
becomes particularly important for wireless networks where
eavesdropping is facilitated by the broadcast nature of the
wireless medium. Traditionally, security is viewed as an
independent feature addressed above the physical layer and
widely-used cryptographic protocols are typically designed
and implemented assuming an error-free physical layer.
Recently, there has been increased interest in using the
attributes of a wireless channel for security. A secret
communication technique using channel state information
(CSI) as secret key was described in [1] and generalized for
the mult-antenna scenario in [2]. In [3] secrecy was achieved
by adding artificially-generated noise to the information signal
in the null space of the intended receivers channel matrix to
degrade the eavesdroppers channel since its null space (if
any) will be different. The security provided by transmit
beamforming was shown in [4] to be highly sensitive to
inaccurate transmit CSI. In [5], a channel-dependent precoder
was used but it does not guarantee a zero diversity order for
the eavesdropper. All of the above approaches assume the
availability of transmit CSI which can be unrealistic due to the
limited resources on wireless devices. Finally, the cross-layer
approach of [6] expands the signal constellation and requires a
secret key from a higher-layer protocol.
In this letter, we propose a new technique to secure
orthogonal space-time block codes (STBC) using a shared key
obtained from RSSI and without requiring transmit CSI or
increasing the transmit signals peak-to-average ratio (PAR).
The proposed method preserves the linear detection
complexity of the orthogonal STBC while guaranteeing full
diversity order for the intended recipient and zero diversity
order for the eavesdropper.
II.

SYSTEM MODEL AND ASSUMPTIONS

A legitimate user named Alice sends digital messages to

another user, Bob, using the Alamouti STBC [7]. To enhance
transmission security, Alice applies random phase rotations to
the symbols at each transmit antenna as illustrated in Figure 1.
For the kth STBC codeword, the encoder transmits s!,! e!!!,!
from antenna one and s!,! e!!!,! from antenna two at time slot

one. At time slot two, the encoder transmits s!,!

e!!!,! from

antenna one and s!,!

e!!!,! from antenna two, resulting in the
following equivalent parameterized STBC

C s!,! , s!,! , !,! , !,! =

s1,k ej1,k s2,k ej2,k

.
s2,k ej1,k s1,k ej2,k

Alice
-S2* S1
S2 S1

Bob
q1

-S2*ejq1 S1ejq1

Alamouti
STBC

Transmitter

(1)

ML
Receiver

S2 S1

Eve
S1* S2

q2

jq2

jq2

S1*e S2 e

ML
Receiver

S2 S1

Figure 1: System Block Diagram

This parameterized STBC is a unitary transformation of the

original Alamouti STBC where orthogonality is preserved.

Moreover, by adding the two random phase rotations as
additional parameters, Alice now has a mechanism to thwart
any unintended recipients.
To initiate the transmission, Alice and Bob, assuming
channel reciprocity, use their received signal strength indicator
(RSSI) to exchange a secret key (more details in Section IIIA) which is used to synchronize the random generation of the
rotation angles between Alice and Bob. The secret key serves
as a seed to a random number generator where the output is
masked (e.g. keeping only the log2(L) most significant bits) to
produce an index associated with a rotation angle from the LPSK signal constellation. Once the generator is seeded, a
random rotation angle (assuming one of L possible values) is
generated for each transmit antenna and is updated every
STBC codeword. In this model, Bob has perfect receive CSI
but Alice has no transmit CSI knowledge.
In our adversarial model, an eavesdropper, Eve, also
receives Alices transmission. We make the following worstcase assumptions that Eve : (1) has perfect receive CSI; (2)
knows Alices signal constellation; (3) knows that Alice is
using the secure Alamouti STBC with random phase rotations
in (1) and that the phase rotations change for each STBC
codeword. However, Eve does not know the rotation angles
and is assumed to be sufficiently spatially separated from Bob
such that their channels fade independently. Throughout this
letter, vectors are underlined and matrices denoted by bold
font and (*) denotes the Hermitian (i.e. complex-conjugate)
operator.
III. PROPOSED SECURE OPEN-LOOP ALAMOUTI STBC
Alice encodes L-PSK symbols = !!!(!!!)/! ! =
1, , using the STBC C s!,! , s!,! , !,! , !,! in (1)
where !,! and !,! are randomly generated (as described
above), each from a finite set of L values unknown to Bob.

A. Encoding
The transmission is initiated by the exchange of a secret
key based on RSSI as follows: Alice and Bob, each
transmitting a known signal to each other. Alice initiates the
process and Bob responds. This two-way probing is performed
over two consecutive time slots (i.e. a duration of 2 symbol
periods which is well within the channel coherence time).
Ideally, Alice and Bob both should measure the RSSI at the
same time. However, they cannot both transmit and receive
the signals simultaneously. Thus, Alice and Bob must measure
the radio channel in one direction at a time. However, as long
as the time between two directional channel measurements is
much smaller than the channel coherence time, they will have
same RSSI estimates. A secret key is exchanged when the
communication session between Alice and Bob is established
and remains valid until the session is terminated. Once seeded,
the random generation of the parameters !,! and !,! is
known to Alice and Bob but unknown to Eve since her
channel experiences independent fading from Bobs channel.
The rotation angles !,! and !,! are chosen such that for
any choice of s!,! and s!,! from the original signal
constellation, the resulting transmitted signals are also from
the same constellation. For example, for L-PSK constellation
!"!!
symbols given by exp ! , l = 0, , 1 , we select
!
!,! = 2l/ where l = 0, , 1. In this case, the resulting
transmitted signals are also L-PSK and, therefore, the PAR is
not increased. Another benefit to the above transmission
scheme is that Alice is not required to know Bobs CSI,
eliminating the need for CSI feedback to the transmitter.
B. Decoding
The two received signal samples for the kth STBC
codeword are given by (assuming a single receive antenna)
,
, ,
=

,
,
,

, ,

,
,

(2)

where the channel matrix is orthogonal and depends on

= !,! , !,! and {hi}. In Bobs receiver, the rotation
angles are known and due to the orthogonality of , the
maximum likelihood (ML) detector is the matched filter
!
and s! = h
r! . Hence, the ML detector
of the secure STBC inherits the full diversity and linear
detection complexity properties from the Alamouti STBC.
B.2 Eves ML Detector
Eves received signal, r! , can also be modeled in a manner
similar to Bob but with independently-fading channel
coefficients {g ! } and noise samples {n!,! } resulting in Eves
version of (3); namely r! = s! + n! .
Although Eve is assumed to have perfect receive CSI,
being an unintended recipient, she does not know .
However, she knows the ! possible values of . Knowing
that changes every STBC codeword removes blind
equalization as an option for Eve. As it was shown in [9] that
several complex orthogonal STBCs (including the Alamouti
STBC) suffer from channel non-identifiability and require
training symbols (which are not available to Eve) to resolve it.
Given the received symbols ! , Eve computes her ML
estimate, denoted by ! , of the transmitted symbols given by
s! , = arg min !! r! s!

(4)

Computing this ML estimate requires an exhaustive search

over the ! pairs of information symbols together with the !
rotation angle pairs resulting in ! search complexity
which becomes formidable as L increases. To reduce this
complexity, Eve can first compute the ML estimate of s!
conditioned on the rotation angle given by the following
simple form due to the orthogonality of G
!!

where hi is the channel coefficient associated with transmit

s! |
=
r! =
g
r! .
!"
antenna , {st,k} for t=1,2 are the two information symbols, and
{nt,k} are the additive noise samples. The channel coefficients
Since Eve has apriori knowledge of Alices transmission
are assumed circularly-symmetric, independent identically
scheme, she employs a conditional ML detector where Eve
distributed (IID) complex-valued Gaussian random variables
conditions on the ! possible values of to implement !
with zero-mean and variance 1/2 per dimension. The channel
parallel matched filters. Using the normalized outputs of these
coefficients are assumed to be fixed over the channel
matched filters, Eves ML detector then selects the symbol
coherence time which spans multiple STBC codewords.
s! |
closest to s! in Euclidean distance which is then
Noise samples are modeled as circularly-symmetric, IID,
!"
complex-valued Gaussian random variables with zero-mean
and
g! ! e!(!!,! ! !!,!,! ) + g ! ! e!!(!!,! ! !!,!,! )
e!( !!,!,! ! !!,!,! ) (e!(!!,! ! !!,!,! ) e!!(!!,! ! !!,!,! ) )
variance
(6)
N0/2 per
e!!( !!,!,! ! !!,!,! ) (e!(!!,! ! !!,!,! ) e!!(!!,! ! !!,!,! ) )
g! ! e!!(!!,! ! !!,!,! ) + g ! ! e!(!!,! ! !!,!,! )
dimensio
n.
passed to an L-PSK slicer. Assuming that Eve performs her
conditional ML detection codeword by codeword with perfect
B.1 Bobs ML Detector
CSI, the outputs of the matched filters have the form
Bobs received signal, modeled by (2), can be converted to
the more convenient form
s = s + n
(5)
!,!

r!,!
h! e!!!,!
=

r!,!
h! e!!!!,!

e!!!,!

h!
h! e!!!!,!

s!,!
s!,!

r! = s! + n!

n!,!
+ n
!,!

(3)

Simplifying the product ! yields the matrix in (6)

where the off-diagonal elements represent inter-codeword
interference that only vanishes when ! = . Conceptually,
we can see from (1) that the rotation angles effectively apply a

phase shift to the assumed known channel coefficients in

Eves ML detector. As a result, even with perfect CSI, the
addition of the unknown random phase rotations results in Eve
having a partially-coherent ML detector which severely
degrades her ability to extract the information carried in the
phase of the signal.
C. Remarks
1) In the Appendix, we show that the diversity order of
Eves ML detector decreases by 1 (compared to Bobs full
diversity order) when a rotation angle is applied to a transmit
antenna. Hence, when phase rotations are only applied to a
subset of the transmit antennas, Eve can improve her
performance by increasing her transmit power or adding more
receive antennas. However, both of these measures fail when
the phase rotations are applied to all transmit antennas
resulting in a diversity order of zero for Eves ML detector.
Proposition 3 in the Appendix provides an example of when
Eves codeword difference matrix has zero rank. The example
shows that, even in the noiseless case, Eves ML detector is
not able to distinguish between symbols with rotations of 0
and . As a result, Eve experiences an irreducible error floor .
2) To exchange a secret key based on RSSI, Alice and
Bob perform a two-way probing over two consecutive time
slots. As long as the channel is fixed over these two time slots,
(which is also a key assumption of the Alamouti scheme [5])
then due to channel reciprocity, Alice and Bob both measure
the same RSSI. Eve listens to both transmissions, but due to
her spatial separation from Bob, her RSSI is independent of
Alices and Bobs RSSI. The RSSI quantizer design as well
as the length of the secret key were left as design trade-off
issues for future work. Increasing the number of probes used
in the construction of the secret key would incur additional
overhead on the transmission rate while providing additional
security. Furthermore, the rate at which secret bits can be
extracted from the channel is fundamentally limited by the
channel coherence time. In highly-dynamic environments, the
reduced channel coherence time will decrease the rate at
which secret bits can be extracted by Alice and Bob from the
channel, but is mitigated by that fact that the key is exchanged
only once at the beginning of the communication session.
3) For M-QAM constellations, the rotation angles take on
the values = 0, 2 , , 3 2 to avoid increasing the PAR.
Using these rotation angles, the diversity orders of Bobs and
Eves ML detectors are the same as in the L-PSK case.
4) The number of possible phase rotations grows with the
square of the signal constellation size. However, because the
phase rotations associated with BPSK, i.e. = 0, , are
common to all other L-PSK constellations, they provide a
minimal set of random phase rotations to guarantee a diversity
order of zero for Eves ML detector while simplifying Bobs
RSSI quantization to two levels only.
5) For 4 transmit antennas, we propose the following
secure rate Octonion orthogonal STBC [10]

C s!,! , s!,! , s!,! , !,! , !,! , !,! , !,! =

s!,! e!!!,! s!,! e!!!,! s!,! e!!!,!
0

s!,!
e!!!,!

s!,!
e!!!,!

s!,! e!!!,!

s!,!
e!!!,!

s!,! e!!!,!

s!,!
e!!!,!

s!,!
e!!!,!

s!,!
e!!!,!

(7)

s!,! e!!!,!

r! = C s!,! , s!,! , s!,! , !,! , !,! , !,! , !,! h + n! .

Eves ML detector must now compare L! symbol triplets
together with L! phase combinations for a total complexity
proportional to L! . We conclude this section by emphasizing
that even at this formidable complexity, Eves ML detector
diversity order is still zero (see Table 1).
IV. SIMULATION RESULTS
Simulation results of bit error probability (BEP) vs. signal
to noise ratio (SNR) for both the 2x1 (Alamouti) and 4x1
Octonion secure STBCs are provided in Figure 2. Notice that
the addition of each rotation angle reduces Eves ML detector
diversity order by 1 (compared to Bobs full ML diversity
order) resulting in a diversity order of zero when random
rotations are applied to all transmit antennas.
V. CONCLUSION
We described a new scheme to secure STBCs without
transmitter CSI. This technique has been applied to L-PSK as
well as M-QAM signals for two and four transmit antennas. It
was shown that the ML detectors diversity order is zero for
the eavesdropper even in the presence of perfect CSI while
still guaranteeing full diversity for the intended receiver.
VI.

APPENDIX : ML RECEIVER DIVERSITY ORDER

To calculate the ML receiver diversity order, we will

consider the codeword difference matrix (CWDM) [8]
B C! , C! = C s!,! , s!,! , !,! , !,! C s!,! , s!,! , !,! , !,!
for
any
two
distinct
codewords;
i.e.
when s!,! , s!,! , !,! , !,! s!,! , s!,! , !,! , !,! .
1) Bobs ML Receivers Diversity Order
For
Bob,
the
rotations
are
known,
hence, !,! = !,! and !,! = !,! . To show that the secure
STBC codeword C! provides full diversity when !,! and !,!
are known to both the transmitter and receiver, we need to
show that the CWDM is a non-singular matrix (i.e. has full
rank).
Proposition 1: The secure STBC encoding scheme has full
ML diversity order of two when !,! = !,! and !,! = !,! .
Proof: In this case, we have
B C! , C! =

s!,! s!,! e!!!,!

!!
!,!

s!,! s!,! e

s!,! s!,! e!!!,!

s!,! s!,! e!!!,!

Assume that the CWDM is singular; hence, its determinant

!

is equal to zero. It follows that s!,! s!,! = 0 which is only

true when s!,! = s!,! i. This contradicts the definition of the
CWDM and proves that our assumption is false.

Proposition 3: The secure STBC encoding scheme has ML

diversity of zero when !,! !,! and !,! !,! .
Proof:
Consider the two distinct codewords
C! = 1, 1, , 0 and C! = 1, 1,0, which are also common
to all L-PSK constellations. By direct substitution, we find
that the CWDM is the all-zero matrix. Hence, Eves ML
diversity order is zero with two rotations for all L-PSK
constellations.
4) Eves ML Diversity Order for Rate- Octonion STBC
The ML diversity orders of Bob and Eve when using the rate
secure STBC for four transmit antennas in Equation (7) can
be derived using the same approach described above and are
summarized in Table 1 (details omitted due to lack of space).
# Phase
Rotations
1

Bobs
Diversity
4

Eves
Diversity
3

Codewords w/
Minimum Rank
C! = 1, 1,1, , 0, 0, 0
C! = 1, 1, 1, 0, 0, 0, 0
C! = 1, 1,1, , , 0, 0
C! = 1, 1, 1, 0, 0, 0, 0
C! = 1, 1,1, , , , 0
C! = 1, 1, 1, 0, 0, 0, 0
C! = 1, 1,1, , , ,
C! = 1, 1, 1, 0, 0, 0, 0

Table 1: ML Diversity Orders for Secure STBC in (7)

REFERENCES
H. Koorapaty, A. A. Hassan, S. Chennakeshu Secure information
transmission for mobile radio, IEEE Trans. Wireless Communications,
pp. 52-55, July 2003
[2] A. E. Hero, Secure space-time communication, IEEE Trans.
Information Theory, pp. 3235-3249, Dec. 2003.
[3] R. Negi, S. Goel, Secret communication using artificial noise, IEEE
Vehicular Technology Conference, Fall 2005. vol. 3, pp. 1906 1910.
[4] A. Mukherjee and A. Swindlehurst, Robust beamforming for security
in MIMO wiretap channels with imperfect CSI, IEEE Trans. Signal
Processing, vol. 59, pp. 351 361, Jan. 2011.
[5] Fakoorian, S.A.A.; Jafarkhani, Hamid; Swindlehurst, A.L., "Secure
space-time block coding via artificial noise alignment," Signals, Systems
and Computers (ASILOMAR), Conference Record of the Forty Fifth
Asilomar Conference on , pp.651-655, Nov. 2011
[6] H. Wen, G. Gong and P.-H. Ho, ``MIMO cross-layer secure
communication architecture based on STBC,'' in Proc. of IEEE
Globecom conference, pp.1-5, Miami, FL, USA, Dec. 2010
[7] S. Alamouti, A simple transmitter diversity scheme for wireless
communications, IEEE JSAC, vol. 16, pp. 1451 1458, Oct. 1998.
[8] V. Tarokh, H. Jafarkhani, and A. R. Calderbank, Space-time block
codes from orthogonal designs, IEEE Trans. Information Theory, vol.
45, pp. 14561467, July 1999.
[9] S. Shahbazpanahi, A. Gershman, and A. Manton, Closed-form blind
MIMO channel estimation for orthogonal spacetime block codes,
IEEE Trans. Signal Procesing, vol. 53, pp 4506-4517, Dec. 2005
[10] S. Diggavi, N. Al-Dhahir, A. Stamoulis, A.R. Calderbank,Great
Expectations : The Value of Spatial Diversity in Wireless Networks,
Proceedings of the IEEE journal, vol. 92, pp 219-270, Feb. 2004
[1]

Figure 2: ML Detector BEP for Bob and Eve

2) Eves ML Receivers Diversity Order (1 phase rotation)

Consider the CWDM when !,! is not known by the receiver
and !,! = 0, meaning that !,! !,! and !,! = !,! = 0.
Proposition 2: The secure STBC encoding scheme has ML
diversity order equal to 1 when !,! !,! and !,! = !,! = 0.
Proof: For any two distinct codewords, the CWDM is not
equal to the all-zero matrix. Hence, the rank cannot be zero.
Now, consider the distinct codewords C! = 1, 1,0, 0 and C! =
1, 1, , 0 . By direct substitution, we find that the CWDM
has a rank of unity. Therefore, using the STBC design rank
criterion [6], we conclude that the ML diversity order is one.
3) Eves ML Receivers Diversity Order (2 phase rotation)
Consider the CWDM when !,! and !,! are not known at the
receiver, meaning that !,! !,! and !,! !,! .