0% found this document useful (0 votes)
55 views9 pages

Secure Joint Communication and Sensing

This document summarizes a research paper on secure joint communication and sensing systems. It introduces a model where a transmitter simultaneously communicates reliably with a legitimate receiver while estimating the channel state, and hides part of the transmitted message from an eavesdropper receiver. The model considers independent and identically distributed channel states that are known to the corresponding receivers through feedback. The paper develops inner and outer bounds for the achievable secrecy-distortion region under this partial secrecy scenario. The bounds match when the channel is physically or reversely physically degraded, characterizing the optimal tradeoff between communication rate, state estimation distortion, and information leakage.

Uploaded by

Osama AL Tayar
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
55 views9 pages

Secure Joint Communication and Sensing

This document summarizes a research paper on secure joint communication and sensing systems. It introduces a model where a transmitter simultaneously communicates reliably with a legitimate receiver while estimating the channel state, and hides part of the transmitted message from an eavesdropper receiver. The model considers independent and identically distributed channel states that are known to the corresponding receivers through feedback. The paper develops inner and outer bounds for the achievable secrecy-distortion region under this partial secrecy scenario. The bounds match when the channel is physically or reversely physically degraded, characterizing the optimal tradeoff between communication rate, state estimation distortion, and information leakage.

Uploaded by

Osama AL Tayar
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

Secure Joint Communication and Sensing

Onur Günlü1 , Matthieu Bloch2 , Rafael F. Schaefer1, and Aylin Yener3


1
Chair of Communications Engineering and Security, University of Siegen, {onur.guenlue, rafael.schaefer}@uni-siegen.de
2
School of Electrical and Computer Engineering, Georgia Institute of Technology, [email protected]
3
Department of Electrical and Computer Engineering, The Ohio State University, [email protected]

Abstract—This work considers mitigation of information leak- output. Follow up works have extended the model to multiple
age between communication and sensing operations in joint access channels [8] and broadcast channels [9].
communication and sensing systems. Specifically, a discrete mem- The nature of JCAS mandates the use of a single modality
arXiv:2202.10790v1 [cs.IT] 22 Feb 2022

oryless state-dependent broadcast channel model is studied in


which (i) the presence of feedback enables a transmitter to for the communication and sensing functions so that sensing
simultaneously achieve reliable communication and channel state signals carry information, which then creates situations in
estimation; (ii) one of the receivers is treated as an eavesdrop- which leakage of sensitive information can occur. For example,
per whose state should be estimated but which should remain a target illuminated for sensing its range has the ability to
oblivious to a part of the transmitted information. The model gather potentially sensitive information about the transmitted
abstracts the challenges behind security for joint communication
and sensing if one views the channel state as a characteristic message [10]. As the sensing performance and secrecy perfor-
of the receiver, e.g., its location. For independent identically mance are both measured with respect to the signal received at
distributed (i.i.d.) states, perfect output feedback, and when part the sensed target, there exists a trade-off between the two [2].
of the transmitted message should be kept secret, a partial To capture and characterize this trade-off, we extend the JCAS
characterization of the secrecy-distortion region is developed. The model in [7] by introducing an eavesdropper in the network.
characterization is exact when the broadcast channel is either
physically-degraded or reversely-physically-degraded. The char- The objective of the transmitter is then to simultaneously
acterization is also extended to the situation in which the entire communicate reliably with the legitimate receiver, estimate
transmitted message should be kept secret. The benefits of a joint the channel state, and hide a part of the message from
approach compared to separation-based secure communication the eavesdropper. The channel state is modeled as a two-
and state-sensing methods are illustrated with a binary joint component state capturing the characteristics of each individual
communication and sensing model.
receiver, the feedback is modeled as perfect output feedback
for simplicity, and the transmitted message is divided into two
I. I NTRODUCTION parts, only one of which should be kept (strongly) secret (this
is called partial secrecy in [11]). We develop inner and outer
The vision for next generation mobile communication net- bounds for the secrecy-distortion region of this partial-secrecy
works includes a seamless integration of the physical and scenario under a strong secrecy constraint when i.i.d. channel
digital world. Key to its success is the network’s ability to states are causally available at the corresponding receivers.
automatically react to changing environments thanks to tight The bounds match when the JCAS channel is physically- or
harmonization of communication and sensing [1]. For instance, reversely-physically-degraded and the outer bound also applies
a mmWave joint communication and radar system can be used to the case of noisy generalized feedback. We also extend these
to detect a target or to estimate crucial parameters relevant to characterizations to the case in which the entire transmitted
communication and adapt the communication scheme accord- message should be kept secret. The proposed secure JCAS
ingly [2]. Joint communication and sensing (JCAS) techniques models can be viewed as extensions of the wiretap channel
are envisioned more broadly as key enablers for a wide range with feedback models [12]–[19]. Our achievability proof lever-
of applications, including connected vehicles and drones. ages the output statistics of random binning (OSRB) method
Several information-theoretic studies of JCAS have been [20]–[22] to obtain strong secrecy. A binary JCAS channel
initiated, drawing on existing results for joint communication example with multiplicative Bernoulli states illustrates how
and state estimation [3]–[6]. Motivated by the integration secure JCAS methods may outperform separation-based secure
of communication and radar for mmWave vehicular applica- communication and state-sensing methods.
tions, [7] considers a model in which messages are encoded
and sent through a state-dependent channel with generalized II. P ROBLEM D EFINITION
feedback both to reliably communicate with a receiver and We consider the secure JCAS model shown in Fig. 1,
to estimate the channel state by using the feedback and which includes a transmitter equipped with a state estimator, a
transmitted codewords. The optimal trade-off between the legitimate receiver, and an eavesdropper (Eve). The transmitter
communication rate and channel-state estimation distortion is attempts to reliably transmit a uniformly distributed message
then characterized for memoryless JCAS channels and i.i.d. M = (M1 , M2 ) ∈ M = M1 × M2 through a memo-
channel states that are causally available at the receiver and ryless state-dependent JCAS channel with known statistics
estimated at the transmitter by using a strictly causal channel PY1 Y2 Z|S1 S2 X and i.i.d. state sequence (S1n , S2n ) ∈ S1n × S2n
 The use of per-letter distortion metrics dj (·, ·) in conjunction
c = Dec(Y1n , S1n )
M c= M
M c1 , M
c2
cn = Estj (X n , Z n )
S j
with i.i.d. states simplifies the problem to a rate distortion
S1,i region characterization [7]–[9]; in fact, past observations are
Y1,i
independent of present and future ones, lending the transmitter
Zi−1 S1,i no state prediction ability to adapt its transmission on the fly.
Xi PY1 Y2 Z|S1 S2 X PS1 S2 Analyzing JCAS models with memory leads to conceptually
S2,i
different results, see, e.g., [23].
Y2,i
Xi = Enci (M, Z i−1 ) S2,i Remark 1. The strong secrecy condition (4) is equivalent
EVE to I(M2 ; Y2n , S2n ) ≤ δ since the transmitted message is
independent of the state sequence.
M = (M1 , M2 )
III. B OUNDS FOR JCAS WITH PARTIAL -S ECRECY
Fig. 1. JCAS model with partial secrecy, where only M2 should be kept
secret from Eve, for j = 1, 2 and i = [1 : n]. We mainly consider JCAS
We next provide inner and outer bounds for the secrecy-
with perfect output feedback, where Zi−1 = (Y1,i−1 , Y2,i−1 ). distortion region RPS,POF ; see Section VI for a proof sketch.
Define [a]+ = max{a, 0} for a ∈ R.
Proposition 1 (Inner Bound). RPS,POF includes the union
generated according to a known joint probability distribution
over all joint distributions PU PV |U PX|V of the rate tuples
PS1 S2 . The transmitter calculates the channel inputs X n as
(R1 , R2 , D1 , D2 ) such that
Xi = Enci (M, Z i−1 ) ∈ X for all i = [1 : n], where Enci (·)
is an encoding function and Z i−1 ∈ Z i−1 is the delayed R1 ≤ I(U ; Y1 |S1 ) (6)
channel output feedback. The legitimate receiver that observes R2 ≤ min{R2′ ,
(I(V ; Y1 |S1 ) − R1 )} (7)
Y1,i ∈ Y1 and S1,i for all channel uses i = [1 : n] should
b
Dj ≥ E[dj (Sj , Sj ))] for j = 1, 2 (8)
reliably decode both M1 and M2 by forming the estimate
c = Dec(Y1n , S1n ), where Dec(·) is a decoding function.
M where
The eavesdropper that observes Y2,i ∈ Y2 and S2,i should
be kept ignorant of M2 . Finally, the transmitter estimates the PUV XY1 Y2 S1 S2 = PU|V PV |X PX PS1 S2 PY1 Y2 |S1 S2 X , (9)
n V

cn = Estj (X n , Z n ) ∈ Sj for
state sequence (S1n , S2n ) as S j
j = 1, 2, where Estj (·, ·) is an estimation function. Unless R2′ = [I(V ; Y1 |S1 , U ) − I(V ; Y2 |S2 , U )]+
specified otherwise, all sets S1 , S2 , Sb1 , Sb2 , X , Y1 , Y2 , and Z + H(Y1 |Y2 , S2 , S1 , V ) (10)
are finite.
For simplicity, we consider the perfect output feedback case and Est∗j (x, y1 , y2 )
= ŝj for j = 1, 2 are per-letter state
in which for all i = [2 : n] we have estimators such that dj (x, y1 , y2 ) is equal to
X
Zi−1 = (Y1,i−1 , Y2,i−1 ). (1) argmin PSj |XY1 Y2 (sj |x, y1 , y2 ) dj (sj , s̃). (11)
b
s̃∈Sj sj ∈Sj
Although this is explicitly used in our achievability proofs, One can limit |U| to
some of our converse results hold for generalized feedback. We
next define the strong secrecy-distortion region for the problem (min{|X |, |Y1 |, |Y2 |}+2) (12)
of interest.
and |V| to
Definition 1. A secrecy-distortion tuple (R1 , R2 , D1 , D2 ) is
(min{|X |, |Y1 |, |Y2 |}+2) · (min{|X |, |Y1 |, |Y2 |}+1). (13)
achievable if, for any δ > 0, there exist n ≥ 1, one encoder,
cn such Proposition 2 (Outer Bound). RPS,POF is included in the union
one decoder, and two estimators Estj (X n , Y1n , Y2n ) = S j
that over all joint distributions PV X of the rate tuples in (8) and
1 R1 ≤ I(V ; Y1 |S1 ) (14)
log |Mj | ≥ Rj − δ for j = 1, 2 (rates) (2) n
n  
 R2 ≤ min H(Y1 , S1 |Y2 , S2 ) − H(S1 |Y2 , S2 , V ) ,
Pr M 6= M c ≤δ (reliability) (3)
o
I(M2 ; Y2n |S2n ) ≤ δ (strong secrecy) (4) I(V ; Y1 |S1 ) − R1 (15)
 n c

E dj (Sj , Sjn ) ≤ Dj +δ for j = 1, 2 (distortions) (5) where (9) with constant U follows and we can apply the
P n deterministic per-letter estimators Est∗j (x, y1 , y2 ) = ŝj for
where dj (sn , sc
n) = 1 dj (s i , b
s i ) for j = 1, 2 are bounded
n i=1 j = 1, 2 by using (11). One can limit the cardinality to
per-letter distortion metrics.
|V| ≤ (min{|X |, |Y1 |, |Y2 |}+1). (16)
The secrecy-distortion region RPS,POF is the closure of the
set of all achievable tuples with partial secrecy and perfect Remark 2. Since we consider perfect feedback as in (1),
output feedback. the outer bound proposed in Proposition 2 is also valid for
the general JCAS problem depicted in Fig. 1, in which the form a Markov chain for these JCAS scenarios. Define
feedback Zi−1 can be any noisy version of (Y1,i−1 , Y2,i−1 ). ′
R2,deg = [I(V ; Y1 |S1 ) − I(V ; Y2 |S2 )]+ + H(Y1 |Y2 , S2 , S1 , V )
We next characterize the exact strong secrecy-distortion (a)
regions for physically-degraded and reversely-physically- = H(V |Y2 , S2 ) − H(V |Y1 , S1 ) + H(Y1 |Y2 , S2 , S1 , V )
degraded JCAS channels with partial secrecy and perfect = H(Y1 , V |Y2 , S2 , S1 ) − H(V |Y1 , S1 ) + H(S1 |Y2 , S2 )
output feedback, defined below; see also [9, Definition 2]. − H(S1 |Y2 , S2 , V )
Definition 2. A JCAS channel PY1 Y2 |S1 S2 X is physically- = H(V |Y2 , S2 , S1 , Y1 ) − H(V |Y1 , S1 ) + H(Y1 |Y2 , S2 , S1 )
degraded if we have + H(S1 |Y2 , S2 ) − H(S1 |Y2 , S2 , V )
PY1 Y2 |S1 S2 X = PS1 PY1 |S1 X PY2 S2 |S1 Y1 . (17) (b)
= H(Y1 , S1 |Y2 , S2 ) − H(S1 |Y2 , S2 , V ) (25)
The channel is reversely-physically-degraded if the degradation
where (a) follows by (23) and (b) follows from the Markov
order is changed and
chain in (24).
PY1 Y2 |S1 S2 X = PS2 PY2 |S2 X PY1 S1 |S2 Y2 . (18) Applying the Fourier-Motzkin elimination [24] to (19)-(22),
for any ǫ > 0 one can achieve
The physically-degraded corresponds to a situation in which
the observations (Y2n , S2n ) of the eavesdropper are degraded R1 = Rv1 = I(V ; Y1 , S1 ) − 2ǫ = I(V ; Y1 |S1 ) − 2ǫ (26)
versions of observations (Y1n , S1n ) of the legitimate receiver
and for any R1 that is less than or equal to (26), one can
with respect to the channel input X n .
achieve
Theorem 1. RPS,POF for a physically-degraded JCAS problem ′
with partial secrecy and perfect output feedback is the region R2 = Rv2 +Rv = min{R2,deg , (I(V ; Y1 |S1 )−R1 )}−3ǫ. (27)
defined in Proposition 2. Furthermore, the proofs for achievable distortions, sufficiency
Proof Sketch: Since the outer bound given in Proposi- of given deterministic estimators, inversion of the problem
tion 2 does not assume any degradedness, the converse proof in the source model into the problem in the channel model,
for Theorem 1 follows from the outer bound. Furthermore, the and elimination of the public indices follow similarly as in
achievability proof for Theorem 1 follows by modifying the Section VI-A, so we omit them.
proof of the inner bound in Proposition 1. We next provide a Lemma 1. RPS,POF for a reversely-physically-degraded JCAS
sketch of the modifications for a physically-degraded JCAS. problem with partial secrecy and perfect output feedback is
First, U n is not used, i.e., U n is eliminated from the the union over all joint distributions PV X of the rate tuples
achievability proof. Second, to each v n (k) we assign four satisfying (8), (14), and
random bin indices (Fv (k), Wv1 (k), Wv2 (k), Lv (k)) such that  
e R2 ≤ min H(Y1 |Y2 , S2 , S1 ), I(V ; Y1 |S1 ) − R1 (28)
Fv (k) ∈ [1 : 2nRv ], Wv1 (k) ∈ [1 : 2nRv1 ], Wv2 (k) ∈ [1 :
nRv2
2 ], and Lv (k) ∈ [1 : 2nRv ] for all k = [1 : b] independently for joint distributions as in (9) but with constant U , and we can
such that M1 (k) = Wv1 (k) and M2 (k) = (Wv2 (k), Lv (k)). As apply the deterministic per-letter estimators Est∗j (x, y1 , y2 ) =
in (53), we impose the reliability constraint ŝj for j = 1, 2 by using (11). One can limit the cardinality to
ev > H(V |Y1 , S1 )
R (19)
|V| ≤ min{|X |, |Y1 |, |Y2 |}. (29)
as in (55) and (56) we impose the strong secrecy constraints
Proof Sketch: The proof follows by showing that the
ev < H(V |Y2 , S2 )
Rv2 + R (20) inner and outer bounds in Propositions 1 and 2, respectively,
n
Rv < H(Y1 |Y2 , S2 , S1 , V ) (21) match after elimination of U from the proof of achievability,
as in the proof for Theorem 1 above. After removal of U , i.e.,
and as in (57) we impose the mutual independence and U is constant, by (7) we have
uniformity constraint
(a)  
ev + Rv < H(V ).
Rv1 + Rv2 + R (22) R2 ≤ min H(Y1 |Y2 , S2 , S1 , V ), I(V ; Y1 |S1 ) − R1
(b)  
We remark that we have H(V |Y2 , S2 ) ≥ H(V |Y1 , S1 ) for = min H(Y1 |Y2 , S2 , S1 ), I(V ; Y1 |S1 ) − R1 (30)
all physically-degraded JCAS channels, i.e., we obtain where (a) follows since V is independent of (S , S ) and since
1 2
[I(V ; Y1 |S1 ) − I(V ; Y2 |S2 )]+ H(V |Y1 , S1 ) ≥ H(V |Y2 , S2 ) for all reversely-physically-
(a) degraded JCAS channels because of the Markov chain
= H(V |Y2 , S2 ) − H(V |Y1 , S1 ) (23)
V − X − (Y2 , S2 ) − (Y1 , S1 ) (31)
where (a) follows because V is independent of (S1 , S2 ) and
since and (b) follows because of the Markov chain
V − X − (Y1 , S1 ) − (Y2 , S2 ) (24) V − X − (Y2 , S2 , S1 ) − Y1 . (32)
Furthermore, by (15) we obtain output feedback are satisfied for some tuple (R, D1 , D2 ). We
then obtain
(a) n 
R2 ≤ min H(Y1 , S1 |Y2 , S2 , V ) − H(S1 |Y2 , S2 , V ) , (a)
nR ≤ I(M ; Y1n |S1n ) + nǫn
o n
I(V ; Y1 |S1 ) − R1 (b) X
≤ H(Y1,i |S1,i ) − H(Y1n |S1n , X n ) + nǫn
(b)  
= min H(Y1 |Y2 , S2 , S1 ), I(V ; Y1 |S1 ) − R1 (33) i=1
n
X
(c) 
where (a) follows from the Markov chain in (31) and (b) = H(Y1,i |S1,i ) − H(Y1,i |S1,i , Xi ) + ǫn
i=1
follows from the Markov chain in (32). n
X
= (I(Xi ; Y1,i |S1,i ) + ǫn ) (38)
IV. B OUNDS FOR JCAS WITH S INGLE S ECURE M ESSAGE i=1

We next give inner and outer bounds for the JCAS problem where (a) follows because M and S1n are independent, and
with perfect output feedback, in which there is a single mes- from Fano’s inequality and (3) for an ǫn > 0 such that ǫn → 0
sage M = M2 that should be kept secret from an eavesdropper, if δn → 0, which is similar to (60) below, (b) follows since
i.e., M1 = ∅ in Fig. 1. For this problem, the definitions M − (X n , S1n ) − Y1n form a Markov chain, and (c) follows
of an achievable secrecy-distortion tuple (R, D1 , D2 ) and because the JCAS channel is memoryless and state sequence
corresponding strong secrecy-distortion region RPOF follow is i.i.d. Furthermore, we also have
similarly as in Definition 1 by eliminating (M1 , R1 ) and by (a)
replacing (M2 , R2 , RPS,POF ) with (M, R, RPOF ), respectively. nR ≤ I(M ; Y1n , Y2n , S1n , S2n ) + nǫn
= H(Y1n , S1n |Y2n , S2n ) + I(Y2n , S2n ; M )
Proposition 3. (Inner Bound): RPOF includes the union over
all joint distributions PV X of the rate tuples (R, D1 , D2 ) − H(Y1n , S1n |Y2n , S2n , M ) + nǫn
satisfying (8) and (b) n
X
≤ H(Y1,i , S1,i |Y2,i , S2,i ) + δn
R ≤ min{R2′′ , I(V ; Y1 |S1 )} (34) i=1
− H(S1n |Y2n , S2n , X n ) + nǫn
where (c)
n
X
= H(Y1,i , S1,i |Y2,i , S2,i ) + δn
PV XY1 Y2 S1 S2 = PV |X PX PS1 S2 PY1 Y2 |S1 S2 X , (35) i=1
n
X
− H(S1,i |Y2,i , S2,i , Xi ) + nǫn (39)
R2′′ = [I(V ; Y1 |S1 ) − I(V ; Y2 |S2 )]+ i=1

(36) where (a) follows from Fano’s inequality and (3) for an
+ H(Y1 |Y2 , S2 , S1 , V )
ǫn > 0 such that ǫn → 0 if δn → 0, (b) follows by
and apply the deterministic per-letter estimators (4) and from Remark 1 after replacing M2 with M for the
Est∗j (x, y1 , y2 ) = ŝj for j = 1, 2 by using (11). One JCAS problem with perfect output feedback, and because
can limit the cardinality as in (16). M −(Y2n , S2n , X n )−S1n form a Markov chain, and (c) follows
because the JCAS channel is memoryless and state sequence
Proposition 4. (Outer Bound): RPOF is included in the union is i.i.d. Thus, by applying (66) given below and introducing
over all PX of the rate tuples satisfying (8) and a uniformly-distributed time-sharing random variable, as being
n  applied in the proof of outer bound for Proposition 2, we prove
R ≤ min H(Y1 , S1 |Y2 , S2 ) − H(S1 |Y2 , S2 , X) , the outer bound for the JCAS problem with perfect output
o
I(X; Y1 |S1 ) (37) feedback by letting δn → 0.
Similar to Section III, we characterize the exact strong
where we can apply the deterministic per-letter estimators secrecy-distortion regions for the JCAS problem with per-

Estj (x, y1 , y2 ) = ŝj for j = 1, 2 by using (11). fect output feedback when the JCAS channel PY1 Y2 |S1 S2 X
is physically-degraded, as in (17), or reversely-physically-
Proof Sketch: The proof of the inner bound in Propo- degraded, as in (18).
sition 3 follows by eliminating U n in the proof of the inner
Theorem 2. RPOF for a physically-degraded JCAS problem
bound for Proposition 1 such that R1 = Rv1 = 0 and by
with perfect output feedback is the region defined in Proposi-
imposing (19)-(22) after replacing Rv2 with Rv since for this
tion 4.
case we have M (k) = (Wv (k), Lv (k)) for all k = [1 : b].
We next prove the outer bound. Assume that for some δn > 0 Proof: Since the outer bound given in Proposition 4 is
and n ≥ 1, there exist an encoder, decoder, and estimators such valid for any JCAS channel, the converse proof for Theorem 2
that all constraints imposed on the JCAS problem with perfect follows from Proposition 4. Furthermore, the achievability
proof for Theorem 2 follows by modifying the proof of the in- Proof Sketch: The proof follows by evaluating the strong
ner bound for Theorem 1 such that we assign V n (k) = X n (k) secrecy-distortion region RPOF defined in Theorem 2 since the
for all k = [1 : b] and then apply the same OSRB steps for JCAS channel considered is physically-degraded. Proofs for
X n (k) rather than V n (k). (46) and (47) follow by choosing Est∗j (1, y1 , y2 ) = yj and
Est∗j (0, y1 , y2 ) = 1{Pr[Sj = 1] > 0.5} for j = 1, 2 that
Lemma 2. RPOF for a reversely-physically-degraded JCAS
result from (11), which are equivalent to the proofs for [9,
problem with perfect output feedback is the union over all PX
Eqs. (27c) and (27d)]. We next have I(X; Y1 |S1 ) = qHb (p),
of the rate tuples in (8) and
which is equivalent to the proof for [9, Eq. (27a)] with r = 1.

R ≤ min H(Y1 |Y2 , S2 , S1 ), I(X; Y1 |S1 ) (40) Furthermore, we obtain
where we can apply the deterministic per-letter estimators H(Y1 , S1 |Y2 , S2 ) − H(S1 |Y2 , S2 , X)
Est∗j (x, y1 , y2 ) = ŝj for j = 1, 2 by using (11). (a)
= H(S1 |S2 ) + H(Y1 |S1 , Y2 , S2 ) − H(S1 |S2 )
Proof Sketch: We show that the inner and outer bounds (b)
in Propositions 3 and 4, respectively, match after assigning = PS1 S2 (1, 0)H(Y1 |S1 = 1, S2 = 0)
V n = X n in the proof of achievability, i.e., we choose V = X (c)
= PS1 S2 (1, 0)H(X) = q(1 − α)Hb (p) (48)
that is allowed by (35), such that by (34) we obtain
(a)  where (a) follows since S1 − S2 − (Y2 , X) form a Markov
R ≤ min H(Y1 |Y2 , S2 , S1 , X), I(X; Y1 |S1 ) chain for the considered JCAS channel, (b) follows since if
(b)  S1 = 0, then Y1 = 0, and if (S1 , S2 ) = (1, 1), then Y1 = Y2 ,
= min H(Y1 |Y2 , S2 , S1 ), I(X; Y1 |S1 ) (41)
and (c) follows since Y1 = X that is because of S1 = 1 and
where (a) follows since X is independent of (S1 , S2 ) and since since X is independent of (S1 , S2 ). Therefore, we have
H(X|Y1 , S1 ) ≥ H(X|Y2 , S2 ) for all reversely-physically- n 
degraded JCAS channels because of the Markov chain in R ≤ min H(Y1 , S1 |Y2 , S2 ) − H(S1 |Y2 , S2 , X) ,
(31) and (b) follows because of the Markov chain in (32). o
I(X; Y1 |S1 )
Furthermore, by (37) we have
(a) n  = q(1 − α)Hb (p) (49)
R ≤ min H(Y1 , S1 |Y2 , S2 , X) − H(S1 |Y2 , S2 , X) ,
o which follows since α ≤ 1.
I(X; Y1 |S1 ) We remark that for the considered example, the rate
(b)  of the securely transmitted message  is upper bounded by
= min H(Y1 |Y2 , S2 , S1 ), I(V ; Y1 |S1 ) (42) H(Y1 , S1 |Y2 , S2 )−H(S1 |Y2 , S2 , X) rather than I(X; Y1 |S1 ),
the latter of which is the upper bound for the rate for the same
where (a) follows from the Markov chain in (31) and (b)
example when there is no secrecy constraint [9, Corollary 4].
follows from the Markov chain in (32).
Thus, the amount of rate loss due to the strong secrecy
V. B INARY JCAS C HANNEL WITH M ULTIPLICATIVE constraint is qαHb (p) for this JCAS example. Furthermore,
B ERNOULLI S TATES E XAMPLE one can show that JCAS methods achieve significantly better
We next consider a JCAS with perfect output feedback performance than separation-based secure communication and
example, in which JCAS channel input and output alphabets state-sensing methods. First, one can show that the maximum
are binary with multiplicative Bernoulli states, i.e., we have secure communication rate in (45) is achieved with p = 0.5,
whereas the minimum distortions in (46) and (47) are achieved
Y1 = S1 · X, Y2 = S2 · X (43) with p = 1 that results in zero communication rate. Then,
applying time sharing between the tuples achieved by the
where PS1 S2 (0, 0) = (1 − q), PS1 S2 (1, 1) = qα, and
separation based methods to convexify and enlarge the region,
PS1 S2 (1, 0) = q(1 − α) for fixed q, α ∈ [0, 1], so the JCAS
we observe that the secrecy-distortion region that can be
channel satisfies (17) [9, Section IV-A].
achieved by applying the JCAS methods is strictly larger than
Define the binary entropy function, for any c ∈ [0, 1], as
the region being achieved by the separation based methods.
Hb (c) = −c log c − (1 − c) log(1 − c). (44) These analyses are analogous to the comparisons between joint
and separation-based secrecy and reliability methods for the
Lemma 3. The strong secrecy-distortion region RPOF for a secret key agreement problem, as discussed in [25]–[27].
binary JCAS channel with multiplicative Bernoulli states char-
acterized by parameters (q, α), and with Hamming distortion VI. P ROOFS FOR P ROPOSITIONS 1 AND 2
metrics is the union over all p = Pr[X = 1] of the rate tuples
A. Inner Bound
R ≤ q(1 − α)Hb (p) (45)
Proof Sketch: We use the OSRB method [21], [22] for
D1 ≥ (1 − p) · min{q, (1 − q)} (46) the achievability proofs, applying the steps in [28, Section 1.6],
D2 ≥ (1 − p) · min{qα, (1 − qα)}. (47) see also [29].
Since in the JCAS problem with partial secrecy and perfect Finally, to each y1n (k − 1), independently and uniformly assign
output feedback an encoder maps messages into length-n code- a random index Ly1 (k − 1) ∈ [1 : 2nRy1 ] = [1 : 2nRv ].
words in the channel model, we first define an operationally Conceptually, the indices F (k) = (Fu (k), Fv (k)) represent the
dual problem to this problem in the source model, as defined public choice of an independent encoder-decoder pair in block
in [21] and which is called Protocol A. Reliability and secrecy k ∈ [1 : b], while the indices W (k) = (Wu (k), Wv (k), Lv (k))
analysis for a proposed random code construction is conducted represent the messages that should be reliably reconstructed at
for Protocol A. We next define a problem in the channel model, the decoder. Only Wv (k) should be directly kept secret from
called Protocol B, that is equivalent to the JCAS problem the eavesdropper. Lv (k) represents a non-secure additional
considered with the addition of a public index. The joint message that should be reliably reconstructed at the decoder
probability distributions in Protocols A and B are shown to be and can be kept secret by applying a one-time pad as used
almost equal under given constraints, which allows to invert in the chosen-secret model [30]–[32]. The role of the index
the random source code proposed for Protocol A to construct Ly1 (k − 1), which is known at all legitimate parties thanks to
a random channel code for Protocol B. The achievability proof the perfect output feedback, is to provide the required key
by using the OSRB method follows by proving that the public for the one-time pad in block k. Secure reconstruction of
index in Protocols A and B can be fixed ahead of transmissions. Lv (k) follows by summing it in modulo-2nRv with Ly1 (k−1).
Protocol A (dual problem in the source model): Similar to Thus, rather than reconstructing Lv (k) directly, the decoder
the dual problem defined in [29] for a wiretap channel, the reconstructs the modulo sum (Lv (k)+Ly1 (k−1)) by estimating
dual of the JCAS problem with partial secrecy and perfect V n (k) since it then can use its observation Y1n (k − 1) from
output feedback is a secret key agreement model, where a the previous transmission block to obtain Lv (k) by applying
source encoder observes X n ∈ X n and independently and modulo-2nRv subtraction. If Ly1 (k−1) is uniformly distributed
uniformly randomly assigns two random bin indices M ∈ and independent of all random variables in the source model
M = M1 × M2 and F ∈ F to it. In the dual source except Y1n (k − 1), then the modulo sum is also uniformly
model, M = (M1 , M2 ) represents a secret key that should distributed and independent of Lv (k), which allows to keep
be reliably reconstructed at a source decoder that observes Lv (k) secret from the eavesdropper. Furthermore, we assign
(Y1n , S1n ) ∈ Y1n × S1n and F to satisfy (3), whereas the for all k = [2 : b] that
eavesdropper observes (Y2n , S2n ) ∈ Y2n × S2n and F , which
M1 (k) = Wu (k), M2 (k) = (Wv (k), Lv (k)). (51)
determines the conditions to satisfy the strong secrecy con-
straint (4). Furthermore, the state sequence estimation at the We next impose conditions on the bin sizes to satisfy all
source encoder by using perfect output feedback should satisfy constraints given in Definition 1.
the distortion constraints (5). Using a Slepian-Wolf [33] decoder, one can reliably recon-
While the strictly causal observation of the i.i.d. state struct U n (k) from (Y1n (k), S1n (k), Fu (k)) for all k = [1 : b]
through feedback does not provide opportunities to improve such that the expected value of the error probability taken over
reliability, feedback offers significant opportunities to improve the random bin assignments vanishes when n → ∞, if we have
secrecy. Hence, we apply a block Markov coding scheme that [21, Lemma 1]
consists of b ≥ 2 transmission blocks, each with n channel eu > H(U |Y1 , S1 ).
R (52)
uses, to transmit (b − 1) independent messages M (k) =
n
(M1 (k), M2 (k)). In every block, secret keys are distilled from Similarly, one can next reliably reconstruct V (k) from
the states and used to protect messages in the subsequent (Y1n (k), S1n (k), Fv (k), U n (k)) for all k = [1 : b] if we have
block. In the following, all n-letter random variables are i.i.d. ev > H(V |Y1 , S1 , U ).
R (53)
according to (9) for all k = [1 : b], obtained by fixing
PU|V , PV |X , and PX so that there exist associated per-letter Thus, (3) is satisfied if (52) and (53) are satisfied and the
cj for j = 1, 2 that satisfy
estimators Estj (x, y1 , y2 ) = S backward decoding is applied, which is a method proposed
in [34] that requires to wait until all b block transmissions
E[dj (Sjn , Estnj (X n , Y1n , Y2n ))] ≤ Dj + ǫn (50) are complete to decode the blocks in the backward order as
k = b, b − 1, . . . , 2 such that reliable reconstruction of Lv (k)
where ǫn > 0 such that ǫn → 0 when n → ∞. The block k is possible by using Y1n (k−1). Furthermore, the large delay of
under consideration is indicated by adding the argument (k) nb channel uses to apply backward decoding can be reduced
to the variables, e.g., M (k) refers to the message in block k, by applying a binning step in the channel model or using the
etc. sliding window decoding [35], as mentioned in [36, pp. 393].
For all blocks k = [1 : b] we construct codes as follows. To The public index Fu (k) and secret key Wu (k) are almost
each un (k) independently and uniformly assign two random independent and uniformly distributed for all k = [1 : b] if we
e
bin indices (Fu (k), Wu (k)) such that Fu (k) ∈ [1 : 2nRu ] and have [21, Theorem 1]
nRu
Wu (k) ∈ [1 : 2 ] for all k = [1 : b]. Furthermore, to
Ru + Reu < H(U ) (54)
each v n (k) independently and uniformly assign three random
e
indices (Fv (k), Wv (k), Lv (k)) such that Fv (k) ∈ [1 : 2nRv ], since then the expected value, which is taken over the random
Wv (k) ∈ [1 : 2nRv ], and Lv (k) ∈ [1 : 2nRv ] for all k = [1 : b]. bin assignments, of the variational distance between the joint
e
probability distributions Unif[1 : 2nRu ] · Unif[1 : 2nRu ] and that 1) the limit of the expectation, defined over the random
PWu Fu vanishes when n → ∞. Furthermore, the public binning operations, of the variational distance between the joint
index Fv (k) and secret key Wv (k) are almost independent probability distributions obtained in Protocol B and required
of (Y2n (k), S2n (k), U n (k)) and uniformly distributed for all for the reliability constraint is 0; 2) the limit of the random
k = [1 : b] if we have probability, defined over the random binning operations, that
ev < H(V |Y2 , S2 , U ). Kullback-Leibler divergence between the joint probability dis-
Rv + R (55)
tributions obtained in Protocol B and required for the secrecy
Similarly, the random bin index Ly1 (k−1) is almost indepen- constraint is greater than 0 is 0. Since the proof steps are
dent of (Y2n (k−1), S2n (k−1), S1n (k−1), V n (k−1), U n (k−1)) standard and mainly repeat the steps in [21], we omit them; see
and uniformly distributed for all k = [2 : b] if we have [29, Section IV] for an extensive proof for a wiretap channel.
Now suppose the public indices F (k) are generated
Ry1 = Rv uniformly at random for all k = [1 : b] indepen-
(a)
< H(Y1 |Y2 , S2 , S1 , V, U ) = H(Y1 |Y2 , S2 , S1 , V ) (56) dently. The encoder generates (U n (k), V n (k)) according
to PU n (k)V n (k)|X n (k)Fu (k)Fv (k) obtained from the binning
where (a) follows because U − V − (Y1 , Y2 , S1 , S2 ) form a scheme above to compute the bins Wu (k) from U n (k) and
Markov chain. Thus, (4) is satisfied by applying the one-time (Wv (k), Lv (k)) from V n (k), respectively, for all k = [1 : b].
padding step above if (55) and (56) are satisfied. Consider next This procedure induces a joint probability distribution that is
the joint condition that (Fu (k), Wu (k), Fv (k), Wv (k), Lv (k)) almost equal to PUV XY1 Y2 S1 S2 fixed above [28, Section 1.6].
are almost mutually independent and uniformly distributed for We remark that the reliability and secrecy metrics considered
all k = [1 : b] if we have above are expectations over all possible realizations F = f .
eu + Rv + R
Ru + R ev + Rv < H(V, U ). (57) Thus, applying the selection lemma [38, Lemma 2.2], these
results prove the inner bound in Proposition 1 by choosing an
Applying the Fourier-Motzkin elimination to (52)-(57), for ǫ > 0 such that ǫ → 0 when n → ∞ and imposing b → ∞.
any ǫ > 0 we can achieve
(a) B. Outer Bound
R1 = Ru = I(U ; Y1 , S1 ) − 2ǫ = I(U ; Y1 |S1 ) − 2ǫ (58)
(b)
Proof Sketch: Assume that for some δn > 0 and n ≥ 1,
R2 = Rv + Rv = min{R2′ , (I(V ; Y1 |S1 ) − R1 )} − 3ǫ (59) there exist an encoder, decoder, and estimators such that (3)-
(5) are satisfied for some tuple (R1 , R2 , D1 , D2 ). Using Fano’s
where (a) follows since U and S1 are independent and (b)
inequality and (3), we have
follows because (U, V ) are mutually independent of (S1 , S2 )
and if H(V |Y2 , S2 , U ) ≤ H(V |Y1 , S1 , U ), then Wv cannot be (a)
H(M |Y1n , S1n ) ≤ H(M |M c) ≤ nǫn (60)
securely reconstructed, i.e., we then have Rv = 0.
We next consider the distortion constraints (5)
where (a) allows randomized decoding and ǫn = δn (R1+R2 )+
on channel-state estimations. Since we assume per-
Hb (δn )/n such that ǫn → 0 if δn → 0.
letter estimators given in (50), (3) is satisfied by
Let Vi , (M1 , M2 , Y1i−1 , S1i−1 ) such that Vi − Xi −
imposing the conditions above on the bin sizes, and all
(Y1,i , Y2,i , S1,i , S2,i ) form a Markov chain for all i ∈ [1 : n]
(un (k), v n (k), xn (k), y1n (k), y2n (k), sn1 (k), sn2 (k)) tuples are
by definition of the channel statistics.
in the jointly typical set with high probability, by applying
the law of total expectation to bounded distortion metrics Bound on R1 : We have
and from the typical average lemma [36, pp. 26], distortion (a)
constraints (5) are satisfied; see also [37]. Furthermore, nR1 ≤ I(M1 ; Y1n |S1n ) + nǫn
X n
without loss of generality one can use the deterministic 
per-letter estimators that result from (11) and the proof ≤ H(Y1,i |S1,i ) − H(Y1,i |M1 , M2 , Y1i−1 , S1n ) + ǫn
i=1
follows from the proof of [7, Lemma 2] by replacing n
b
(S, Z, Ŝ, d) with (Sj , (Y1 , Y2 ), Sj , dj ), respectively, since (b) X 
= H(Y1,i |S1,i ) − H(Y1,i |M1 , M2 , Y1i−1 , S1i ) + ǫn
c
Sj (k) − (X(k), Y1 (k), Y2 (k)) − Sj (k) form a Markov chain i=1
for all j = 1, 2 and k = [1 : b]. n
(c) X
Protocol B (main problem in the channel model with = I(Vi ; Y1,i |S1,i ) + ǫn ) (61)
common randomness assistance): In Protocol B, we consider i=1
the JCAS problem with partial secrecy and perfect output
where (a) follows by (60) and because M1 and S1n are
feedback, and assist the problem with the public index F (k)
independent, (b) follows since
for all k = [1 : b] such that (3)-(5) are satisfied also for
Protocol B by choosing R1 and R2 as in (58) and (59), n
S1,i+1 − (M1 , M2 , Y1i−1 , S1i ) − Y1,i (62)
respectively. The proof of this result follows mainly by proving
that the joint probability distribution obtained in Protocol A is form a Markov chain, and (c) follows from the definition of
almost preserved in Protocol B, i.e., we prove for n → ∞ Vi .
Bound on (R1 + R2 ): Similar to (61), we obtain ACKNOWLEDGMENT
(a) This work has been supported by the German Federal
n(R1 + R2 ) ≤ I(M1 , M2 ; Y1n |S1n ) + nǫn Ministry of Education and Research (BMBF) under the Grant
n
(b) X 16KIS1242, German Research Foundation (DFG) under the

≤ H(Y1,i |S1,i ) − H(Y1,i |M1 , M2 , Y1i−1 , S1i ) + ǫn Grant SCHA 1944/9-1, and National Science Foundation
i=1 (NSF) under the Grant CCF 1955401.
n
(c) X
= I(Vi ; Y1,i |S1,i ) + ǫn ) (63) R EFERENCES
i=1
[1] T. Wild, V. Braun, and H. Viswanathan, “Joint design of communication
where (a) follows because (M1 , M2 , S1n ) are mutually inde- and sensing for beyond 5G and 6G systems,” IEEE Access, vol. 9, pp.
pendent and by (60), (b) follows since (62) form a Markov 30 845–30 857, Feb. 2021.
[2] Z. Wei, F. Liu, C. Masouros, N. Su, and A. P. Petropulu, “Towards multi-
chain, and (c) follows from the definition of Vi . functional 6G wireless networks: Integrating sensing, communication
Bound on R2 : We obtain and security,” July 2021, [Online]. Available: arxiv.org/abs/2107.07735.
[3] W. Zhang, S. Vedantam, and U. Mitra, “Joint transmission and state
(a) estimation: A constrained channel coding approach,” IEEE Trans. Inf.
nR2 ≤ I(M2 ; Y1n , Y2n , S1n , S2n ) + nǫn Theory, vol. 57, no. 10, pp. 7084–7095, Oct. 2011.
[4] H. W. et al., “Integration of communication and sensing in 6G: A joint
≤ H(Y1n , S1n |Y2n , S2n ) + H(Y2n , S2n ) industrial and academic perspective,” in IEEE Annu. Int. Symp. Pers.,
− H(Y2n , S2n |M2 ) − H(Y1n , S1n |Y2n , S2n , M1 , M2 )+nǫn Indoor Mobile Radio Commun., Helsinki, Finland, Sep. 2021, pp. 1–7.
[5] S. Buzzi, C. D’Andrea, and M. Lops, “Using Massive MIMO arrays
= H(Y1n , S1n |Y2n , S2n ) + I(Y2n , S2n ; M2 ) for joint communication and sensing,” in Asilomar Conf. Signals, Syst.,
Comput., Pacific Grove, CA, Nov. 2019, pp. 5–9.
− H(Y1n , S1n |Y2n , S2n , M1 , M2 )+nǫn [6] G. Fettweis et al., “Joint communications & sensing - Common radio-
n
(b) X communications and sensor technology,” VDE Positionspapier, July
≤ H(Y1,i , S1,i |Y2,i , S2,i ) + δn + nǫn 2021.
[7] M. Kobayashi, G. Caire, and G. Kramer, “Joint state sensing and
i=1
n
communication: Optimal tradeoff for a memoryless case,” in IEEE Int.
X Symp. Inf. Theory, Vail, CO, June 2018, pp. 111–115.
− H(S1,i |Y2,i , S2,i , M1 , M2 , Y1i−1 , S1i−1 ) [8] M. Kobayashi, H. Hamad, G. Kramer, and G. Caire, “Joint state sensing
i=1 and communication over memoryless multiple access channels,” in IEEE
(c) n
X Int. Symp. Inf. Theory, Paris, France, July 2019, pp. 270–274.
≤ H(Y1,i , S1,i |Y2,i , S2,i ) + δn + nǫn [9] M. Ahmadipour, M. Wigger, and M. Kobayashi, “Joint sensing and
communication over memoryless broadcast channels,” in IEEE Inf.
i=1 Theory Workshop, Riva del Garda, Italy, Apr. 2021, pp. 1–5.
n
X [10] N. Su, F. Liu, and C. Masouros, “Secure radar-communication systems
− H(S1,i |Y2,i , S2,i , Vi ) (64) with malicious targets: Integrating radar, communications and jamming
i=1 functionalities,” IEEE Trans. Wireless Commun., vol. 20, no. 1, pp. 83–
95, Jan. 2021.
where (a) follows by (60), (b) follows by (4) and from [11] J. D. D. Mutangana, R. Tandon, Z. Goldfeld, and S. Shamai, “Wiretap
channel with latent variable secrecy,” in IEEE Int. Symp. Inf. Theory,
Remark 1, and because Melbourne, Australia, July 2021, pp. 837–842.
[12] R. Ahlswede and N. Cai, “Transmission, identification and common
randomness capacities for wire-tape channels with secure feedback from
n\i n\i the decoder,” Electron. Notes Discrete Math., vol. 21, pp. 155–159, Aug.
(Y2 , S2 ) − (Y2,i , S2,i , M1 , M2 , Y1i−1 , S1i−1 ) − S1,i (65) 2005.
[13] A. Cohen and A. Cohen, “Wiretap channel with causal state information
form a Markov chain, and (c) follows from the definition of and secure rate-limited feedback,” IEEE Trans. Commun., vol. 64, no. 3,
Vi . pp. 1192–1203, Mar. 2016.
[14] M. Bloch, O. Günlü, A. Yener, F. Oggier, H. V. Poor, L. Sankar, and R. F.
Distortion Bounds: We have for j = 1, 2 Schaefer, “An overview of information-theoretic security and privacy:
Metrics, limits and applications,” IEEE J. Sel. Areas Inf. Theory, vol. 2,
(a)    
cn ) = 1 E dj (Sj,i , S
no. 1, pp. 5–22, Mar. 2021.
(Dj +δn ) ≥ E dj (Sjn , S d j,i ) (66) [15] B. Dai, A. J. H. Vinck, Y. Luo, and Z. Zhuang, “Capacity region of non-
j
n degraded wiretap channel with noiseless feedback,” in IEEE Int. Symp.
Inf. Theory, Cambridge, MA, July 2012, pp. 244–248.
where (a) follows by (5), which is achieved by using the [16] X. He and A. Yener, “The role of feedback in two-way secure communi-
deterministic per-letter estimators that result from (11). cations,” IEEE Trans. Inf. Theory, vol. 59, no. 12, pp. 8115–8130, Dec.
Introduce a uniformly distributed time-sharing random vari- 2013.
[17] G. Bassi, P. Piantanida, and S. Shamai, “The wiretap channel with
able Q ∼ Unif[1 : n] that is independent of other random generalized feedback: Secure communication and key generation,” IEEE
variables, and define Y1 = Y1,Q , S1 = S1,Q , Y2 = Y2,Q , Trans. Inf. Theory, vol. 65, no. 4, pp. 2213–2233, Apr. 2019.
S2 = S2,Q , and V = (VQ ,Q), so V − X − (Y1 , Y2 , S1 , S2 ) [18] E. Ardestanizadeh, M. Franceschetti, T. Javidi, and Y.-H. Kim, “Wiretap
channel with secure rate-limited feedback,” IEEE Trans. Inf. Theory,
form a Markov chain. The proof of the outer bound follows vol. 55, no. 12, pp. 5353–5361, Dec. 2009.
by letting δn → 0. [19] A. Gohari and G. Kramer, “An upper bound for wiretap multi-way
Cardinality Bounds: We use the support lemma [39, channels,” Sep. 2020, [Online]. Available: arxiv.org/abs/2009.14814.
[20] R. Ahlswede and I. Csiszár, “Common randomness in information theory
Lemma 15.4] to prove the cardinality bound, which is a and cryptography - Part I: Secret sharing,” IEEE Trans. Inf. Theory,
standard procedure, so we omit the proof. vol. 39, no. 4, pp. 1121–1132, July 1993.
[21] M. H. Yassaee, M. R. Aref, and A. Gohari, “Achievability proof via [30] O. Günlü, O. İşcan, V. Sidorenko, and G. Kramer, “Reliable secret-key
output statistics of random binning,” IEEE Trans. Inf. Theory, vol. 60, binding for physical unclonable functions with transform coding,” in
no. 11, pp. 6760–6786, Nov. 2014. IEEE Global Conf. Sign. Inf. Process., Washington, DC, Dec. 2016, pp.
[22] J. M. Renes and R. Renner, “Noisy channel coding via privacy amplifi- 986–991.
cation and information reconciliation,” IEEE Trans. Inf. Theory, vol. 57, [31] T. Ignatenko and F. M. J. Willems, “Biometric systems: Privacy and
no. 11, pp. 7377–7385, Nov. 2011. secrecy aspects,” IEEE Trans. Inf. Forensics Security, vol. 4, no. 4, pp.
[23] M.-C. Chang, T. Erdogan, S.-Y. Wang, and M. Bloch, “Rate and detection 956–973, Dec. 2009.
error-exponent tradeoffs of joint communication and sensing,” in IEEE [32] O. Günlü, “Key agreement with physical unclonable functions and
Int. Hybrid Symp. Joint Commun. & Sensing, Seefeld, Austria, Mar. biometric identifiers,” Ph.D. dissertation, TU Munich, Germany, Nov.
2022, to appear. 2018, published by Dr.-Hut Verlag in Feb. 2019.
[24] A. Schrijver, Theory of Linear and Integer Programming. Chichester, [33] D. Slepian and J. Wolf, “Noiseless coding of correlated information
England: John Wiley & Sons, June 1998. sources,” IEEE Trans. Inf. Theory, vol. 19, no. 4, pp. 471–480, July
[25] O. Günlü, O. İşcan, V. Sidorenko, and G. Kramer, “Code constructions 1973.
for physical unclonable functions and biometric secrecy systems,” IEEE [34] F. Willems and E. van der Meulen, “The discrete memoryless multiple-
Trans. Inf. Forensics Security, vol. 14, no. 11, pp. 2848–2858, Nov. 2019. access channel with cribbing encoders,” IEEE Trans. Inf. Theory, vol. 31,
[26] R. A. Chou and M. R. Bloch, “Separation of reliability and secrecy no. 3, pp. 313–327, May 1985.
in rate-limited secret-key generation,” IEEE Trans. Inf. Theory, vol. 60, [35] A. Carleial, “Multiple-access channels with different generalized feed-
no. 8, pp. 4941–4957, Aug. 2014. back signals,” IEEE Trans. Inf. Theory, vol. 28, no. 6, pp. 841–850, Nov.
[27] O. Günlü, P. Trifonov, M. Kim, R. F. Schaefer, and V. Sidorenko, 1982.
“Randomized nested polar subcode constructions for privacy, secrecy, [36] A. E. Gamal and Y.-H. Kim, Network Information Theory. Cambridge,
and storage,” in IEEE Int. Symp. Inf. Theory Appl., Kapolei, HI, Oct. U.K.: Cambridge University Press, 2011.
2020, pp. 475–479. [37] O. Günlü, R. F. Schaefer, and H. V. Poor, “Biometric and physical
[28] M. Bloch, Lecture Notes in Information-Theoretic Security. Atlanta, identifiers with correlated noise for controllable private authentication,”
GA: Georgia Inst. Technol., July 2018. July 2020, [Online]. Available: arxiv.org/abs/2001.00847.
[29] M. Nafea and A. Yener, “A new wiretap channel model and its strong [38] M. Bloch and J. Barros, Physical-layer Security. Cambridge, U.K.:
secrecy capacity,” IEEE Trans. Inf. Theory, vol. 64, no. 3, pp. 2077–2092, Cambridge University Press, 2011.
Mar. 2018. [39] I. Csiszár and J. Körner, Information Theory: Coding Theorems for
Discrete Memoryless Systems, 2nd ed. Cambridge, U.K.: Cambridge
University Press, 2011.

You might also like