1378 IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 15, NO.

6, NOVEMBER 2021

Joint Transmit Waveform and Receive Filter Design

for Dual-Function Radar-Communication Systems
Christos G. Tsinos , Senior Member, IEEE, Aakash Arora , Student Member, IEEE,
Symeon Chatzinotas , Senior Member, IEEE, and Björn Ottersten , Fellow, IEEE

Abstract—In this paper, the problem of joint transmit waveform interconnected devices [1] are expected, spanning from regular
and receive filter design for dual-function radar-communication cellular networks to small-scale smart devices under the concept
(DFRC) systems is studied. The considered system model involves of the so-called “Internet of Things (IoT)” [2], [3].
a multiple antenna base station (BS) of a cellular system serving
multiple single antenna users on the downlink. Furthermore, the Thus, additional frequency spectrum resources are more than
BS simultaneously introduces sensing capabilities in the form of necessary to wireless communication systems in order to meet
point-like target detection from the reflected return signals in a those demands. To that end, a promising solution is the efficient
signal-dependent interference environment. A novel framework utilization of the electromagnetic spectrum, currently occupied
based on constrained optimization problems is proposed for the by other applications. Among the different propositions [4]- [6],
joint design of the transmit waveform and the radar receive filter
such that different constraints related to the power amplifiers the case of joint communications-radar spectrum sharing has
and the radar waveform are satisfied. In contrast to the existing recently concentrated great interest in the literature [7]–[9]. The
approaches in the DFRC systems’ literature, the proposed ap- latter is usually implemented via two different approaches: 1)
proach does not require the knowledge of a predetermined radar Coexistence of the radar and the communication systems and 2)
beampattern in order to optimize the performance of the radar part Dual-function radar-communication (DFRC) system design.
through its approximation. Instead, a beampattern is generated by
maximizing the radar receive signal-to-interference ratio (SINR) In the first one, the design of the transmit signals from the
thus, enabling a more flexible design. Moreover, the radar receive two systems is done independently while managing the cross
filter processing and its optimization is considered for the first interference to each other in order to avoid a possible degradation
time on DFRC systems, enabling the effective exploitation of the on their performance [10]–[15]. Therefore, cooperation between
available degrees of freedom in the radar receive array. Efficient the two systems is required in real-time that perplexes the system
algorithmic solutions with guaranteed convergence are developed
for the defined constrained nonconvex optimization problems. The design and results in increased complexity and communication
effectiveness of the proposed solutions is verified via numerical overhead.
results. The second one, aims at the design of systems that can
Index Terms—Radar-communication, beamforming, precoding,
jointly handle the operations of the radar and the communi-
waveform design, MIMO, spectrum sharing, multiuser cation systems. Such a design may apply in real-time the joint
interference, nonconvex optimization, alternating optimization, sensing/communication operations via a single hardware setup.
gradient-projection (GP). To that end, the DFRC system is implemented via designing
the transmit waveform such that both radar/communication-
I. INTRODUCTION related performance metrics are optimized and spatial/temporal
constraints are satisfied, respectively. This approach appropri-
HE rise of the 5th Generation (5G) and beyond wire-
T less technologies brings huge demands for high quality
wireless communications services, since numerous different
ately exploits the available spatial degrees of freedom in the
DFRC system, i.e., multiple transmit antennas. For this multiple
input-multiple output (MIMO) system, the designed transmit
devices are expected to be interconnected. By 2022, 28.5 billion waveform enables the transmission at high information rates to
the intended users while maintaining a reliable operation for the
Manuscript received February 23, 2021; revised June 6, 2021; accepted radar system.
August 27, 2021. Date of publication September 14, 2021; date of current version
December 2, 2021. This work was supported by FNR, Luxembourg under the In [16], the transmit waveform is directly designed by min-
FNR projects DISBuS, ECLECTIC, RISOTTI, and SPASAT. The guest editor imizing the multiuser interference (MUI) such that a desired
coordinating the review of this manuscript and approving it for publication was radar beampattern is achieved by imposing constraints on the
Prof. Zhiyong Feng. (Corresponding author: Christos Tsinos.)
Christos G. Tsinos is with the Interdisciplinary Centre for Security, Reliability covariance matrix of the transmit signals. In [17], linear precod-
and Trust (SnT), University of Luxembourg, Luxembourg, and also with the ing solutions are derived such that the designed signals to be
General Department, National and Kapodistrian University of Athens, 15772 transmitted match the desired radar beampattern, guarantee the
Athens, Greece (e-mail: [email protected]).
Symeon Chatzinotas and Bjorn Ottersten are with the Interdisciplinary required signal-to-interference ratio (SINR) level at the intended
Centre for Security, Reliability and Trust (SnT), University of Luxem- users and satisfy a transmission power constraint. A similar ap-
bourg, Luxembourg (e-mail: [email protected]; [email protected]; proach with improved performance was followed in [18] where
[email protected]).
Digital Object Identifier 10.1109/JSTSP.2021.3112295 the precoding matrix is decomposed into two parts, one for the
communication and the other for the radar system. A different

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
TSINOS et al.: JOINT TRANSMIT WAVEFORM AND RECEIVE FILTER DESIGN FOR DFRC SYSTEMS 1379

view is presented in [19]. There, a vehicular DFRC system is (GP) frameworks that have fast convergence to a stationary
proposed that transmits a single waveform and extracts both point. Furthermore, we establish the convergence of the
radar and communication parameters at the intended receiver proposed solutions to a Karush-Kuhn-Tucker (KKT)-point
through efficient processing. of the respective problems.
In the existing literature on the DFRC systems [16]–[18], r Numerical results verify the excellent performance of the
the waveform designs are derived such that only transmission proposed solutions under different simulation setups. We
energy/power related constraints are mainly satisfied. On the also compare the performance of the proposed algorithms
other hand, the radar waveform usually has to satisfy additional to known benchmarks that independently optimize the
spatial/temporal constraints, e.g., constant-modulus [20], [21], transmit signals of the communication and the radar sys-
similarity to a known waveform [22], [23], etc. Furthermore, the tems.
existing works on the DFRC systems aim to optimize the perfor- The rest of the paper is organized as follows. Section II
mance of the radar part by minimizing the distance to/imposing a describes the considered system model for the DFRC system.
predetermined beampattern in the designed radar waveform. On In Section III, the optimization problems for the joint transmit
the contrary, in the MIMO radar-only literature, there are several waveform and radar receive filter design are formulated for the
works that optimize the radar performance without the need of a different cases of the considered spatial/temporal constraints.
predetermined beampattern [24]–[29]. In those cases, the radar Section IV derives the algorithmic solutions to the previously
transmission waveform is designed by maximizing the radar defined optimization problems. Section V presents the numerical
receive signal-to-interference ratio (SINR). Since the detection results and Section VI concludes this work.
probability of a target is an monotonically increasing function Notation. Upper-case and lower-case bold letters are used to
of the radar output SINR [26], the radar performance can be op- define matrix and vector variables, respectively. The N × N
timized without the knowledge of a predetermined beampattern, identity matrix is denoted by IN . The N × 1 vector of zeros is
leading to a more flexible design. Moreover, the existing works denoted by 0N . C and R are the complex and the real numbers’
on the DFRC systems do not consider the application/design of domains, respectively. | · | denotes the modulus of a complex
a radar receive filter in their system model. In literature so far, number. ∇x denotes the gradient with respect to the variable
the application/design of radar receive filters has been solely x. E{·} is the expectation operator.  · 2 ,  · ∞ , and  · F
considered for radar-only systems [24]–[29]. To that end, the denote the l2 , l∞ and Frobenius norms, respectively. (·)H and
aim of the present work is to propose a novel framework for the (·)T denote the Hermitian and the transpose of a vector/matrix,
joint design of the transmit waveform and the radar receive filter respectively. vec(·) is the vectorization operator. A ⊗ B is the
of a DFRC system based on the MUI minimization/radar receive Kronecker product of matrices A and B. Re{·} is the real part
SINR maximization criteria. The proposed framework is able of the complex operand.
to design transmit waveforms that simultaneously satisfy both
communications and radar specific spatial/temporal constraints. II. SYSTEM MODEL
The contributions of the present work are summarized as
Let us consider a narrow-band DFRC system equipped with an
follows.
r A novel design for a DFRC system with radar receive uniform linear array (ULA) of T elements with half-wavelength
inter-element spacing. The DFRC system serves M single
filtering is proposed.
r The transmit waveform and the radar receive filter of the antenna user terminals (UTs) on the downlink operation and
simultaneously transmits radar probing waveforms for point-like
DFRC system are jointly designed such that both the MUI
target detection. Let us further assume that the DFRC system is
in the communications system and the receive SINR at
also equipped with a dedicated ULA of R elements for receiving
the radar’s side are optimized. The SINR maximization
the radar signals, following the model of a radar system with
criterion for optimizing the performance of the radar part
collocated transmit and receive antenna arrays [24]–[29]. The
is considered for the first time in the literature of DFRC
system model is shown in Fig. 1. In Subsections II-A and
systems.
r The transmit waveform is designed while satisfying differ- II-B, the operation of the communication and the radar part
are described in detail. Then, in Subsection II-C, the operation
ent spatial/temporal constraints for the first time in the liter-
protocol of the DFRC system is presented.
ature. The cases of 1) total transmission energy, 2) constant-
modulus, 3) total transmission energy-plus-similarity and
A. Communication Model
4) constant-modulus-plus-similarity constraints are con-
sidered. By dropping the time index for simplicity, the received signals
r In order to jointly design the transmit waveform and the at the communication UTs in N symbol times are given by,
radar receive filter subject to the aforementioned spa-
Y = HX + Z, (1)
tial/temporal constraints, four different nonconvex con-
M ×N
strained optimization problems are formulated. This is the where Y ∈ C is a matrix whose entry ym,n is the signal
first time that these problems are studied in the open liter- received at the mth UT on the nth symbol time, 1 ≤ m ≤ M ,
ature and as a consequence, they have no known solutions. 1 ≤ n ≤ N , X = [x1 , . . . , xN ] ∈ X is the T × N transmit sig-
To that end, we develop novel algorithmic solutions based nal matrix for the N symbol times, X is a set of spatial/temporal
on the alternating optimization and gradient-projection constraints on the transmitted signals which are discussed in
1380 IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 15, NO. 6, NOVEMBER 2021

Fig. 1. DFRC system model.

Section III, xn denotes the transmit vector from the DFRC sys- where sm,n is the (m, n)th entry of the symbol matrix S. The
tem’s antennas at symbol time n, H = [h1 , . . . , hM ]T ∈ C M ×T expectation operator in the numerator is applied on sm,n and in
is the matrix of the frequency-flat fading channels between the denominator is applied on sm,n and xn , as well.
the DFRC system and the UTs, hm is the T × 1 vector of It has been shown in [30] that the achievable information rate
channel coefficients between the DFRC transmitter and the mth for the mth user is a function of ξm in (4). Thus, the achievable
UT, modeled as CN (0, IT ), Z = [z1 , . . . , zN ] is the Additive sum-rate of the communication part can be expressed as,
White Gaussian Noise (AWGN) matrix variable that corrupts M

the corresponding transmissions, zn ∼ CN (0, σz2 IM ) and σz2 r= log2 (1 + ξm ). (5)
denotes the noise variance. m=1
The communication part of the system aims at the transmis-
Let us assume that 1) the symbols sm,n , 1 ≤ m ≤ M , 1 ≤
sion of a desired symbol s(m, n) ∈ O, 1 ≤ m ≤ M , 1 ≤ n ≤
n ≤ N are drawn from the same constellation set O and have
N , from the DFRC system to the mth UT in the nth symbol
fixed energy and 2) the channel matrix H is perfectly estimated at
time. O is the set of the employed constellation points. In each
the DFRC system, e.g. via a training-based method [31]. Then,
symbol time, the vector of the symbols to be transmitted to
the signal power E{|sm,n |2 } in (4) is also fixed and thus, the
the UTs is denoted by sn ∈ OM ×1 . Given the CSI and sn , the
SINR expression per user can be maximized by minimizing the
communication part of the DFRC system aims at the design of
MUI energy in (2). From (5), it is straightforward to see that
a transmit signal vector, xn , such that the received vector signal
by maximizing the SINR of the UTs, their achievable rate is
yn is as close to sn , as possible.
maximized, as well. That is, the MUI energy (2) minimization
The latter may be achieved by minimizing the so-called “MUI
is a suitable criterion for optimizing the performance of the
energy” [16], [30], given by,
communication system.
f (X) = HX − S2F , (2) Note that the signals to be transmitted are designed in a
block basis, i.e., for N symbol times (matrix X). Furthermore,
where S = [s1 , . . . , sN ] is the matrix of the stacked symbols observe that the optimal transmit signals that convey the desired
over the N symbol times. Observe that the received signal of symbols to the intended UTs are directly designed through
the mth UT in the nth symbol time can be written as, the minimization of the least-squares function in (2). That is,
ym,n = sm,n + hTm xn − sm,n +zm,n . (3) the transmit signal matrix X is derived in a nonlinear manner
   without the need of a linear precoder.
MUI at the mth user

Then, according to (3), the receive SINR per block of N symbols B. Radar Model
for the mth user is defined as
We may now move to the description of the radar model. For
E{|sm,n |2 } the transmitted T × 1 signal xn , 1 ≤ n ≤ N by the antennas of
ξm = , (4)
E{|hTm xn − sm,n |2 } + N0 the DFRC system, the signal seen at a location with angle θ is
TSINOS et al.: JOINT TRANSMIT WAVEFORM AND RECEIVE FILTER DESIGN FOR DFRC SYSTEMS 1381

given by, function of the output SINR [23], [26]. It is assumed that a linear
RN × 1 finite impulse response (FIR) filter w is applied at the
aTt (θ)xn , n = 1, . . . , N, (6)
radar receive array for maximizing the output SINR. The filter
where at (θ) denotes the T × 1 transmit steering vector. As output is given by,
discussed in the beginning of the present section, here, we
assume that a ULA with half-wavelength inter-element spacing
is employed and thus, the steering vector is given by, c = wH q
1  T K

at (θ) = √ 1, e−jπ sin(θ) , . . . , e−jπ(T −1) sin(θ) . (7) = α0 wH A(θ0 )x+wH αk A(θk )x+wH u. (12)
T
k=1
Let us assume that the target is located at angle θ0 and that
there are also K signal-dependent interference sources located at
angles θk = θ0 , k = 1, . . . , K. We further assume that any pos- Then, the output SINR can be expressed as,
sible self-interference between the transmit and the radar receive
array is mitigated by employing an appropriate method [32]. The
baseband vector of signals at the radar receive array in the nth σ02 |wH A(θ0 )x|2
γ(x, w) =   .
symbol time is given by, K
wH k=1 σk2 A(θk )xxH AH (θk ) w+ σu2 wH w
qn = α0 ar (θ0 )aTt (θ0 )xn (13)
K
Under the assumption that the DFRC system is able to estimate
 θk and σk2 for k = 0, 1, . . . K, the optimal waveform x and filter
+ αk ar (θk )aTt (θk )xn + un , (8)
w can be derived via the SINR maximization criterion. This
k=1
way, we optimize the performance at the radar’s side in terms
where α0 and αk are the complex amplitudes of the target and the of its target detection probability, as well. This is achieved by
kth interference source with E{|α0 |2 } = σ02 , and E{|αk |2 } = designing a transmit waveform and radar receive filter such that
σk2 , respectively, un is a circular symmetric complex white deep nulls are placed in specific directions to reject the signals
Gaussian noise vector with zero mean and covariance matrix coming from the interfering sources while keeping distortionless
equal to σu2 IR , σu2 is the noise variance and ar (θ) is the R × 1 response in the target signal’s direction. That is, a beampattern
vector of the propagation delays from a source located at angle is automatically generated that aims at the maximization of
θ to the radar receive elements. Again, under the assumption the radar receive SINR. Note that the required information
of a ULA with half-wavelength inter-spacing between adjacent regarding the interfering sources may be obtained by using an
elements, we have that environmental dynamic database, following a cognitive radar
1  −jπ sin(θ) T
paradigm, as in [33]–[35].
ar (θ) = √ 1, e , . . . , e−jπ(R−1) sin(θ) . (9)
R
Note that in the previous, the angle of arrival in the radar
receive array is assumed to be the same with the angle of C. Operation Protocol of the DFRC System
departure of the transmit signal from the DFRC system. This is This section is closed with the description of the operation
a reasonable assumption within the relevant literature [24]–[29] protocol of the DFRC system. The operation flow-chart is pre-
given that the transmit and the radar receive array are collocated. sented in Fig. 2. As we can see, the DFRC system’s operation
Of course in practice, these two angles might differ, due to is divided into four stages. In the first one, the DFRC system
multi-path phenomena, for example. Nevertheless, if both the estimates the required information related to the communication
actual angles of arrival and departure can be estimated, the system’s channel matrix and the angles and the complex gains of
proposed methodology is directly applicable for the design of the target and interference sources, as well. In the second stage,
the DFRC system. based on the estimated information and the desired symbols
By setting x = vec(X), q = vec(Q) and u = vec(U), where to be transmitted to the UTs in the next N symbol times, the
Q = [q1 , . . . , qN ] and U = [u1 , . . . , uN ], (8) can be rewritten DFRC system jointly designs the corresponding transmit signal
as matrix X and the radar receive filter w. In the third one, the
K
 derived signals in X are transmitted for N symbol times. Based
q = α0 A(θ0 )x + αk A(θk )x + u, (10) on the received signals, the UTs are able to directly detect the
k=1 transmitted symbols. During this period, the DFRC radar receive
where A(θ) is defined for angle θ as array, receives also the reflected signals from the target and the
interference sources. In the final stage, the transmission of the
A(θ) = IN ⊗ ar (θ)aTt (θ) . (11)
signals in matrix X is completed. The filter w is applied on the
A common approach to design the radar waveform is based on radar receive signals and the detection of the target source is
the maximization of the output SINR. This is the case since examined such that a predetermined false alarm probability is
the detection probability is usually an monotonically increasing preserved.
1382 IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 15, NO. 6, NOVEMBER 2021

Fig. 2. Operation flow-chart of the DFRC system.

III. PROBLEM FORMULATION of f (x). Thus, the DFRC system favors the performance of
In this section, we formulate four optimization problems the communications part over the radar one. Contrariwise, as ρ
via which the optimal transmit waveform and the radar re- moves towards 0, a higher weight is allocated to the minimization
ceive filter are jointly designed. These optimization prob- of 1/γ(x, w) or equivalently to the the maximization of the
lems differ in the spatial/temporal constraints that are im- SINR expression γ(x, w). Therefore, the DFRC system opts for
posed on the transmit waveform. The cases of total transmis- a better radar performance over the one of the communication
sion energy (TE), constant-modulus (CM), total transmission part. Thus, a flexible performance trade-off between the radar
energy-plus-similarity (TE+SIM) and constant-modulus-plus- and the communication objectives is achieved.
similarity (CM+SIM) constraints are considered. Before pro- We may now proceed to the different problem formulations
ceeding further, let us first focus on how the function of both considering the four different cases of spatial/temporal con-
the communication and the radar part can simultaneously be straints, discussed above.
optimized.
At this point, it is instructive to highlight that the DFRC system A. Total Transmission Energy Constraint
optimization requires the knowledge of the channel matrix H
and the radar parameters θk and σk2 for k = 0, 1, . . . , K. Note The TE or total transmission power constraint has been ex-
that the estimation methods that are employed by the indepen- tensively employed in the communications’ literature to model
dent communication and radar systems can be employed by the the limitations of the power amplifiers (PAs) at the transmitter’s
DFRC system, as well. That is, the channel matrix can be esti- side [37]. First works on the MIMO radar domain have also
mated via a training based scheme [31] and the radar parameters considered the SINR maximization problem under a constraint
can be estimated as discussed in the end of Section II-B. on the total transmission energy [24], [38], [39]. In case of the
As discussed in Sections II-A and II-B, the communication DFRC system, the optimization problem to be solved under the
part of the DFRC system requires the minimization of the TE constraint is defined as,
objective function in (2), in order to design the transmit signals in
X such that the desired symbols in S are conveyed to the intended 1
(P1 ) : min ρf (x) + (1 − ρ)
UTs. Furthermore, the radar part of the DFRC system may x,w γ(x, w)
maximize the detection probability by maximizing the SINR s.t. x22 ≤ Pmax ,
expression in (13). Clearly, the simultaneous optimization of
both of the communication and radar parts is not a straightfor-
where Pmax is the maximum transmission energy, supported by
ward task. Among the different approaches for solving such
the DFRC system.
a multi-objective optimization problem [36], we opt for the
weighted sum method that combines the two objective functions
into a composite one. Thus, the composite objective function is B. Constant-Modulus Constraint
given by
The CM constraint is highly desirable in real-world commu-
1 nication as well as radar systems due to its low Peak-to-Average-
fc (x, w) = ρf (x) + (1 − ρ) , (14)
γ(x, w) Ratio (PAR) property. The possible large envelope fluctuations
of the transmission signals could lead to significant nonlinear
where f (x) = H̃x − s22 , H̃ = IN ⊗ H and ρ ∈ R is a pa- distortions by the transmitter’s PAs. In order to avoid the undesir-
rameter that enables the trade-off between the communication able distortion, each PA has to work in its linear region, resulting
and the radar performance. in low power efficiency, [40]. On the contrary, CM transmission
It is evident from (14) that by selecting the ρ value to be signals are not affected by the distortions produced by amplitude
close to 1, a higher weight is imposed on the minimization nonlinearities when appropriate filtering is used, [41]. In this
TSINOS et al.: JOINT TRANSMIT WAVEFORM AND RECEIVE FILTER DESIGN FOR DFRC SYSTEMS 1383

case, the optimization problem to be solved is defined, as and

1 1
(P2 ) : min ρf (x) + (1 − ρ) (P4 ) : min ρf (x) + (1 − ρ) + (1 − ρ)λx − x0 22
x,w γ(x, w) x,w γ(x, w)

Pmax Pmax
s.t. |xn | = , 1 ≤ n ≤ T N, s.t. |xn | = , 1 ≤ n ≤ T N,
TN TN
respectively. λ ∈ R is a penalty parameter. That is, by appropri-
where xn is the nth entry of x and the modulus |xn | is set
ately tuning the ρ and λ values, we can achieve the desired oper-
to Pmax /(T N ) in order to have the same total transmission
ation point regarding to the performance of the communication-
energy per block of N symbols with the case of (P1 ).
radar parts and the similarity to a reference waveform, as well.

C. Similarity Constraint IV. ALGORITHMIC FRAMEWORK FOR THE JOINT TRANSMIT

While the maximization of the SINR expression in (13) WAVEFORM-RADAR RECEIVE FILTER DESIGN
achieves the maximization of the detection probability, as well, In this section, we develop algorithmic solutions to problems
it does not necessarily results in a transmit waveform that has (P1 ) − (P4 ). The aforementioned problems are nonconvex due
good ambiguity properties, i.e., a narrow main peak both in to the involved nonconvex cost function and the nonconvex
range and in Doppler and relatively low sidelobe peaks. Transmit CM constraint in (P2 ) and (P4 ). Before proceeding towards
waveforms with good ambiguity properties are desired in the the algorithmic solutions development, first we observe that
radar systems since they enable the reliable estimation of the all the problems involve a differentiable objective function.
so-called “Doppler frequency”. The latter is very important since Moreover, later in the present section, we show that the orthog-
it is directly related to the target radial velocity which is required onal projection on the constraint sets of problems (P1 ) − (P4 ),
for tracking the target and classifying its dangerousness [23], can be computed in closed-form. These two facts lead to the
[42]. development of algorithmic solutions based on the alternating
A common approach followed in the literature is to consider minimization and GP frameworks.
a similarity constraint on the SINR maximization problem. The
similarity constraint forces the designed waveform to be close
A. Algorithms
to a reference one in some sense. As a reference waveform
is used a known one that possesses the desirable properties. Problems (P1 ) and (P2 ) share the same objective function
Nevertheless, the addition of the similarity constraint comes which is differentiable in both w and x. We refer to this common
at the cost of inferior receive SINR and as a consequence, objective function with the term g(x, w). Also, problems (P3 )
lower target detection probability. As it is evident, the closer a and (P4 ) have the same objective function which has a penalty
waveform is to the reference one, the better is the approximation term to enforce similarity constraints in addition to the objective
of its desirable properties and the higher is the degradation on function of problems (P1 ) − (P2 ). The term g (x, w) is used to
the detection performance. denote that function. Furthermore, (P1 ) and (P3 ) have to satisfy
In radar-only studies, the similarity constraint is usually en- the TE constraint which can be written in set form as, XT E =
forced via a constraint on the l2 or the l∞ norm of the difference {x ∈ C T N ×1 |x22 ≤ Pmax }. In a similar manner, (P2 ) and
of the designed waveform to the reference one, i.e., (P4 ) have to satisfy the CM constraint which is expressed in
set form as XCM = {x ∈ C T N ×1 ||xn | = Pmax /(T N ), 1 ≤
x − x0 22 (x − x0 ∞ ) ≤ , (15) n ≤ T N }. As all the problems are nonconvex but have dif-
ferentiable objective functions with respect to both w and x,
where x0 is the reference waveform and  ∈ R is a parameter we adopt the alternating minimization framework, as discussed
for controlling the degree of the similarity. In the case of a above. The steps of the derived algorithm are given by,
communication system, such a strict constraint would have
severe impact on the performance of the system, as well. Since x(k+1) = arg min g(x, w(k) ) or g (x, w(k) )
x∈XT E or x∈XCM
the similarity constraint is related to the radar system only, a
(16)
more flexible approach would be to connect this constraint to
the trade-off parameter ρ, used to tune the performance between 1
w(k+1) = arg min(1 − ρ) , (17)
the radar and communication parts of the DFRC system. This w γ(x(k+1) , w)
can be done by relaxing the similarity constraint and bringing
it as a penalty term on the cost function of the optimization where x(k) and w(k) are the solutions available to problems (16)
problem under consideration. More specifically for the cases of and (17), respectively, at the k-th iteration. As can be seen from
the TE and CM constraints we have, (17), the first term of the original objective functions g(x, w)
and g (x, w) is ignored, as it is independent of w.
1 Problem (17) can be simplified as,
(P3 ) : min ρf (x) + (1 − ρ) + (1 − ρ)λx − x0 22
x,w γ(x, w)
1
w(k+1) = arg max γ(x(k+1) , w). (18)
s.t. x22 ≤ Pmax , w (1 − ρ)
1384 IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 15, NO. 6, NOVEMBER 2021

Let us now proceed with the derivation of the projection

Algorithm 1: Joint Waveform and Filter Design for the
operator ΠXT E {·}. The constraint related to set XT E may be
DFRC System.
enforced by solving the following optimization problem,
1: Initialize k = 0, w ∈ C RN ×1 and x ∈ XT E for (P1 )
and (P3 ) or x ∈ XCM for (P2 ) and (P4 ); (P5 ) : min x† − x22
x†
2: while not converged do
3: k ← k + 1 s.t. x† ∈ XT E ,
4: Compute the gradient of g for (P1 ) and (P2 ) from where x† is the projection of x onto XT E . Problem (P5 ) is
(23) or g for (P3 ) and (P4 ) from (24); convex and from its KKT conditions, it can be shown that
5: Update d(k+1) using (21); 
6: Compute ΠXT E {d(k+1) } for (P1 ) and (P3 ) from † x, x22 ≤ Pmax
x = √ . (25)
(25) or ΠXCM {d(k+1) } for (P2 ) and (P4 ) from (26); x
Pmax x2 , x22 > Pmax
7: Update w(k+1) from (20);
8: end while The projection operator ΠXCM {·} may be derived by solving
return x, w a similar problem to (P5 ), i.e.,
(P6 ) : min x‡ − x22
x‡
Now problem (18) can be further written as, s.t. x‡ ∈ XCM ,
K 
1  where x‡ is the projection of x onto the set XCM . As can be seen,
max wH σk2 A(θk )x(k+1) (x(k+1) )H AH (θk ) w
w (1 − ρ) (P6 ) is separable in each one of the elements x‡n and xn , 1 ≤
k=1
n ≤ T N , of vectors x and x‡ , respectively. One may show that
+ σu2 wH w these decoupled scalar problems admit a closed-form solution
given by
s.t. wH A(θ0 )x = 1. (19) ⎧ 
⎨ Pmax
The above problem admits the following closed-form solution, ‡ TN , xn = 0
xn =  . (26)
⎩ Pmax xn , x = 0
(B(k) )−1 A(θ0 )x(k+1) T N |xn | n
w(k+1) = , (20)
(x(k+1) )H AH (θ0 )(B(k) )−1 A(θ0 )x(k+1) The complete algorithm summarizing the steps for solving prob-
where B = (k) K
σk2 AH (θk )x(k+1) (x(k+1) )H A(θk )
+ lems (P1 )-(P4 ) is presented in Algorithm 1.
k=1
σu2 I. Now, we focus on solving (16) via the GP framework. We close this subsection with a brief discussion on the compu-
The core of the GP framework [43] is the iterative steps, given tational complexity of Algorithm 1. As can be seen, the variable
by w is updated at every iteration using a closed-form solution
in (20) and the vector x is updated using a GP step which
d(k+1) = x(k) − α(k) ∇x ḡ(x(k) , w(k) ) (21) requires the computation of the objective function’s gradient.
  These two steps involve basic matrix-vector and matrix-matrix,
x(k+1) = ΠX d(k+1) , (22) multiplications and additions. The per-iteration complexity of
Algorithm 1 is O(T 3 N 3 ) as the primary operation that dom-
where α(k) is a suitable step-size parameter in the k-th iter- inates the computational complexity is the matrix inversion,
ation, ΠX {·} denotes the orthogonal projection onto set X , (B(k) )−1 . Note that the complexity of the proposed method is
ḡ(x(k) , w(k) ) can be either g(x(k) , w(k) ) or g (x(k) , w(k) ) and comparable to the one of the state of the art works in MIMO radar
X can be either XT E or XCM based on the optimization problem literature [24], [26], [29], [38]. As presented in the Introduction,
under consideration. Furthermore, the gradient vectors for those those works develop solutions for the simpler problem of the
two cases are given by, joint transmit waveform and radar receive filter design in radar
(|b(k) |2 G(k) x(k) ) only systems. That is, the proposed DFRC system design is able
∇x g(x(k) , w(k) ) = 2(1 − ρ) to accommodate the functions of both the radar and the com-
|b(k) |4
munication systems without significant impact on the required
(x(k) )H G(k) |b(k) |2 computational overhead.
− 2(1 − ρ)
|b(k) |4
B. Convergence Analysis
+ 2ρ(−H̃H s + H̃H H̃x(k) ), (23)
First we note that problems (P1 )-(P4 ) are solved by using the
and corresponding version of the derived Algorithm 1 which is based
∇x g (x(k) , w(k) ) = ∇x g(x(k) , w(k) ) + 2(1 − ρ)λ(x − x0 ), on the alternating optimization and GP frameworks. Moreover,
(24) all of them have a differentiable objective function. Thus, for
simplicity we prove the convergence for problem (P1 ) based on
respectively, where b(k) = (x(k) )H AH (θ0 )w(k) and G(k) = the results in [44] and a similar proof follows for the remaining
K 2 (k)
k=1 σk A(θk )w (w(k) )H AH (θk ). problems, as well.
TSINOS et al.: JOINT TRANSMIT WAVEFORM AND RECEIVE FILTER DESIGN FOR DFRC SYSTEMS 1385

Theorem 1: A sequence of iterates generated by Algorithm 1, where μ(∞) ∈ R+ denotes the optimal dual solution associated
{x(k) , w(k) }, converges to a KKT-point of problem (P1 ). with the TE constraint. Now utilizing (30) in (41), we obtain,
Proof: Problem (17) admits a closed-form solution and prob-
lem (16) is solved using a GP step. Therefore, first we analyze ∇x g(x(∞) , w(∞) ) + 2μ(∞) x(∞) = 0T N . (42)
the solution in (21), (22) of problem (16). It can be seen that the The condition (41) shows that x(∞) is a block-wise minimizer
solution in (21), (22) is a minimizer of the following function of problem (16). On the similar lines, we can show that,
over the considered constraint set, XT E ,
g(x(kj ) , w) ≥ g(x(kj ) , w(kj ) ) (43)
g̃(x; x(k) , w(k) ) = g(x(k) , w(k) )
  g(x(∞) , w) ≥ g(x(∞) , w(∞) ). (44)
+ Re ∇x g(x(k) , w(k) )H (x − x(k) )
The inequality (44) shows that w(∞) is a block-wise minimizer
1  2
  of function g and satisfies the following partial KKT-condition,
+ x − x(k)  , (27)
2α(k) 2
∇w g(x(∞) , w(∞) ) = 0RN . (45)
(k) 1 (k)
where α < L(k) and L is the smallest Lipschitz constant
Combining the partial KKT-conditions (42) and (45), we prove
of the function g at the k-th iteration. Now, it can be seen that
that a solution sequence generated by Algorithm 1 converges to
the function g̃ satisfies the following properties,
a KKT-point of problem (P1 ).
g(x, w(k) ) ≤ g̃(x; x(k) , w(k) ), ∀x (28)
V. NUMERICAL RESULTS
g(x(k) , w(k) ) = g̃(x(k) ; x(k) , w(k) ) (29)
In this section, numerical results are presented for evaluating
∇x g(x(k) , w(k) ) = ∇x g̃(x(k) ; x(k) , w(k) ), (30) the performance of the proposed techniques. A DFRC system
1 with T = 16 antennas at the transmitter and R = 8 antennas
where (28) follows from α(k) < L(k) . Now considering the
at the radar receiver is considered. The DFRC system serves
updates in (16) and (17), we have the following inequalites,
M = 4 UTs while aiming at the detection of a target source at
g(x(k) , w(k) ) = g̃(x(k) ; x(k) , w(k) ) (31) spatial angle θ0 = 15o . Furthermore, 3 interfering sources are
assumed to be located at spatial angles θ1 = −50o , θ2 = −10o
≥ g̃(x(k+1) ; x(k) , w(k) ) (32) and θ3 = 40o . The power of target signals is set to σ02 = 10 dB.
The power of the interference signals is set to σk2 = 30 dB for
≥ g(x(k+1) , w(k) ) (33)
k = 1, 2, 3. The noise variance at the radar receiver is equal to
≥ g(x(k+1) , w(k+1) ), (34) σu2 = 0 dB. The employed antenna ULAs are considered to have
half-wavelength inter-element separation. The symbols to be
where (31) follows from (29), (32) holds because of (16), (33) transmitted to the UTs are drawn uniformly from a Binary Phase
follows from (28) and (34) is obtained because the problem with Shift Keying (BPSK) modulation. The constellation points are
respect w is solved in closed-form as given by (20). normalized to unit power. The number of symbols per block
Let us now consider a convergent subsequence {x(kj ) , w(kj ) } is set to N = 20. The maximum allowed power per block of
with a limit point (x(∞) , w(∞) ). Based on the minimization steps symbols is set to Pmax = 20 W. The orthogonal linear frequency
(16) and (17) we have, modulation (LFM) is considered as a reference waveform. The
g̃(x; x(kj ) , w(kj ) ) ≥ g̃(x(kj +1) ; x(kj ) , w(kj ) ) (35) space-time matrix X0 of the LFM waveform is given by,

≥ g(x (kj +1)

,w (kj )
) (36) Pmax 2jπ (n−1)(l+n−1)
X0 (l, n) = e N , (46)
TN
(kj +1) (kj +1)
≥ g(x ,w ) (37)
where l = 1, . . . , T , n = 1, . . . , N and X0 (l, n) is the (l, n)th
≥ g(x(kj+1 ) , w(kj+1 ) ) (38) entry of X0 . The reference sequence x0 ∈ C T N ×1 is calculated
as x0 = vec(X0 ). Note that LFM waveforms present good pulse
= g̃(x(kj+1 ) ; x(kj+1 ) , w(kj+1 ) ), (39) compression and ambiguity properties while performing well in
where (35)–(39) can be shown by following similar arguments to distinguishing point targets [45]. On the other hand, they may
the ones used to prove (31)–(34), in the above. Now, by utilizing present poor performance in environments with clutter. Fixed
the continuity of the function g̃ and taking the limit j → ∞ on step-size parameters a(k) for all the algorithms are employed
both the sides of (39), we obtain, such that the optimal performance is attained. The results are
averaged over 100 different blocks of symbols and 1000 channel
g̃(x; x(∞) , w(∞) ) ≥ g̃(x(∞) ; x(∞) , w(∞) ). (40) realizations. The channel coefficients are modeled as CN (0, 1).
The inequality (40) implies that x(∞) is a block-wise minimizer
A. Benchmarks
of the function g̃(·) and satisfies the following partial-KKT
condition with respect to x as given by, Prior proceeding to the presentation of the results, let us pro-
vide a brief discussion about the benchmarks used for compari-
∇x g̃(x(∞) ; x(∞) , w(∞) ) + 2μ(∞) x(∞) = 0T N , (41) son purposes in the present paper. We start with the benchmark
1386 IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 15, NO. 6, NOVEMBER 2021

for the communication part of the DFRC system. This involves

the design of the transmit signals such that the MUI function in
(2) is minimized subject to the TE and CM constraints. The latter
is equivalent to setting ρ = 1 in (P1 ) and (P2 ), respectively. The
resulting optimization problems are defined as
(P1 ) : min f (x)
x

s.t. x22 ≤ Pmax ,

and
(P2 ) : min f (x)
x

Pmax
s.t. |xn | = , 1 ≤ n ≤ T N,
TN
for the TE and CM constraints, respectively. Note, that (P1 ) Fig. 3. Convergence study of the proposed algorithmic solutions for a DFRC
and (P2 ) can be directly handled by Algorithm 1, though they system with T = 16 antennas at the transmitter serving M = 4 UTs. The trade-
off and similarity parameters are set to ρ = 0.5 and λ = 1, respectively.
can be solved via more efficient approaches since they are much
simpler problems than (P1 ) and (P2 ). More specifically, (P1 )
admits a solution given by,
literature and thus, the approaches in [24], [26], [29], [38] can
x = (H̃H H̃ + ψIT )−1 H̃H s, (47) be also applied for deriving their solution.
where the Lagrange multiplier ψ ∈ R+ is set such that the
TE constraint is met, e.g. by the application of a bisection B. Convergence Study
method [43]. Problem (P2 ) can be efficiently solved via the In Fig. 3, the convergence of the proposed algorithmic so-
application of Algorithm 2 in [46]. lutions is experimentally studied when the trade-off control
In a similar manner for the radar part, the considered bench- parameter is set to ρ = 0.5. The average values of the objective
marks are involving the SINR maximization of the radar-only functions per iteration for the different approaches are examined.
system under the considered constraints. This is equivalent to As it is observed, all of the considered approaches are rapidly
setting ρ = 0 in optimization problems (P1 )-(P4 ), respectively. converging to a stationary point. It is noteworthy to point out
The resulting optimization problems are given by, that the case of the TE constraint (“TE” curve) converges to a
(P1 ) : max γ(x, w) stationary point that corresponds to a lower objective value than
x,w the one of the CM constraint (“CM” curve). This is the case,
s.t. x22 ≤ Pmax , since the feasible set of solutions is smaller for the CM case,
than the TE one.
The former is a subset to the one defined for the case of the
TE constraint. In other words, the performance of the CM case
(P2 ) : max γ(x, w)
x,w is lower bounded by that of the TE one. Similar conclusions can
be reached when the similarity constraints are included in the
Pmax previous two cases (“TE+SIM” and “CM+SIM,” respectively).
s.t. |xn | = , 1 ≤ n ≤ T N,
TN The similarity parameter is set to λ = 1. The case involving the
TE constraint converges to a “better” optimal point compared to
1 the one of the CM constraint for the common objective function
(P3 ) : min + λx − x0 22
x,w γ(x, w) in (P3 ) and (P4 ).
s.t. x22 ≤ Pmax ,
C. Bit Error Rate Performance of the Communication Part
and We start with the evaluation of the performance of the com-
1 munication part of the DFRC system. In Fig. 4, the average
(P4 ) : min + λx − x0 22
x,w γ(x, w) uncoded Bit Error Rate (BER) achieved by different techniques
versus the transmit signal-to-noise ratio (SNR) is depicted. The
Pmax trade-off parameter is set to ρ = 0.9. The transmit SNR is defined
s.t. |xn | = , 1 ≤ n ≤ T N,
TN as E{x22 }/σz2 . In more detail, a DFRC system is considered
for the TE, CM, TE+SIM, and CM+SIM constraints, respec- for the cases of 1) TE (“TE-DFRC”), 2) CM (“CM-DFRC”),
tively. The proposed algorithmic solutions in the present paper 3) TE+SIM (“TE+SIM-DFRC”) and 4) CM+SIM (“CM+SIM-
can handle directly (P1 ), (P2 ), (P3 ) and (P4 ), as well. Further- DFRC”) constraints. The similarity parameter is set to λ = 0.1 in
more, (P1 ) and (P2 ) have been extensively studied in the radar the corresponding cases. For comparison purposes, the cases of a
TSINOS et al.: JOINT TRANSMIT WAVEFORM AND RECEIVE FILTER DESIGN FOR DFRC SYSTEMS 1387

TABLE I
RADAR RECEIVE SINR AND DETECTION PROBABILITY - TOTAL TRANSMISSION
ENERGY CONSTRAINT

the two cases in Table I. In the latter table, the performance of the
corresponding radar-only systems is also shown for comparison
purposes (ρ = 0). These cases involve the performance of the
radar-only system under the TE and TE+SIM constraints. The
joint waveform and filter design in the aforementioned cases
is based on the solutions of (P1 ) and (P3 ), respectively, as
Fig. 4. BER versus transmit SNR curves of the DFRC system for the different described in Section V-A. The values of the trade-off parameter
techniques. The DFRC system has T = 16 transmit antennas and serves M = 4
UTs. The trade-off and similarity parameters are set to ρ = 0.9 and λ = 0.1,
ρ = {0.1, 0.5, 0.9} are considered for all the examined cases in
respectively. Fig. 5 and the Table I. For the TE+SIM case, results are presented
for two values of the similarity parameter, λ = 0.1 and λ = 1.
The previous applies both on the results shown in Figs. 5.(a)-(b)
communication-only system under the TE (“TE-COM-ONLY”) and Table I, as well. The column for λ = 0 in Table I corresponds
and the CM (“CM-COM-ONLY”) constraints are considered, to the TE constraint case.
as well. The waveform design for these cases is based on the In Fig. 5.(a), the performance of the DFRC system for the
solutions of (P1 ) and (P2 ), respectively. As it was expected, the TE constraint is depicted. As we can see, a decrease in the
performance is worse for the approaches involving the similarity value of ρ has in general small impact on the performance of
constraint (“TE+SIM-DFRC,” “CM+SIM-DFRC”) compared to the communication system while adding more weight on the
the ones without it (“TE-DFRC,” “CM-DFRC”). performance of the radar part. On the other hand, decreasing
This is the case, since the similarity constraint enforces a the ρ parameter’s value from 0.9 to 0.5 results in a significant
specific structure to the transmission signals. In general, this improvement on the radar receive SINR (Table I). As a conse-
structure does not promote the minimization of the MUI energy quence, the probability of detection is significantly improved,
(2) which is related to the performance of the communication as well. Thus, it can be concluded that the DFRC system with
part, as discussed in Section III. Furthermore, the dual radar- the TE constraint has a communication part that appears to be
communication function comes at a cost on the performance robust to the changes of the trade-off parameter ρ. That is, the
of the communication part, as it can be seen by comparing ρ parameter can be tuned easily to a value where both the radar
the curves of the DFRC systems to the communication-only and the communication parts can perform well and very close
counterparts. This impact on the performance is more severe to the radar- and communication-only counterparts.
for the TE constraint, as the “TE-DFRC” curve presents ap- In Fig. 5.(b), the achievable sum-rate of a DFRC system
proximately 4 dB loss on the performance when compared to under the TE+SIM constraint is presented. As can be seen, the
its communication-only counterpart (“TE-COM-ONLY”). On similarity constraint has severe impact on the performance of
the other hand, for the CM constraint case, the DFRC system the DFRC system due to the imposed strict structure on the
(“CM-DFRC”) presents very small performance loss compared transmitted waveforms. This was also discussed in the results
to its communication-only counterpart (“CM-COM-ONLY”). of Fig. 4 in Section V-C. Thus, if high similarity is sought,
Note that in general, the systems under the TE constraint achieve significant degradation is observed on the achievable rate. This
better performance than the ones under the CM constraint for the can be verified by checking the results for ρ = 0.1 and λ = 1
reasons mentioned in the discussion of the convergence results in Fig. 5.(b). Similar conclusions can be reached by inspecting
in Fig. 3. Table I. High values of the λ parameter, impose high similarity
to the reference waveform though this degrades the radar receive
SINR and the detection probability, as well. Of course, by
D. Total Transmission Energy Constraint appropriately tuning the ρ and λ parameters, one may select
We now move to the evaluation of the performance of the the desired point of function based on the performances of the
DRFC system under the TE and the TE+SIM constraints. To radar and the communication parts. Furthermore, the results in
that end, we plot the achievable sum-rate (5) of the two cases Table I verify the relation between the radar receive SINR and
versus the transmit SNR in Figs. 5.(a)–(b). The performance the detection probability. That is, the latter is a monotonically
of the communication-only system under the TE constraint is increasing function of the former.
depicted in both figures for comparison purposes. In addition, for For further insights on the radar part performance, the beam-
evaluating the performance of the radar part, as well, we present pattern P (θ) is plotted for both the TE and TE+SIM cases in
the radar receive SINR and the detection probability achieved by Figs. 6.(a)-(b). By denoting by x and w , the optimal waveform
1388 IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 15, NO. 6, NOVEMBER 2021

Fig. 5. Achievable sum-rate versus transmit SNR for a DFRC system with T = 16 transmit antennas serving M = 4 users. (a) Total Transmission Energy
Constraint, (b) Total Transmission Energy-plus-Similarity Constraint.

Fig. 6. Beampattern for a system with T = 16 transmit and R = 8 radar receive antennas. (a) Total Transmission Energy Constraint, (b) Total Transmission
Energy-plus-Similarity Constraint.

Fig. 7. Achievable sum-rate versus transmit SNR for a DFRC system with T = 16 transmit antennas serving M = 4 users. (a) Constant-Modulus Constraint,
(b) Constant-Modulus-plus-Similarity Constraint.
TSINOS et al.: JOINT TRANSMIT WAVEFORM AND RECEIVE FILTER DESIGN FOR DFRC SYSTEMS 1389

Fig. 8. Beampattern for a system with T = 16 transmit and R = 8 radar receive antennas. (a) Constant-Modulus Constraint, (b) Constant-Modulus-plus-Similarity
Constraint.

Fig. 9. Phase of the derived waveforms for different values of the trade-off parameter ρ. A system with T = 16 transmit and R = 8 radar receive antennas is
considered. (a) Total Transmission Energy Constraint, λ = 1, (b) Total Transmission Energy Constraint, λ = 0.1, (c) Constant-Modulus Constraint, λ = 1, (d)
Constant-Modulus Constraint, λ = 0.1.
1390 IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 15, NO. 6, NOVEMBER 2021

TABLE II In Fig. 7.(b), the achievable sum-rate of a DFRC system with

RADAR RECEIVE SINR AND DETECTION PROBABILITY - CONSTANT-MODULUS
CONSTRAINT
the CM+SIM constrainted is presented. In a similar manner to
the results in Fig. 5.(b), a high value for the similarity parameter
results in degradation on achievable sum rate of the communi-
cation part. The results in Table II, show also a similar impact on
the SINR and, as a consequence, to the probability of detection
of the system. Again, it can be deduced that by appropriately
tuning the ρ and λ parameters, one may achieve satisfactory
performance for both the communication and the radar part of
and radar receive filter, respectively, the beampattern is given by, the DFRC system. It is noteworthy that the results in Table II,
verify also the monotonically increasing relation between the
P (θ) = | (w )H A(θ)x |2 , (48) radar receive SINR and the detection probability.
where A(θ) is defined in (11). In Figs. 8.a-b, the beampattern P (θ) is plotted for both the CM
Both of the examined techniques are performing well, in a and CM+SIM constraint case. Similar conclusions are drawn
sense that they are placing relative deep nulls in the spatial with the ones of Figs. 6.a-b. Furthermore, the results are in line
directions of the interfering sources. Furthermore, the results with those reported in Table II.
are in line with the ones presented in Table I. The lower the ρ
value is, the deepest is the null placed on an interfering source. F. Impact of the Similarity Constraint on the Transmit
As a consequence, the higher is the radar receive SINR/detection Waveform
probability. Furthermore, high values for the similarity param- As the final part of this section, the impact of the similarity
eter λ result, in general, in less deep nulls on the interfering constraint on the transmit waveform is examined. To do so, the
sources. This results in the degradation of the radar receive angles of the transmit waveforms designed by the techniques
SINR/detection probability performance. involving the TE+SIM and CM+SM constraints are presented
in Figs. 9.(a)-(b) and Figs. 9.(c)-(d) for λ = 0.1 and λ = 1, re-
E. Constant-Modulus Constraint spectively. For comparison purposes the angles of the reference
signal x0 are depicted on all these figures. The first 100 samples
We move now to the case of the DFRC system that func-
of each signal are shown for ρ = {0.1, 0.5, 0.9}. As expected,
tions under the CM and CM+SIM constraints. The achievable
high λ values result in a better approximation of the x0 angles.
sum-rate is plotted for the latter cases versus the transmit SNR in
On the contrary, small λ values result in a poorer approximation
Figs. 7.(a)–(b). In the same figures, the performance of the corre-
of the reference signal. Moreover, a decrease in the trade-off
sponding communication-only systems is shown. The radar re-
control parameter ρ, results in a better approximation of the x0
ceive SINR and the probability of target detection are presented
angles. This is the case since more weight is considered on the
in Table II. Following the presentation of the results in Table I,
performance of the radar part over the communication one.
Table II includes also the SINR and the probability of detection
achieved by the radar-only systems (ρ = 0). That is, problems
VI. CONCLUSION
(P2 ) and (P4 ) are solved for the CM and CM+SIM constraints
case, respectively. The values of the trade-off parameter are set In this paper, joint designs for the transmit waveform and
again to ρ = {0.1, 0.5, 0.9} for the results in Figs. 7.(a)–(b) and the radar receive filter of DFRC systems are proposed. A
Table II. The similarity parameter is set again to λ = 0.1 and multiple antenna BS serving multiple single antenna users on
λ = 1 for the CM+SIM case. The column for λ = 0 in Table II the downlink while aiming at the simultaneous detection of
corresponds to the CM constraint case. a radar target is assumed. The joint optimal designs for the
The performance of the DFRC system under the CM con- transmit waveform and the radar receive filter are derived such
straint only is shown in Fig. 7.(a). For the cases of ρ = 0.9 that different spatial/temporal constraints are satisfied. Four
and ρ = 0.5, the DFRC system achieves performance close to multi-objective optimization problems are formulated aiming
the one of the communication-only counterpart. For the case at jointly optimizing the MUI energy and radar receive SINR
of ρ = 0.1, a significant degradation on the performance is of the the communication and the radar part of the DFRC
presented for high SNR values. By inspecting the corresponding system, respectively. The problems are formulated based on the
values in Table II, it can be seen that both the radar receive weighted sum method that combines the two objective functions
SINR and the detection probability are quite low for ρ = 0.9. into a composite one. The latter approach enables a flexible
Furthermore, they are improving significantly when decreasing performance trade-off between the radar and the communication
the ρ parameter to the 0.5 value. Further gains can be seen on objectives. The resulting constrained optimization problems are
the SINR and the detection probability values when ρ = 0.1. difficult and nonconvex. The different cases of TE, CM and
Thus, it can be deduced that for the values of ρ that try to similarity to a known radar waveform constraints are consid-
balance the performance of the communication and the radar ered. Efficient algorithmic solutions are developed based on the
parts, satisfactory performance may be achieved for both of alternating optimization and GP frameworks. The convergence
them. If more weight is given to the communication side, the of the proposed solutions to a KKT-point of the respective
radar performance deteriorates and vice-versa. problems is theoretically established. The performance of the
TSINOS et al.: JOINT TRANSMIT WAVEFORM AND RECEIVE FILTER DESIGN FOR DFRC SYSTEMS 1391

proposed methods is studied via numerical results where it is [20] Y. Wang, X. Wang, H. Liu, and Z. Luo, “On the design of constant modulus
shown that one may achieve satisfactory performance for both probing signals for MIMO radar,” IEEE Trans. Signal Process., vol. 60,
no. 8, pp. 4432–4438, Aug. 2012.
the radar and communication parts, tailored for the needs of the [21] M. Soltanalian and P. Stoica, “Designing unimodular codes via quadratic
considered application, by appropriately tuning the values of the optimization,” IEEE Trans. Signal Process., vol. 62, no. 5, pp. 1221–1234,
corresponding parameters. Mar. 2014.
[22] J. Li, J. R. Guerci, and L. Xu, “Signal waveform’s optimal-under-restriction
design for active sensing,” IEEE Signal Process. Lett., vol. 13, no. 9,
pp. 565–568, Sep. 2006.
[23] A. De Maio, S. De Nicola, Y. Huang, S. Zhang, and A. Farina, “Code design
REFERENCES to optimize radar detection performance under accuracy and similarity
constraints,” IEEE Trans. Signal Process., vol. 56, no. 11, pp. 5618–5629,
[1] V. Cisco, “Cisco visual networking index: Forecast and trends, 2017- Nov. 2008.
2022,” White Paper, vol. 1, pp. 1–38, 2018. [24] C. Chen and P. P. Vaidyanathan, “MIMO radar waveform optimization
[2] J. Lin, W. Yu, N. Zhang, X. Yang, H. Zhang, and W. Zhao, “A survey on in- with prior information of the extended target and clutter,” IEEE Trans.
ternet of things: Architecture, enabling technologies, security and privacy, Signal Process., vol. 57, no. 9, pp. 3533–3544, Sep. 2009.
and applications,” IEEE Internet Things J., vol. 4, no. 5, pp. 1125–1142, [25] M. M. Naghsh, M. Soltanalian, P. Stoica, M. Modarres-Hashemi, A. De
Oct. 2017. Maio, and A. Aubry, “A doppler robust design of transmit sequence
[3] A. Zanella, N. Bui, A. Castellani, L. Vangelista, and M. Zorzi, “Internet of and receive filter in the presence of signal-dependent interference,” IEEE
Things for smart cities,” IEEE Internet Things J., vol. 1, no. 1, pp. 22–32, Trans. Signal Process., vol. 62, no. 4, pp. 772–785, Feb. 2014.
Feb. 2014. [26] G. Cui, H. Li, and M. Rangaswamy, “MIMO radar waveform design with
[4] Federal Communications Commission (FCC), Connecting America: The constant modulus and similarity constraints,” IEEE Trans. Signal Process.,
national broadband plan, 2010. [Online]. Available: https://www.fcc.gov/ vol. 62, no. 2, pp. 343–353, Jan. 2014.
general/national-broadband-plan [27] A. Aubry, A. De Maio, and M. M. Naghsh, “Optimizing radar waveform
[5] Federal Communications Commission (FCC), FCC Proposes Innovative and Doppler filter bank via generalized fractional programming,” IEEE J.
Small Cell Use in 3.5 GHz Band, 2012. [Online]. Available: https://www. Sel. Topics Signal Process., vol. 9, no. 8, pp. 1387–1399, Dec. 2015.
fcc.gov/document/fcc-proposes-innovative-small-cell-use-35-ghz-band [28] B. Tang and J. Tang, “Joint design of transmit waveforms and receive filters
[6] Defense Advanced Research Projects Agency (DARPA), Shared spec- for MIMO radar space-time adaptive processing,” IEEE Trans. Signal
trum access for radar and communications (SSPARC), 2016. [On- Process., vol. 64, no. 18, pp. 4707–4722, Sep. 2016.
line]. Available: http://www.darpa.mil/program/sharedspectrum-access- [29] L. Wu, P. Babu, and D. P. Palomar, “Transmit waveform/receive filter
for-radar-and-communications design for MIMO radar with multiple waveform constraints,” IEEE Trans.
[7] B. Paul, A. R. Chiriyath, and D. W. Bliss, “Survey of RF communications Signal Process., vol. 66, no. 6, pp. 1526–1540, Mar. 2018.
and sensing convergence research,” IEEE Access, vol. 5, pp. 252–270, [30] S. K. Mohammed and E. G. Larsson, “Per-antenna constant envelope
Dec. 2017. precoding for large multi-user MIMO systems,” IEEE Trans. Commun.,
[8] L. Zheng, M. Lops, Y. C. Eldar, and X. Wang, “Radar and communication vol. 61, no. 3, pp. 1059–1071, Mar. 2013.
coexistence: An overview: A review of recent methods,” IEEE Signal [31] M. Biguesh and A. Gershman, “Training-based MIMO channel estimation:
Process. Mag., vol. 36, no. 5, pp. 85–99, Sep. 2019. A study of estimator tradeoffs and optimal training signals,” IEEE Trans.
[9] F. Liu, C. Masouros, A. P. Petropulu, H. Griffiths, and L. Hanzo, “Joint Signal Process., vol. 54, no. 3, pp. 884–893, Mar. 2006.
radar and communication design: Applications, state-of-the-art, and the [32] A. Sabharwal, P. Schniter, D. Guo, D. W. Bliss, S. Rangarajan, and
road ahead,” IEEE Trans. Commun., vol. 68, no. 6, pp. 3834–3862, R. Wichman, “In-band full-duplex wireless: Challenges and opportu-
Jun. 2020. nities,” IEEE J. Sel. Areas Commun., vol. 32, no. 9, pp. 1637–1652,
[10] R. Saruthirathanaworakun, J. M. Peha, and L. M. Correia, “Opportunistic Sep. 2014.
sharing between rotating radar and cellular,” IEEE J. Sel. Areas Commun., [33] J. R. Guerci, “Cognitive radar: A knowledge-aided fully adaptive ap-
vol. 30, no. 10, pp. 1900–1910, Nov. 2012. proach,” in Proc. IEEE Radar Conf., 2010, pp. 1365–1370.
[11] J. A. Mahal, A. Khawar, A. Abdelhadi, and T. C. Clancy, “Spectral [34] A. Aubry, A. DeMaio, A. Farina, and M. Wicks, “Knowledge-aided
coexistence of MIMO radar and MIMO cellular system,” IEEE Trans. (potentially cognitive) transmit signal and receive filter design in signal-
Aerosp. Electron. Syst., vol. 53, no. 2, pp. 655–668, Apr. 2017. dependent clutter,” IEEE Trans. Aerosp. Electron. Syst., vol. 49, no. 1,
[12] B. Li, A. P. Petropulu, and W. Trappe, “Optimum co-design for spectrum pp. 93–117, Jan. 2013.
sharing between matrix completion based MIMO radars and a MIMO [35] G. Cui, X. Yu, V. Carotenuto, and L. Kong, “Space-time transmit code
communication system,” IEEE Trans. Signal Process., vol. 64, no. 17, and receive filter design for colocated MIMO radar,” IEEE Trans. Signal
pp. 4562–4575, Sep. 2016. Process., vol. 65, no. 5, pp. 1116–1129, Mar. 2017.
[13] B. Li and A. P. Petropulu, “Joint transmit designs for coexistence of [36] E. Bjornson, E. A. Jorswieck, M. Debbah, and B. Ottersten, “Multiobjec-
MIMO wireless communications and sparse sensing radars in clutter,” tive signal processing optimization: The way to balance conflicting metrics
IEEE Trans. Aerosp. Electron. Syst., vol. 53, no. 6, pp. 2846–2864, in 5G systems,” IEEE Signal Process. Mag., vol. 31, no. 6, pp. 14–23,
Dec. 2017. Nov. 2014.
[14] F. Liu, C. Masouros, A. Li, and T. Ratnarajah, “Robust MIMO beamform- [37] M. Alodeh et al., “Symbol-level and multicast precoding for multiuser
ing for cellular and radar coexistence,” IEEE Wireless Commun. Lett., multiantenna downlink: A state-of-the-art, classification and challenges,”
vol. 6, no. 3, pp. 374–377, Jun. 2017. IEEE Commun. Surv. Tut., vol. 20, no. 3, pp. 1733–1757, Jul./Aug. 2018.
[15] F. Liu, C. Masouros, A. Li, T. Ratnarajah, and J. Zhou, “MIMO radar and [38] B. Friedlander, “Waveform design for MIMO radars,” IEEE Trans. Aerosp.
cellular coexistence: A power-efficient approach enabled by interference Electron. Syst., vol. 43, no. 3, pp. 1227–1238, Jul. 2007.
exploitation,” IEEE Trans. Signal Process., vol. 66, no. 14, pp. 3681–3695, [39] T. Naghibi and F. Behnia, “MIMO radar waveform design in the presence
Jul. 2018. of clutter,” IEEE Trans. Aerosp. Electron. Syst., vol. 47, no. 2, pp. 770–781,
[16] F. Liu, L. Zhou, C. Masouros, A. Li, W. Luo, and A. Petropulu, “To- Apr. 2011.
ward dual-functional radar-communication systems: Optimal waveform [40] C. Rapp, “Effects of HPA-nonlinearity on a 4-DPSK/OFDM-signal for
design,” IEEE Trans. Signal Process., vol. 66, no. 16, pp. 4264–4279, a digital sound broadcasting signal,” ESA Special Publication, vol. 332,
Aug. 2018. pp. 179–184, 1991.
[17] F. Liu, C. Masouros, A. Li, H. Sun, and L. Hanzo, “MU-MIMO com- [41] S. Haykin, Communication Systems. Hoboken, NJ, USA: John Wiley Sons,
munications with MIMO radar: From co-existence to joint transmis- Inc., 2008.
sion,” IEEE Trans. Wireless Commun., vol. 17, no. 4, pp. 2755–2770, [42] A. Farina and S. Pardini, “Track-while-scan algorithm in a clutter en-
Apr. 2018. vironment,” IEEE Trans. Aerosp. Electron. Syst., vol. AES-14, no. 5,
[18] X. Liu, T. Huang, N. Shlezinger, Y. Liu, J. Zhou, and Y. C. Eldar, “Joint pp. 769–779, Sep. 1978.
transmit beamforming for multiuser MIMO communications and MIMO [43] D. P. Bertsekas, “Nonlinear programming,” J. Oper. Res. Soc. vol. 48, no. 3,
radar,” IEEE Trans. Signal Process., vol. 68, pp. 3929–3944, Aug. 2020. pp. 334–334, 1999.
[19] S. H. Dokhanchi, B. S. Mysore, K. V. Mishra, and B. Ottersten, “A [44] J. Tranter, N. D. Sidiropoulos, X. Fu, and A. Swami, “Fast unit-modulus
mmWave automotive joint radar-communications system,” IEEE Trans. least squares with applications in beamforming,” IEEE Trans. Signal
Aerosp. Electron. Syst., vol. 55, no. 3, pp. 1241–1260, Jun. 2019. Process., vol. 65, no. 11, pp. 2875–2887, Jun. 2017.
1392 IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL. 15, NO. 6, NOVEMBER 2021

[45] M. A. Richards, Fundamentals of Radar Signal Processing. New York, Symeon Chatzinotas (Senior Member, IEEE) is cur-
NY, USA: McGraw-Hill Educ., 2014. rently a Full Professor / Chief Scientist I and the Head
[46] S. Domouchtsidis, C. G. Tsinos, S. Chatzinotas, and B. Ottersten, of the SIGCOM Research Group at SnT, University
“Symbol-level precoding for low complexity transmitter architectures of Luxembourg, Luxembourg City, Luxembourg. He
in large-scale antenna array systems,” IEEE Trans. Wireless Commun., is coordinating the research activities on communi-
vol. 18, no. 2, pp. 852–863, Feb. 2019. cations and networking, acting as a PI for more than
20 projects and main representative for 3GPP, ETSI,
DVB. In the past, he was a Visiting Professor with the
University of Parma, Italy, lecturing on 5G Wireless
Networks. He has coauthored more than 450 technical
papers in refereed international journals, conferences
and scientific books. He was involved in numerous R&D projects for NCSR
Demokritos, CERTH Hellas and CCSR, University of Surrey, Guildford, U.K.
He was the co-recipient of the 2014 IEEE Distinguished Contributions to
Christos G. Tsinos (Senior Member, IEEE) received Satellite Communications Award and Best Paper Awards at EURASIP JWCN,
the Diploma degree in computer engineering and in- CROWNCOM, ICSSC. He is currently in the Editorial Board of the IEEE
formatics, the M.Sc. and Ph.D. degrees in signal pro- TRANSACTIONS ON COMMUNICATIONS, IEEE OPEN JOURNAL OF VEHICULAR
cessing and communication systems, and the M.Sc. TECHNOLOGY, and the International Journal of Satellite Communications and
degree in applied mathematics from the University Networking.
of Patras, Greece, in 2006, 2008, 2013, and 2014,
respectively.
From 2014 to 2020, he held a Postdoctoral Re- Björn Ottersten (Fellow, IEEE) received the M.S.
searcher, Research Associate and Research Scientist degree in electrical engineering and applied physics
Positions with the University of Patras and with the from Linköping University, Linköping, Sweden, in
University of Luxembourg, Luxembourg City, Lux- 1986, and the Ph.D. degree in electrical engineering
embourg. He has recently been elected as an Assistant Professor with the from Stanford University, Stanford, CA, USA, in
University of Athens, Greece. He is also a Research Fellow with the SnT, Uni- 1990. He has held a research positions with the De-
versity of Luxembourg. He was involved in a number of different Research and partment of Electrical Engineering, Linköping Uni-
Development projects funded by national and/or EU funds. His current research versity, the Information Systems Laboratory, Stan-
interests include optimization and machine learning for signal processing and ford University, the Katholieke Universiteit Leuven,
communications. Leuven, Belgium, and the University of Luxembourg,
Luxembourg. From 1996 to 1997, he was the Director
of Research with ArrayComm, Inc., a start-up in San Jose, CA, USA, based on
his patented technology. In 1991, he was appointed as a Professor of signal pro-
cessing with the Royal Institute of Technology (KTH), Stockholm, Sweden. He
is the Head of the Department for Signals, Sensors, and Systems, KTH, and the
Dean of the School of Electrical Engineering, KTH. He is currently the Director
of the Interdisciplinary Centre for Security, Reliability and Trust, University
of Luxembourg. He was the recipient of the IEEE Signal Processing Society
Technical Achievement Award, the EURASIP Group Technical Achievement
Award, and the European Research Council advanced research grant twice.
Aakash Arora (Student Member, IEEE) received He has coauthored journal papers that were the recipient of the IEEE Signal
the M.S. (Research) degree in electrical engineering Processing Society Best Paper Award in 1993, 2001, 2006, 2013, and 2019, and
from the Indian Institute of Technology Delhi, New eight IEEE conference papers best paper awards. He is a Board Member of the
Delhi, India, in 2017, and the Ph.D. degree in com- IEEE Signal Processing Society, the Swedish Research Council and currently
puter science from the Interdisciplinary Centre for serves of the boards of EURASIP, and the Swedish Foundation for Strategic
Security, Reliability and Trust, University of Lux- Research. Dr. Ottersten has served as the Editor-in-Chief of the EURASIP Signal
embourg, Luxembourg. He is currently holds a Re- Processing, and acted on the Editorial Boards of the IEEE TRANSACTIONS
search Associate position with the Interdisciplinary ON SIGNAL PROCESSING, IEEE SIGNAL PROCESSING MAGAZINE, IEEE OPEN
Centre for Security, Reliability and Trust, Univer- JOURNAL FOR SIGNAL PROCESSING, EURASIP Journal of Advances in Signal
sity of Luxembourg. His research interests include Processing, and Foundations and Trends in Signal Processing. He is a Fellow
optimization algorithms, statistical signal processing, of EURASIP.
wireless communications, and machine learning.