2021 IEEE/CIC International Conference on Communications in China (ICCC Workshops)

Hybrid Beamforming for DFRC System Based on

SINR Performance Metric
2021 IEEE/CIC International Conference on Communications in China (ICCC Workshops) | 978-1-6654-3944-2/21/$31.00 ©2021 IEEE | DOI: 10.1109/ICCCWorkshops52231.2021.9538857

Yuanzhe Dai1 , Kaifeng Han2 , Guo Wei1 , Ying Du2 , Zhiqin Wang2
1
Department of Electronic Engineering and Information Science,
University of Science and Technology of China, Hefei, China
2
China Academy of Information and Communications Technology, Beijing, China
Email: 1 [email protected], 1 [email protected], 2 {hankaifeng, duying1, zhiqin.wang}@caict.ac.cn

Abstract—Dual-functional radar-communication (DFRC) is a proposed in [7], and a manifold optimization-based algorithm

promising solution to simultaneously implement radar detection was proposed in [5] using the minimum sum-MSE criterion.
and communication. To further reduce the hardware cost, the In addition, the phased-MIMO radar, which trades-off be-
hybrid analog-digital (HAD) beamforming problem on DFRC
system is investigated in this paper. Specifically, the problem tween phased-array and MIMO radars, is also based on the
is to maximize the SINR of radar receiver under the non- hybrid analog-digital (HAD) structure. The work in [8] proved
convex individual SINR constraints of communication users that phased-MIMO radar with overlapped subarrays combines
(CUs) while guaranteeing the constant modulus constraints of the advantages of the phased array and MIMO radars. More-
analog beamformer. To solve this problem, we propose an itera- over, in the case of nonoverlapped subarrays, the transmit
tion algorithm, which optimizes digital and analog beamformer
iteratively. Both digital and analog beamforming designs are beamformer of was optimized in [9] to improve the Cramér-
formulated as non-convex quadratically constrained quadratic Rao bound (CRB) on target parameter estimation.
programs (QCQPs), and we use semidefinite relaxation (SDR) It can be observed that phased-MIMO radar and hybrid
and feasible point pursuit successive convex approximation (FPP- beamforming (HBF) techniques are able to combine naturally
SCA) techniques to solve them respectively. Finally, we evaluate in DFRC system from the perspective of hardware. The design
the proposed algorithm by numerical simulations, which show
that the performance of the proposed hybrid beamforming design of DFRC system in HAD structure is still an open problem and
is close to that of the full-digital beamforming. has both challenges from DFRC and HAD structure, such as
Index Terms—Spectrum sharing, radar-communication, hy- the coupling between radar and communication performances,
brid beamforming, signal to interference and noise ratio. the coupling between digital and analog beamformers and the
non-convex nature of constant modulus constraints of analog
I. I NTRODUCTION beamformer. It is worth pointing out that there have been some
researches on hybrid beamforming for DFRC systems [10]–
Communication and radar spectrum sharing has brought [13]. These works are all focused on the transmit beampattern
about the wide attention due to the increasingly congested or similarity constraints of transmit beamformer. However, the
frequency spectrum. One promising solution to spectrum shar- radar detection rate is directly determined by the SINR of radar
ing between radar and communication is dual-function radar- receiver [14]. Therefore, we consider using SINR as the radar
conmmunication (DFRC) system, where the two systems share performance metric to design the HAD beamformer.
the same hardware platform and spectral resources. And the In this paper, we propose a hybrid transmit beamforming
design of DFRC system has been studied extensively [1]. design for DFRC system, where the SINR of radar receiver is
Beamforming is an essential subject in the design of DFRC maximized under the SINR constraints of single antenna users.
systems. To improve the performance of the radar while The problem is formulated as a non-convex optimization prob-
guaranteeing the signal to interference and noise ratio (SINR) lem. To solve it, this paper proposes an iteration algorithm,
constrains of communication users (CUs) , the transmit beam- which optimizes digital beamformer and analog beamformer
former was optimized to match the desired radar beampattern iteratively. Both digital and analog beamforming optimizations
in [2]. In addition to transmit beampattern, the mean squared can be formulated as non-convex quadratically constrained
cross correlation pattern also is considered in [3]. Moreover, quadratic programs (QCQPs), and can be solved by semidef-
the work in [4] jointly designed transmit and receive beam- inite relaxation (SDR) and feasible point pursuit successive
forming to maximize the SINR of radar receiver. convex approximation (FPP-SCA) techniques respectively. In
Furthermore, to deal with the high hardware cost and the end, the performance of the proposed algorithm is evalu-
power consumption caused by large-scale antenna array, hy- ated by numerical simulations.
brid beamforming was studied in many literatures [5]–[7]. The
II. P ROBLEM S TATEMENT
work in [6] proposed a spatially sparse precoding algorithm
via orthogonal matching pursuit (OMP). To further improve A. System Model
the performance of analog beamformer, an iterative coordinate In order to share spectrum and hardware between radar and
descent algorithm which maximizes the spectral efficiency was communication functions, we consider a DFRC system which

978-1-6654-3944-2/21/$31.00
Authorized ©2021
licensed use limited to: INDIAN INSTITUTE TECHNOLOGY DELHI.82
OF IEEE Downloaded on April 05,2023 at 21:29:25 UTC from IEEE Xplore. Restrictions apply.
 2021 IEEE/CIC International Conference on Communications in China (ICCC Workshops)

UE where λ is wavelength, and d is the spacing between adjacent

BS transformer 1
antennas of ULA. Then BS uses a full-digital receive beam-
RF chain Nt
s
former w ∈ C Nr ×1 to maximize the SINR of radar receiver.

...

...
T
Let A (θ) = ar (θ) at (θ) , the SINR of radar receiver can be
Digital Analog
Beamformer Beamformer UE expressed by
...

...
...
FBB FRF K

α0 wH A (θ0 ) x2
 
K
NRF
Target of interest γr = "  2 #
 PQ 
BS receiver E wH αi A (θi ) x + wH w
i=1 (4)
 2
Digital α0 wH A (θ0 ) FRF FBB s
Probing signal r MVDR = ,
Filter wH Rin w
w ...
Reflaction signal
Interference where
Nr
Q
X 2 H
Rin = |αi | A (θi ) FRF FBB FH H
BB FRF A (θi ) + I. (5)
Fig. 1. System model of the HAD structure DFRC system
i=1

Notice that we can resort to the well-known minimum

implements radar target detection and downlink communica- variance distortionless response (MVDR) beamformer for
tion simultaneously. As illustrated in Fig.1, the DFRC system SINR maximization [15], which minimize the interference and
has a base station (BS) and K single-antenna CUs. Assume noise power while maintaining a distortionless response to the
that BS transmitter uses the fully connected HAD structure direction of target of interest. The closed form solution of
with NRF radio frequency (RF) chains and Nt antennas, and MVDR receive beamformer can be expressed as
BS receiver is a full-digital receiver with Nr antennas. The
R−1
in A (θ0 ) x
transmit and receive array are placed close to each other thus wopt = , (6)
the direction of departure (DoD) and direction of arrival (DoA) x A (θ0 ) R−1
H H
in A (θ0 ) x
are same for BS. then according to (1), (4) and (6), the average SINR of the
Giving the baseband signals s ∈ C K×1 , in which si ∼ radar receiver can be written as
CN (0, 1), i ∈ {1, ..., K} is the i.i.d. information symbol of Ns
the i-th CU. The transmit signal can be given by   X
γr = E xH Φx = H
fBB,k FH


RF ΦFRF fBB,k , (7)
k=1
x = FRF FBB s, (1)
where
2
where FBB ∈ C NRF ×K is the digital beamforming matrix, Φ = |α0 | AH (θ0 ) R−1
in A (θ0 ) , (8)
and FRF ∈ C Nt ×NRF is the analog beamforming matrix.
Beamforming matrixes satisfy kFRF FBB k2 6 P0 , where P0 and fBB,k is the k-th column of FBB .
is the transmit power constraint of BS. Furthermore, analog Further, we consider the receive signal of each single
beamformer is implemented by analog phase shifters, thus antenna CU. The receive signal of the k-th CU can be written
each element of analog beamforming matrix satisfy constant as
modulus constraint |FRF (i, j)| = 1 ∀i, j. K
X
Suppose there is a target of interest which located at angle yk = hH
k FRF fBB,k sk + hH
k FRF fBB,i si +nk , (9)
θ0 , and Q signal-dependent interference source that are located i6=k
at angle θi , i ∈ {1, ..., Q}. According to (1), the received signal where hk ∈ C Nt ×1 is the channel vector between BS and the
of BS can be expressed as k-th CU, and nk ∼ CN (0,1) is the AWGN of the k-th CU
Q
X receiver. Then the SINR of k-th CU can be written as
y= α0 ar (θ0 ) aTt (θ0 ) x + αi ar (θi ) aTt (θi ) x + n, (2) h FRF fBB,k 2
 H 
k
i=1 γk = N
. (10)
Ps  H 2
where α0 and αi is complex reflection coefficients of interest 1+ h FRF fBB,i 
k
i6=k
target and interference targets, respectively. n is additive
white Gaussian noise (AWGN) with each element ni ∼ B. Problem Formulation
CN (0, 1), i ∈ {1, ..., Nr }. at (θ) and ar (θ) are transmit and For DFRC system, there is a tradeoff between radar and
receive steering vector. For an N -antenna uniform linear array communication performance. We use SINR as the criterion of
(ULA), the steering vector can be written as beamforming design, maximizing the SINR of radar receiver
h 2πd 2πd(N −1)
iT under the individual SINR constraints of CUs. Meanwhile, the
a(θ) = 1, ej λ sin(θ) , ..., ej λ sin(θ)
∈ C N ×1 , (3) transmitter satisfies the power constraint and constant modulus

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI.83

Downloaded on April 05,2023 at 21:29:25 UTC from IEEE Xplore. Restrictions apply.
 2021 IEEE/CIC International Conference on Communications in China (ICCC Workshops)

constraints of analog beamforming matrix. Then the transmit solved through SDR technique by dropping the rank-1 con-
H
beamforming design problem can be expressed as straints, i.e. FBB,k = fBB,k fBB,k , then the problem (P2.1)
Ns can be relaxed as
X
H
FH


max fBB,k RF ΦFRF fBB,k K  
{fBB,k },FRF
X
k=1 max tr Φ̃0 FBB,k
{FBB,k }
h FRF fBB,k 2
 H 
k=1
k
s.t.  > Γk , ∀k,
 
h FRF fBB,i 2 tr H̃k FBB,k
P H
(P1.1) 1+ k (11) X  
i6=k (P2.2) s.t. − tr H̃k FBB,i ≥ 1, ∀k,
Γk
i6=k
|FRF (i, j)| = 1, ∀i, j,
K
Ns X
tr FH


RF FBB,k FRF 6 P0 , FBB,k  0, ∀k,
X
H
fBB,i FH
RF FRF fBB,i 6 P0 ,
k=1
i=1 (13)
where Γk is the threshold of the k-th CU’s SINR. where H̃k = h̃k h̃H k . Problem (P2.2) is a convex problem and
This problem involves joint optimization over the digital it can be efficiently solved by CVX toolbox in MATLAB.
beamformer and analog beamformer. However, it is usually The computational
complexity of solving this SDR problem
difficult to find their solutions simultaneously. Moreover, the is O K 4 NRF 2.5
[16]. Notice that the rank-1 constraints have
non-convex nature of individual SINR constraints of CUs and been dropped. However, it can be shown that the optimal
constant modulus constraint of analog beamforming matrix solution of problem (P2.2) has the rank-1 property by proving
makes this problem more challenging to solve. To address the following proposition with the theorem proposed in [17].
these challenges, this paper proposes an iterative algorithm Proposition 1: If the problem (P2.2) is feasible and bounded,
for obtaining a sub-optimal solution to (11). then there is always a optimal solution this problem that satisfy
rank(F?BB,k ) = 1, ∀k.
III. P ROPOSED B EAMFORMING D ESIGN A LGORITHM
Proof: See Appendix A.
In this section, we respectively provide digital and analog Therefore the optimal digital beamforming vector for k-th
beamforming design method, and then propose an overall CU can be given by fBB,k ? ?
with fBB,k ?H
fBB,k = F?BB,k .
iterative algorithm to solve the problem (P1.1). Notice that
both digital and analog beamforming design are non-convex B. Analog Beamforming Design
problem and there is no closed-form solution. Thus, we use
SDR technique and FPP-SCA algorithm to optimize the digital For analog beamforming, there are not only SINR con-
and analog beamforming matrices respectively. straints of CUs, but also constant modulus constraints of
analog beamforming matrix, and all of these constrains are
A. Digital Beamforming Design non-convex. Assume that we have obtained a fixed digital
Assuming that we already have a feasible solution of analog beamformer FBB . Then we use the sequential optimization
beamforming matrix FRF . For the sake of conciseness, let and define the constant matrix Φ0 as we did in III-A. The
Φ̃ = FH RF ΦFRF , h̃k = hk FRF , k ∈ {1...K}. Notice that
analog beamforming design problem can be written as
Φ̃ is not a constant matrix, thus the SINR of radar receiver K
(7) is a nonlinear function of the digital beamforming vector
X
tr FH


max RF Φ0 FRF FBB,k
fBB , which makes this problem difficult to solve. Therefore, FRF
k=1
we adopt the sequential optimization algorithm to optimize h FRF fBB,k 2
 H 
digital beamformer in an iterative fashion, where we fix Φ̃ k
s.t. P  H  2 > Γk ,
as a constant matrix Φ̃0 in each iteration process. Then the (P3.1) 1+ hk FRF fBB,i  (14)
digital beamforming optimization problem in each iteration i6=k
K
can be written as X
tr FH


Ns RF FBB,k FRF 6 P0 , ∀k,
X k=1
H
max fBB,k Φ̃0 fBB,k (12a)
{FBB,k }
k=1
|FRF (i, j)| = 1,
 2
 H H
h̃k fBB,k 
 where FBB,k = fBB,k fBB,k . Due to the SINR constraints of
(P2.1) s.t. P  H 2 > Γk , ∀k, (12b) CUs and constant modulus constraints, this is a non-convex
1+ h̃k fBB,i  QCQP so it is difficult to solve it directly. Nevertheless, using

i6=k the FPP-SCA method proposed in [16], a suboptimal solution
Ns
X
H
to (P3.1) can be obtained. To adopt FPP-SCA method, first
fBB,i FH
RF FRF fBB,i 6 P0 . (12c) we need to rewrite (P3.1) into the general form in [16]. By
i=1
the property of Kronecker product
This is still a non-convex QCQP problem because (12b)

T
tr (ABCD) = vec DT CT ⊗ A vec (B) ,


are non-convex constraints. Nevertheless, it can be efficiently (15)

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI.84

Downloaded on April 05,2023 at 21:29:25 UTC from IEEE Xplore. Restrictions apply.
 2021 IEEE/CIC International Conference on Communications in China (ICCC Workshops)

where A, B, C and D are arbitrary matrices, the problem Algorithm 1 Hybrid Transmit Beamforming Design Algo-
(P3.1) can be equivalently rewritten as rithm
Initialize: {θ0 , θ1 , ..., θQ }, {α0 ,q
α1 , ..., αQ }, [h1 , ..., hK ].
max uH BT ⊗ Φ0 u


(0)
u
  Initialization: Let {FBB,k } = NtPN0RF [1, ..., 1]T , and for
T K (0) (0)
F BB,k ⊗ Hk X (i, j)-th element of FRF we let FRF (i, j) = ejϕ(i,j) , where
s.t. uH  − FTBB,j ⊗ Hk  u
Γk φ(i, j) is the phased of the (i, j)-th element of the conjugate
(P3.2) j6=k (16) transpose of the composite downlink channel, i.e., H =
> 1, ∀k, [h1 , ..., hK ]. Set l = 0.
uH BT ⊗ IN t u 6 P0 ,


Repeat
uH Em u 6 1, uH Em u > 1, ∀m , Set l = l + 1.
(l)
PK 1) Calculate digital beamforming matrix FBB as follows.
where B = i=1 FBB,k , u = vec (FRF ). And Em ∈ • Initialize m = 0 and convergence threshold δ1 .
C Nt NRF ×Nt NRF , where the m-th diagonal element of Em Repeat
is 1, and the remaining elements are 0. Then the following
– Set m = m + 1.
proposition shows the problem relaxed by FPP-SCA. (m−1)
– Calculate Φ̃0 = Φ̃(m−1) and γr according
Proposition 2: The problem (P3.2) can be relaxed
PK into the
T to (8) and (7), respectively.
problem (P3.4) by FPP-SCA, where B̄ = i6=k BB,i ⊗
F
– Calculate the optimal {F?BB,k } according to (13),
Hk , ∀i. Moreover, r, sk , pm and vm are slack variables. λ is
and decompose F?BB,k = fBB,k ? ?H
fBB,k to get the
penalty variable. z ∈ C Nt Nrf is an arbitrary vector, for which ?
optimal fBB,k .
we set z(n) = u(n−1) in the n-th iteration. (m)
Proof: See Appendix B. – Calculate Φ̃(m) and γr according to (8) and
The problem (P3.4) is a convex problem and can be solved (7).
(m) (m−1)
by CVX toolbox in MATLAB. The computational complexity Until |γr − γr | 6 δ1 .
of solving this problem is O([K + Nt Nrf ]3.5 ) [16]. (l)
2) Calculate analog beamforming matrix FRF as follows.
Compared to the traditional SCA method, FPP-SCA adds (l−1)
•Initialize m = 0, z(m) = vec(FRF ), and conver-
slack variables to sustain feasibility thus it does not need
gence threshold δ2 .
a feasible initial point. And the penalty method is used to
Repeat
force the slack variables toward zero, which helps pushing the
iterates towards the feasible point of problem (P3.1). Finally – Set m = m + 1.
(m−1)
the solution will converge to a KKT point of the original – Calculate Φ̃0 = Φ̃(m−1) and γr according
problem (P3.1), and a suboptimal solution to (P3.1) can be to (8) and (7), respectively.
obtained. – Calculate the optimal u(m) according to (17), and
set z(m) = u(m) .
(m)
C. Overall Transmit Beamforming Design Algorithm – Calculate Φ̃(m) and γr according to (8) and
Above all, the overall iteration algorithm of hybrid transmit (7).
(m) (m−1)
beamforming design can be proposed as Algorithm 1. The Until |γr − γr | 6 δ2
digital beamforming optimization and analog beamforming Until convergence.
optimization are alternately repeated until convergence.
Remark 1: Assuming that T0 is average number of outer
iteration between digital and analog beamforming optimiza-
tion, and in each outer iteration, TD is the average number of the reflection coefficient of interference is set as |αi |2 = 30dB,
inner iteration of digital beamforming optimization, and TA is i = 1, 2, 3. The channel vector of each CU is scattering
the average number of inner iteration of analog beamforming channel which is randomly generated.
design. Then the computational complexity of Algorithm 1 is The overall beampatterns for different antenna numbers of
O(T0 [TD (K 4 NRF2.5
)) + TA (K + Nt NRF3.5
)]) the transmitter and receiver is shown in Fig. 2. Assume that
there are 2 CUs and the SINR constraint of each CU is 15dB.
IV. S IMULATION R ESULTS And the number of RF-chains at BS transmitter is set to 2. It
In this section, we evaluate the proposed hybrid transmit is easy to see that the main lobes of the beampatterns point to
beamforming design algorithm via numerical simulations. As- θ0 = 10°, where the target of interest located. And nulls are
sume that both the BS transmitter and receiver are equipped placed in the direction of interferences, i.e., θ1 = −60°, θ2 =
with ULAs with half-wavelength interval between adjacent −30°, θ3 = 40°. As the number of the DFRC BS antennas
antennas. We set the BS transmit power as P0 = 50dBm. increases, the width of the main lobe decreases and the PSLR
A target of interest is located at the spatial angle θ0 = 0°, increases. This suggests that the increasing of the number of
and there are 3 interference located at the spatial angles BS antennas leads to better radar performance.
θ1 = −60°, θ2 = −30°, and θ3 = 40°. The reflection In Fig. 3, the tradeoff between the SINR constraints of CUs
coefficient of target of interest α0 is set as |α0 |2 = 10dB, and and the average SINR of radar is shown from 1,000 Monte-

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI.85

Downloaded on April 05,2023 at 21:29:25 UTC from IEEE Xplore. Restrictions apply.
 2021 IEEE/CIC International Conference on Communications in China (ICCC Workshops)

max t − λr − λksk1 − λkpk1 − λkvk1

a,t,r,s,p,v

s.t. 2Re zH BT ⊗ Φ0 u − zH BT ⊗ Φ0 z > t − r, r > 0,




     
 FTBB,k ⊗ Hk  FTBB,k ⊗ Hk
(P3.4) uH −B̄ ⊗ Hk u + 2Re zH   u − zH 


 z > 1 − sk , sk > 0, ∀k,
 Γk  Γk

uH Em u 6 1 + pm , −zH Em z + 2Re zH Em u > 1 − vm , pm > 0, vm > 0, ∀m,



uH BT ⊗ IN t u 6 P0


(17)

20
41
0
40
-20

Radar SINR (dB)

-40 39
Beampattern(dBi)

-60
38
-80
37
-100
FPP-SCA
-120 36 Codebook
Full Digital
-140
35
-160 15 16 17 18 19 20 21 22 23 24 25
-80 -60 -40 -20 0 20 40 60 80 CU SINR (dB)
Angle(°)
Fig. 3. Tradeoff between the SINR constraints of CUs and the average SINR
Fig. 2. The overall beampatterns for different antenna numbers of DFRC BS, of radar for different design methods, Nt = Nr = 4, K = 2.
Nt = Nr , NRF = 2, K = 2, Γk = 15dB.

radar receiver has been maximized while satisfying the SINR

Carlo trials. It is clear to see that the average SINR of radar
constraints of CUs. To solve the joint HAD beamforming
receiver exponentially decreases as the SINR constraint thresh-
optimization problem, we have proposed an iteration algorithm
olds of CUs increase. In addition to our proposed method,
in which digital and analog beamformer were optimized
we also consider the full-digital structure proposed in [4] and
alternately. Both the digital and analog beamforming design
the codebook method which optimizes analog beamforming
problem have been formulated as non-convex QCQPs. The
matrix by exhaustive searching of DFT codebook. We assume
SDR technique have been adopted to obtain the optimal so-
that there are 2 CUs, and the BS transmitter is equipped
lution to digital beamforming optimization and the FPP-SCA
with 2 RF-chains and 4 transmit antennas. In full-digital
have been used to obtain the suboptimal solution to analog
case, the number of RF-chains is the same as the number
beamforming optimization. Numerical simulation results have
of antennas, and there are no constant modulus constraints.
shown that the performance of the proposed algorithm was
Therefore, the average SINR of radar receiver of full-digital
closed to that of the full-digital structure.
structure is higher than that of the HAD structure. However,
the performance of our proposed method is very close to A PPENDIX A
that of the full-digital structure. When the SINR constraint P ROOF OF P ROPOSITION 1
thresholds of CUs are 25dB, the performance loss of the
average SINR of radar is only about 0.5dB. Besides, the Consider a general form separable semidefinite programm-
performance of radar SINR of the codebook method is much ing (SDP) as follows:
worse than that of our proposed method. When the SINR L
X
constraints of CUs are 25dB, the radar SINR of the codebook min tr (Cl Xl )
{Xl }
method is 4.5dB lower than that of the FPP-SCA method. l=1
L (17)
V. C ONCLUSION X
s.t. tr (Aml Xl ) Cm bm , m = 1, ... , M,
In this paper, we have proposed a hybrid transmit beam- l=1
forming design for the DFRC system, in which the SINR of Xl  0, l = 1, ... , L,

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI.86

Downloaded on April 05,2023 at 21:29:25 UTC from IEEE Xplore. Restrictions apply.
 2021 IEEE/CIC International Conference on Communications in China (ICCC Workshops)

where {Cl }, {Aml } and {Xl } are Hermitian matrices (not [5] T. Lin, J. Cong, Y. Zhu, J. Zhang, and K. B. Letaief, “Hybrid beam-
necessarily positive semidefinite), bm ∈ R, Cm ∈ {>, 6, =}, forming for millimeter wave systems using the MMSE criterion,” IEEE
Trans. Commun., vol. 67, no. 5, pp. 3693–3708, May 2019.
∀m. [6] O. El Ayach, S. Rajagopal, S. Abu-Surra, Z. Pi, and R. Heath,
Suppose the above SDP and its dual is solvable. According “Spatially sparse precoding in millimeter wave MIMO systems,” IEEE
to [17], it always has an optimal solution {X?l } such that Trans.Wireless Commun., vol. 13, no. 3, pp. 1499–1513, Mar. 2014.
[7] F. Sohrabi and W. Yu, “Hybrid digital and analog beamforming design
L
X for large-scale antenna arrays,” IEEE J. Sel. Topics Signal Process., vol.
[rank(X?l )]2 6 M . (18) 10, no. 3, pp. 501–513, Apr. 2016.
[8] A. Hassanien and S. A. Vorobyov, “Phased-MIMO radar: A tradeoff
l=1 between phased-array and MIMO radars,” IEEE Trans. Signal Process.,
Then it is easy to prove that there is always an optimal solution vol. 58, no. 6, pp. 3137–3151, Jun. 2010.
[9] D. Wilcox and M. Sellathurai, “On MIMO radar subarrayed trans-
to the problem (P2.2) such that mit beamforming,” IEEE Trans. Signal Process., vol. 60, no. 4, pp.
K 2076–2081, Apr 2012.
X [10] F. Liu, C. Masouros, A. P. Petropulu, H. Griffiths, and L. Hanzo, “Joint
[rank(F?BB,k )]2 6 K + 1. (19) radar and communication design: Applications, State-of-the-art, and the
k=1 road ahead,” IEEE Trans. Commun., vol. 68, no. 6, pp. 3834–3862, Jun.
2020.
Because there is individual SINR constraint for each CU, the [11] F. Liu and C. Masouros, “Hybrid beamforming with sub-arrayed MIMO
optimal solution {F?BB,k }, satisfy rank(F?BB,k ) > 1, ∀k. Then radar: Enabling joint sensing and communication at mmwave band,” in
according to (19), proposition 1 can be proved. Proc. IEEE Int. Conf. Acoust. Speech Signal Process., Brighton, U.K.,
May 2019, pp. 7770–7774.
A PPENDIX B [12] Z. Cheng, B. Liao, Z. He, “Hybrid Transceiver Design for Dual-
Functional Radar-Communication System”, in Proc. 11th IEEE Sensor
P ROOF OF P ROPOSITION 2 Array Multichannel Signal Process. Workshop, Jun. 2020, pp. 8-11.
In order to account for potential infeasibility, the slack [13] Z. Cheng, Z. He and B. Liao, “Hybrid Beamforming for Multi-
carrier Dual-Function Radar-Communication System,” IEEE Trans.
variables can be added to the problem (P3.2), then we obtain Cogn. Commun. Netw., doi: 10.1109/TCCN.2021.3063110.
[14] C. Y. Chen and P. P. Vaidyanathan, “MIMO radar waveform optimization
max uH BT ⊗ Φ0 u − λr − λksk1 − λkpk1 − λkvk1


a,t,r,s,p,v with prior information of the extended target and clutter,” IEEE Trans.
  Signal Process., vol. 57, no. 9, pp. 3533–3544, Sep. 2009.
T K [15] G. Cui, H. Li, and M. Rangaswamy, “MIMO radar waveform design
FBB,k ⊗ Hk X
s.t. uH  − FTBB,j ⊗ Hk  u with constant modulus and similarity constraints,” IEEE Trans. Signal
Γk Process., vol. 62, no. 2, pp. 343–353, Jan. 2014.
j6=k
[16] O. Mehanna, K. Huang, B. Gopalakrishnan, A.Konar, and N. D.
(P3.3) > 1 − sk , sk > 0, ∀k, Sidiropoulos, “Feasible point pursuit and successive approximation of
non-convex QCQPs,” IEEE Signal Process. Lett., vol. 22, no. 7, pp.
uH BT ⊗ IN t u 6 P0 ,


804-808, Jul. 2015.
[17] Y. Huang and D. P. Palomar, “Rank-constrained separable semidefinite
uH Em u 6 1 + pm , uH Em u > 1 + vm , ∀m. programming with applications to optimal beamforming,” IEEE Trans.
(20) Signal Process., vol. 58, no. 2, pp. 664–678, 2010.
Problem (P3.3) is still a non-convex QCQP, and the SCA tech-
nique can be adopted. For any positive semi-definite matrix
C ∈ C n×n and any z, u ∈ C n×1 ,we have (u−z)H C(u−z) >
0. Expanding the above inequality, we can obtain
uH Cu > 2Re zH Cu − zH Cz

(21)
We notice that B, Φ0 , {FBB,k } and {Hk } in (20) are all
positive semi-definite matrices. Thus using the liner restriction
like (21) around the point z, the non-convex problem P(3.3)
can be replaced by convex problem P(3.4). Proposition 2 is
proved.
R EFERENCES
[1] A. Hassanien, M. G. Amin, E. Aboutanios and B. Himed, “Dual-
Function Radar Communication Systems: A Solution to the Spectrum
Congestion Problem,” IEEE Signal Process. Mag., vol. 36, no. 5, pp.
115-126, Sep. 2019
[2] F. Liu, C. Masouros, A. Li, H. Sun, and L. Hanzo, “MU-MIMO commu-
nications with MIMO radar: From co-existence to joint transmission,”
IEEE Trans. Wireless Commun., vol. 17, no. 4, pp. 2755–2770, Apr.
2018.
[3] X. Liu, T. Huang, N. Shlezinger, Y. Liu, J. Zhou, and Y. C. Eldar, “Joint
transmit beamforming for multiuser MIMO communication and MIMO
radar,” IEEE Trans. Signal Process., vol. 55, no. 3, pp. 1213–1226, Jun.
2019.
[4] L. Chen, F. Liu, J. Liu and C. Masouros, “Composite Signalling
for DFRC: Dedicated Probing Signal or Not?” arXiv preprint arXiv:
2009.03528, 2020.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI.87

Downloaded on April 05,2023 at 21:29:25 UTC from IEEE Xplore. Restrictions apply.