1

Radar and Communication Co-existence: an

Overview
Le Zheng, Member, IEEE, Marco Lops, Fellow, IEEE, Yonina C. Eldar, Fellow, IEEE, and Xiaodong Wang,
Fellow, IEEE

Abstract—Increased amounts of bandwidth are required to Research Projects Agency (DARPA) [5] and the demands of
guarantee both high-quality/high-rate wireless services (4G and sensing and communication for self-driving cars [6]. As a
5G) and reliable sensing capabilities such as automotive radar,
arXiv:1902.08676v1 [eess.SP] 22 Feb 2019

result, a number of studies have been conducted, based on

air traffic control, earth geophysical monitoring and security
applications. Therefore, co-existence between radar and com- a variety of scenarios, degrees of cooperation between the
munication systems using overlapping bandwidths has been coexisting systems, and design strategies.
a primary investigation field in recent years. Various signal The goal of this paper is to review existing results in this
processing techniques such as interference mitigation, pre-coding context and define a taxonomy of the different philosophies
or spatial separation, and waveform design allow both radar
and communications to share the spectrum. This article reviews
proposed so far. Three major architectures for co-existence
recent work on co-existence between radar and communication have been henceforth defined:
systems, including signal models, waveform design and signal (a) Co-existence in spectral overlap;
processing techniques. Our goal is to survey contributions in
this area in order to provide a primary starting point for new (b) Co-existence via cognition;
researchers interested in these problems. (c) Functional co-existence.
Index Terms—Radar/communication co-existence, spectrum Category (a) includes architectures wherein both radar and
sharing. communication systems are equipped with active transmitters
using the same frequency spectrum. Here, the major problem is
I. I NTRODUCTION to eliminate or mitigate mutual interference while guaranteeing
satisfactory performance for both functions. Different degrees
The use of radar has been widened to numerous civilian
of cooperation between the active systems have been so
applications including traffic control, remote sensing, car
far accounted for. Absolute lack of cooperation is assumed.
cruise control and collision avoidance. On a parallel track,
For example, in [7], [8] where the inherent resilience of
the quest for ever increasing rates in wireless communications
properly designed coherent Multiple-Input Multiple-Output
has pushed the carrier frequencies towards bands traditionally
(MIMO) radars to the interference is exploited and attention
assigned to radar systems. This, along with the need to limit
is paid to the performance of the radar system only. A similar
electromagnetic pollution, results in the scenario of co-existing
“radar-centric” philosophy is adopted in [9], [10], wherein co-
radar and communication systems [1], [2]. Emerging technolo-
existing communication users are safeguarded by limiting the
gies in this field rely on concepts such as passive sensing,
amount of interference produced by the radar on given sub-
waveform diversity, co-design and the so called “bio-inspired”
bandwidths. Symmetrically, uncooperative “communication-
strategies, wherein each part of a given architecture is seen as a
centric” approaches have been suggested in a number of more
sub-system whose design choices must be negotiated with the
recent studies, wherein countermeasures against the radar-
other constituent subsystems. To this last philosophy belong
induced interference are taken either at the communication
the class of cognitive systems, which are in turn intimately
receiver [11] or, in the presence of some prior information,
linked to the concept of Bayesian learning as a means to
directly at the transmitter [12], [13].
facilitate and sometimes enable individual decision-making
[1], [3], [4]. Cooperation between the active systems, possibly operating
The last few years have seen the growth of vibrant industrial in full spectral overlap, in order to negotiate the respec-
and academic interest towards the convergence of sensing tive transmit policies and adjust the corresponding detec-
and communication functions. This has been affirmed by tion/demodulation strategies is the idea underlying co-design,
the announcement of the Shared Spectrum Access for Radar first introduced in [3], and further developed in [14]–[19].
and Comm (SSPARC) program by the Defense Advanced In this approach, which we define holistic, the co-existing
systems are seen as constituent parts of a whole, so that the
Le Zheng is with Electronics & Safety, Aptiv, Agoura Hills, CA, 91301, degrees of freedom under the designer’s control are both the
e-mail: [email protected]. waveform(s) transmitted by the sensing systems and the code-
Marco Lops is with the Dipartimento di Ingegneria Elettrica e delle
Tecnologie dell’Informazione (DIETI) University of Naples “Federico II” Via book(s) employed by the communication systems. These are
Claudio 21, 80125 Naples (Italy), e-mail: [email protected]. jointly optimized so as to guarantee that both the communi-
Yonina C. Eldar is with the Department of Electrical Engineering, Technion, cation and the radar performance are satisfactory. Co-design
Israel Institute of Technology, Haifa 32000, Israel.
Xiaodong Wang is with Electrical Engineering Department, Columbia allows taking into account in the transceiver design effects
University, New York, USA, 10027, e-mail: [email protected]. such as the reverberation produced by the radar, due to clutter
2

or targets moving in close proximity to the communication II. C O - EXISTENCE IN SPECTRAL OVERLAP
receiver, range ambiguities and (random) Doppler frequencies. A. System Model
It is important to underline that these schemes are heavily
In the discussion below, we unify the Single-Input Single-
knowledge-based and rely on information exchange between
Output (SISO) and MIMO settings as they are amenable to
the constituent systems: this presupposes, on one hand, the
similar approaches. Thus, to keep the discussion as general as
presence of a “fusion center” accessible to both systems, and,
possible, we consider a scenario wherein a MIMO radar with
on the other, the accessibility of a common database, wherein
MT transmit and MR receive (typically, but not necessarily
the basic channel parameters are made available.
co-located) antennas should co-exist with a MIMO commu-
In dynamic scenarios co-design may greatly benefit from nication system equipped with NT transmit and NR receive
cognitive paradigms. Here channel state is learned through antennas, respectively, as illustrated in Fig.1.
suitable algorithms, which is conducive to the philosophy
of co-existence via channel sensing put forth in [4] and,
more generally, to category (b) of the classification above. In
fact, category (b) comprises systems wherein spectral overlap
between the communication and radar transmitters is avoided
through cognition, so that the corresponding channels are Colocated MIMO radar
interference-free. Starting from the idea, proposed in [14] and
borrowed from cognitive radio networks, of using pilot signals U CR U RC
to estimate the channels and share the channel information
between the subsystems, new approaches have been recently
proposed wherein the radar and/or the communication system H
are able to “learn” the environment without transmitting pilots
or avoiding the need for coordination [20]–[23]. In [4], for
example, the SpeCX system combines sub-Nyquist multi-band Communication TX Communication RX

sensing with sub-Nyquist radar [24] so as to enable the radar

to sense the communication channel at very low rates. Fig. 1. MIMO Communication system sharing spectrum with a MIMO radar
Category (c) comprises architectures wherein there is only system.
one active transmitter, whereby co-existence is functional, but
The MIMO radar transmits MT signals, where the signal
no interference is produced and no real resource negotiation
transmitted from the i-th transmit element is characterized by
takes place. Dual Function Radar Communication (DFRC)
a fast-time code ci = [ci (0), . . . , ci (Pr − 1)] ∈ CPr . The
systems rely on combining radar and communication transmit-
continuous-time waveform for the i-th transmit element is then
ters in the same hardware platform, which should be designed
given by
so as to guarantee the performances of both systems: the PX
r −1
information is embedded [25]–[28] in the radar signal, and a c̃i (t) = ci (p)ψ(t − pTr ). (1)
MIMO radar transmitter uses a combination of beam-forming p=0
and waveform diversity in order to direct information bits
Here ψ(·) is a Nyquist waveform of bandwidth1 B = T1r ,
towards multiple communication receivers, without affecting
i.e., such that its autocorrelation Rψ (·) satisfies the condition
the performance of the sensing function, while guaranteeing
Rψ (kTr ) = δ(k), with δ(·) denoting the Kronecker delta,
satisfactory Bit Error Rate (BER) performance. Opportunistic
and T1r is the fast-time coding rate. The product between the
sensing systems, instead, consist of a receiver co-located with
bandwidth and the “effective” duration of these coded pulses
the communication transmitter and a dedicated software chain
is typically much larger than one. Therefore, these signals
aimed at processing the received signal. The receiver can
are sometimes referred to as “sophisticated waveforms”, as
avail itself of some side information, such as timing and
opposed to conventional un-sophisticated signals whose band-
transmitted data. This architecture has been proposed and
width is on the order of the inverse of their duration.
theoretically assessed with reference to the 802.11ad format
In this architecture of Figure 1, every radiating element is
used in conjuction with a sensing system in an automotive
allowed to transmit a train of N coded pulses of the form
environment [29], [30]. Passive radar systems also can be
(1), spaced the pulse repetition time T apart, and amplitude-
thought of as belonging to category (c) since they exploit other
modulated by a slow-time code gi = [gi (0), gi (1), ..., gi (N −
transmissions (communications, broadcast, or radionavigation)
1)]T ∈ CN . Thus the i-th element transmits the signal
rather than having their own dedicated radar transmitter [1],
[31]. N
X −1
si (t) = gi (n)c̃i (t − nT ). (2)
The article structure reflects the above categorization: in
n=0
Sec. II we review systems in spectral overlap, by considering
the cases of uncoordinated and coordinated transmission. Sec. 1 Nyquist waveforms with bandwidth B = 1 are strictly band-limited,
Tr
III is devoted to cognition-based systems and Sec. IV focuses and therefore not time-limited. In practice, they are generated by truncation
of an ideal waveform, whereby the discretization may incur some degree of
on functional duality. Concluding remarks and suggestions for aliasing: however, by allowing some excess bandwidth, this effect can be kept
future investigations are presented in Sec. V. under control. A detailed discussion can be found in [32].
 3

Some special cases of the radar signal model (2) are as signal at the j-th antenna of the radar receiver (RX) can be
follows: cast in the form
1) A single-antenna transmitter using a single signal MT
X NT
X
with fast-time code c = [c(0), . . . , c(Pr − 1)]T , rj (t) = ai,j si (t − τi,j ) + (uCR
i,j ∗ xi )(t)
corresponding to N = MT = 1. i=1 i=1
MT
2) A single-antenna transmitter using an amplitude- X
modulated train of pulses, corresponding to MT = 1, + (aIi,j ∗ si )(t) + nR,j (t), (4)
Pr = 1. The train is uniquely determined by the i=1

slow-time code g = [g(0), . . . , g(N − 1)]T ∈ CN . where ai,j is the target complex backscattering coefficient,
The usual pulsed-radar corresponds to an all-one including the path loss and the phase shift due to the target
slow-time code. angle and position with respect to the transmit and receive
3) A multi-antenna transmitter wherein each antenna antennas; uCR
i,j (t) is the response of the channel from the
transmits a single sophisticated signal. As a conse- communication Transmitter (TX) to the radar RX; τi,j is the
quence, N = 1, si (t) = ci (t) and the Pr × MT delay of the target from the i-th TX to the j-th RX; aIi,j (t) is
space-time code matrix C = [c1 , . . . , cMT ] is the the response of the clutters; ∗ is the convolution operation; and
degree of freedom to be employed at the transmitter nR,j (t) denotes the noise at the j-th RX antenna. Likewise,
side [33]. the signal received at the j-th antenna of the communication
4) A multi-antenna transmitter wherein each an- RX is given by
tenna transmits a train of unsophisticated signals, NT MT
amplitude-modulated by the slow-time code. In this
X X
yj (t) = (hi,j ∗ xi )(t) + (uRC
i,j ∗ si )(t) + nC,j (t), (5)
case, Pr = 1 and the N ×MT space-time code matrix i=1 i=1
G = [g1 , . . . , gMT ] is the degree of freedom at the where hi,j (t) is the channel response from the i-th com-
transmitter side [17]. munication TX to the j-th communication RX; uRC i,j (t) is
Radars use radio waves to determine the range, angle, or the response of the interfering channel from radar TX to
velocity of objects. The operation of a typical MIMO radar communication RX; and nC,j (t) denotes the noise of the j-th
receive chain is summarized in the box of Page 4. The radar communication RX antenna.
range resolution is dictated, for a given Signal-to-Noise Ratio In (4), the transmitted signal si (t) is known and uCR
i,j (t) can
(SNR), by the transmit bandwidth, i.e., 1/Tr in (1). The be estimated via pilot training. On the other hand, xi (t) and
velocity resolution is determined by the duration of coherent aIi,j (t) are unknown at the radar RX. The radar needs to detect
integration, i.e., N T in (2). In situations 1) and 3) no Doppler the presence of the target, i.e., ai,j = 0 for H0 and ai,j 6= 0
processing is undertaken, mainly due to the fact that typical for H1 , and estimate the paramters τi,j and aIi,j (t). For the
single-pulse durations are too short to allow measuring the communication system given by (5), hi,j (t) can be estimated
Doppler shift induced by targets in moderate radial motion. via pilot training. In coordinated architectures, where the radar
In settings 2) and 4) moving objects generate steering vectors transmits pilots and communicates with the communication
and Doppler shifts up to T1 can be unambiguously measured. RX, uRC i,j and si (t) are known at the communication RX, while
Likewise, pulse trains with Pulse Repetition Time (PRT) in uncoordinated scenarios uRC i,j and si (t) are both unknown.
T generate range ambiguities whereby scatterers located at Based on the models (4) and (5), different co-existence
distances corresponding to delays which are integer multiples scenarios can be analyzed. Sec. II-B discusses a radar-centric
of T contribute to the same range cell. approach wherein a single-antenna radar transmits a single
The signal model for the communication system is simpler sophisticated signal with fast-time code, i.e. situation 1).
in that we just have to distinguish between the case of single Sec. II-C reviews some communication-centric approaches,
and multiple transmit antennas. In particular, we assume that assuming different degrees of prior knowledge as to the radar
the communication system operates on the same frequency interference (i.e., scenarios 2) and 3) ). Sec. II-D focuses on
band as the radar, occupying a fraction B L of its dedicated coordinated design of the radar waveform(s) and the commu-
bandwidth. Setting Tc = L/B, the signal radiated by the i-th nication code-books, assuming the most general scenario (i.e.,
transmit element is written as scenarios 3) and 4) ) of multiple transmit and receive antennas
∞
X for both systems, with either slow-time or fast-time coding.
xi (t) = vi (p)ψL (t − pTc ), (3)
p=−∞ B. Uncoordinated design: radar centric
where vi (p) is the data sequence to be transmitted, and ψL (·) We begin by discussing a “radar-centric” approach in which
satisfies the Nyquist criterion with respect to Tc = LTr . the radar function is considered primary, while unlicensed
The situation of full spectral overlap corresponds to L = 1. users are allowed to transmit in partial spectral overlap on
We note that there may be a multiplicity of narrow-band the same bandwidth. Following [9], [10], we assume NI
communication systems, each occupying a fraction of the radar interferers of the form (3). Their presence is acknowledged
bandwidth. by limiting the amount of interference the radar produces on
Assume that the radar and the communication receivers are the shared bandwidths. The focus is on the design of the radar
equipped with MR and NR receive antennas, respectively. The system, assumed to employ a single coded pulse according to
4

Classic collocated MIMO radar processing traditionally includes the following stages:
1) Sampling: At each radar RX 1 ≤ j ≤ MR , the signal rj (t) is projected onto the orthonormal system {ψ(t −
mTr )}P r −1
m=0 and sampled at its Nyquist rate B = Tr , creating the samples rj (m), 0 ≤ m ≤ Pr − 1.
1

2) Matched filter: The sampled signal is convolved with the transmitted radar codes ci , 1 ≤ i ≤ MT . The time
resolution attained in this step is 1/B.
3) Beamforming: The correlations between the observation vectors from the previous step and the steering vectors
corresponding to each azimuth on the grid defined by the array aperture are computed.
4) Doppler detection: The correlations between the resulting vectors and Doppler vectors, with Doppler frequencies
lying on the grid defined by the number of pulses, are computed. The Doppler resolution is 1/N T .
5) Peak detection: A heuristic detection process, in which knowledge of the number of targets, targets’ powers, clutter
location, and so on, may help in discovering targets’ positions. For example, if we know there are κ targets, then
we can choose the κ-strongest points in the map. Alternatively, constant false alarm (FA) rate detectors determine
a power threshold, above which a peak is considered to originate from a target so that a required probability of
FA is achieved.

situation 1) of the previous section, designed so as to guarantee with xk = [xk (0), xk (1), ..., xk (Pr − 1)]T the k-th
the maximum possible Signal to Interference-plus-Noise Ratio communication user occupying the bandwidth, nI =
(SINR) at the radar RX. [nI (0), nI (1), ..., nI (Pr − 1)]T ∈ CPr the clutter, and nR =
Assume that the radar RX is equipped with a single antenna [nR (0), nR (1), ..., nR (Pr − 1)]T ∈ CPr the noise term.
and the interference is dominated by the direct path between Equation (7) describes the model for the signal in the radar
the radar and the communication: the subscript j can thus RX. Next, we discuss the interference from the radar to the
be removed from the variables in (4). Thus, rj (t) becomes communication users, i.e., the second term in (5). As to the
r(t), and uCRi (t) = δ(t − τi ), with τi
RC RC
dictated by the communication users coexisting with the radar of interest, we
distance between the i-th communication TX and the radar suppose that each of them is operating over a frequency band
RX. Such a model holds for narrowband systems where the flat [f1k , f2k ], where f1k and f2k denote the lower and upper normal-
fading assumption is valid [14], and can be extended to more ized frequencies for the k-th system, respectively. Following
sophisticated situations by using different forms of channel (2) and (3) in [9], the interfering energy produced on the k-th
responses [34]. For simplicity, we assume there is only one communication user is given by cH Rk c where
target and let the target delay be τ = 0.
f2 − f1k , if m = n
( k
Plugging (2) into (4) and projecting the equation onto the Rk (m, n) = k k
ej2πf2 (m−n) −ej2πf1 (m−n) (8)
orthonormal system {ψ(t − mTr )}P r −1 , if m 6= n
m=0 leads to j2π(m−n)
2
(m, n) ∈ {1, 2, ..., Pr } .
r(m) = hr(t), ψ(t − mTr )i
PX
r −1
The covariance matrix M of the exogenous interference, i.e. of
= ha ci (p)ψ(t − pTr ), ψ(t − mTr )i the
PNsignal-independent component of the overall interference
k=1 uk xk + nR , is assumed to be known or perfectly
I
p=0
NT estimated.
The objective thus becomes to design the radar code c so as
X
+ uk hx(t − τkCR ), ψ(t − mTr )i
k=1
| {z } to maximize the SINR at the radar RX while ensuring that the
xk (m)
interference produced on the co-existing communication users
MT
X is smaller than a constrained value. Additional constraints to
+h (aIi ∗ si )(t), ψ(t − mTr )i
be enforced are an energy constraint on the radar code c,
i=1
| {z } and its “closeness” to some reference code c0 with prescribed
nI (m) correlation properties [9], [10]: the latter is also referred to as
+ hnR (t), ψ(t − mTr )i (6) a “similarity constraint”. The design then reduces to solving
the following constrained maximization problem:
| {z }
nR (m)

max SINR = a2 cH M−1 c (9)

with a the target complex backscattering coefficient, including c∈CN ×1
the path loss, and uk the coefficient of the interfering channel NR
X
for user k. Denoting r = [r(0), r(1), ..., r(Pr − 1)]T , we have s.t. ωk cH Rk c ≤ EI ,
k=1
NI
X (1 − η)ρ ≤ cH c ≤ ρ,
r = ac + uk xk + nI + nR ∈ CPr , (7)
k=1
kc − c0 k2 ≤ ,
 5

In the above equation, the terms cH Rk c represent the inter- i.e., L = 1, Tc = Tr and ψL (t) = ψ(t). The signal transmitted
ference produced onto the k communication receiver, k = by the communication system in the interval [0, Pr Tr ] is
1, 2, . . . , NR , EI the maximum interference that can be tol- assumed to have the form
erated by the coexisting communication networks, ωk ≥ 0 PX
r −1
for k = 1, 2, ..., NR are weights that can be assigned to the x(t) = v(p)ψ(t − pTr ).
coexisting wireless users based, for instance, on their distance p=0
from the radar and their tactical importance, 0 ≤ η ≤ 1 is
a design parameter which introduces some tolerance on the In the above equation, v = [v(0), . . . , v(Pr − 1)]T ∈
nominal interference level, ρ is the transmit energy of the radar. CPr is tied to a generic P -dimensional data vector b0 =
With relaxation, the optimization problem (9) can be trans- [b0 (0), . . . , b0 (P − 1)]T to be transmitted as v = Ab0 , with
formed into a convex optimization amenable to Semi-Definite A ∈ CPr ×P a suitable matrix. Relevant special cases of the
Programming (SDP), which entails polynomial computational above model are the OFDM transmission format, wherein
complexity [10]. Pr = P and A takes on the form of an Inverse Discrete
The scenario leading to problem (9) holds true only when Fourier Transform (IDFT) matrix, and a CDMA system with
the clutter is either absent or has rank one covariance matrix, P active users, wherein A contains the users’ signatures [11].
i.e. is modeled as a specular image of the transmitted signal Here, in order to keep the discussion simple, we confine our
reflected towards the receiver by a point-like scatterer. If, attention to the case of direct transmission of the constellation
conversely, more complex channel models are considered, points in full spectral overlap, so that P = Pr , b0 = b ∈ CPr ,
and the clutter covariance has rank larger than one (i.e., the A = IPr (IPr denotes the identity matrix of order Pr ).
point-like model does not carry over to reverberation), then Suppose a single antenna communication RX, and single-tap
constrained maximization of the SINR results in a fractional model for both communication and interference channels. It is
non-convex problem [16]. also assumed that the (typically high-power) radar transmitter
is not saturating the front-end of the communication receiver.
The communication signal in (5) can thus be re-written as
C. Uncoordinated design: communication centric
PX
r −1
The approach of optimizing radar waveforms, although
y(t) = h b(p)ψ(t − pTr )
theoretically well established, is not always applicable, mainly
p=0
due to the fact that governmental and military agencies are
MT PX
r −1
unwilling to make major changes in their radar deployments, X
+ um cm (p)ψ(t − pTr − τm ) + nC (t). (10)
which may impose huge costs. Thus, coexisting communica-
m=1 p=0
tion systems must be equipped with proper counter-measures
to guarantee required Quality of Service (QoS) when the radar Here a flat-fading channel is assumed for the communication
system(s) do not modify their transmission policy. Attention network where h is the channel coefficient, τm and um
is thus shifted back to the communication transceiver, which denote the (unknown) delay and complex coupling coefficient
explains the name “communication-centric” design. The ap- for the m-th radar, respectively. When um = 0, the m−th
proaches so far available in the literature focus either on the transmitter is idle. We also assume that in each frame Pr
receiver [11], when prior information on the radar signals is symbols are transmitted and that the frame sychronization
not available, or on the transmitter [13], when the structure of between the radar and communication is guaranteed, i.e., the
the radar transmitted waveform is known. communication system is made aware of the beginning of the
Assume first the scenario considered in [11], wherein a radar train pulse. This is a low-rate information, which can be
multiplicity of radars may be potentially active in full spectral shared once and for all, and regularly updated to account for
overlap with a communication system. Each radar is allowed possible timing drifts.
to transmit a sophisticated waveform, but no prior knowledge The communication RX has to accomplish jointly the two
as to the number of active systems, their distance from the tasks of interference estimation/removal and data demodu-
communication receiver or the channel gains is available. The lation. For interference removal, we need to estimate τm
scenario is thus akin to the one outlined in situation 3) of and um cm (p) so as to substract the second term from (10).
the list of Sec. II-A, wherein MT now plays the role of the Obviously, data demodulation and interference estimation are
maximum number of potentially active emitters. The antennas inherently coupled. In [35], an iterative procedure is proposed
of such a “multiple input” system are widely spaced, so that for joint data demodulation and interference estimation, and
the delays with which their signals arrive at the communication a direct demodulation function b̂(0) = Ψ({y(t)}0≤t≤Pr Tr ) is
receiver are all different and unknown. used as the initial step. In a general uncoordinated scenario, the
As to the communication signal, the scenario assumed in communication receiver may not know the exact form of the
[11] is fairly general. The transmitted symbols are assumed interfering radar signals, but only rely on a coarse information
to undergo suitable pre-coding, where the choice of the pre- of the family they belong to. A viable means to account for this
coding matrix dictates the type of system, ranging from Code- uncertainty is to assume that cm lives in a low-dimensional
Division Multiple Access (CDMA) to Orthogonal Frequency- subspace of CPr , spanned by the columns of a known Pr × K
Division Multiplexing (OFDM). In particular, suppose the matrix Φ = [φ0 , φ1 , ..., φPr −1 ]T ∈ CPr ×K with K  Pr ,
communication and radar systems have the same bandwidth, i.e., cm = Φαm for some unknown αm ∈ CK , tied to the
6

0
10
amplitudes. If the radar transmit code is a phase-only one (or
if, more realistically, the pulse complex amplitudes vary signif-
icantly only in the phase), then a narrow-band communication
system experiences an interference which is approximately a
−1
10 constant-envelope additive√ jθ(t)signal. Specifically, the interference
is (uRC
i,j ∗ si )(t) = Ie , t ∈ Ξ where θ(t) is the interfer-
SER

ence phase, assumed uniform in [0, 2π), I = |ug|2 denotes

−2
10
the average power of the radar interference, assumed known,
and Ξ designates the time intervals where the communication
Iteration 0
CS−L1
system is interfered. The communication transmitter in turn
CS−AN
Known radar delay
randomly selects the symbols to be transmitted from the set
B = {b̃1 , b̃2 , ..., b̃Q } of unit-energy and equally-likely points.
−3
10
8 10 12 14 16 18 20
SNR (dB)
Exploiting the statistical independence between these symbols
and θ(t), the optimal decoding regions can be obtained, and the
Fig. 2. Comparison of the algorithm SERs. Figure is taken from [11].
constellation B can be designed to maximize the transmission
rate and/or minimize the error rate.
corresponding minimal and maximum distances of all of the
potential radar transmitters from the receiver.
Following [11], the signal z (`) (t) = y(t) − D. Coordinated design
Pr −1 (`)
b̂ (p)ψ(t − pTr ) contains the superposition of the
P
h p=0
residual communication signal (due to demodulation errors), The major drawback of the previous approaches is that
the residual radar interference and noise. To understand they rely on a simplified scenario wherein several important
the joint interference removal and symbol demodulation phenomena are not accounted for:
algorithm proposed in [11], let us refer to the first iteration: - The radar system, especially when operating in search
PX
r −1   mode, generates reverberation from the surrounding en-
z (1) (t) = h b(p) − b̂(0) (p) ψ(t − pTr ) vironment, the so-called clutter, which impairs not only
p=0 | {z } its own performance, but also the performance of the
β (0) (p) communication system.
MT PX
X r −1 - The scattering centers generating clutter could have
+ φTm um αm ψ(t − pTr − τm ) +nC (t),(11) radial motion with respect to both the radar and the
m=1 p=0 communication receivers, thus generating Doppler shifts
that should be accounted for if slow-time coding is
| {z }
X(t)
considered.
where τm for m = 1, 2, ..., MT are the desired unknown
delays: in the above equation, the quantities φm and h are Cooperation between the active systems, possibly operating
known, while the objects of interest to be estimated are τm , in full spectral overlap, in order to negotiate the respec-
β (0) (p) and um αm . tive transmit policies and adjust the corresponding detec-
Define β (`) = [β (`) (0), β (`) (1), ..., β (`) (Pr − 1)]T . Notice tion/demodulation strategies is the idea underlying co-design,
that building up iterations may rely on two types of sparsity, in first introduced in [3], and further developed in [14], [16], [17].
that a) X(t) in (11) is a combination of at most MT compo- It is generally assumed that the radar and the communication
nents with unknown modulation um αm , and MT  Pr , and system may exchange information. The availability of large
b) kβ (`) k0 has to be as small as possible. The problem can be data-bases accurately mapping the scattering characteristics of
solved by using the recently developed mathematical theory large areas has allowed the development of cognitive systems
of continuous sparse recovery for super-resolution and in par- (see, e.g., [36], [37]): joint design of the radar waveform(s) and
ticular of Atomic-Norm (AN) minimization techniques [11]. the communication system codebook thus appears as a natural
Figure 2 illustrates the achievable results in terms of Symbol means to allow co-existence by preserving the performances
Error Rate (SER) for AN-based and CS-based methods, and of both.
allows assessing the loss due to the lack of prior knowledge Consider an NT × NR communication system co-existing
as to the delays of the radar systems. in full spectral overlap with an MT × MR MIMO radar
A fairly different scenario is the one considered in [13], with closely spaced antennas and co-located transmitter and
where it is assumed that the radar transmits a pulse train, receiver. We denote by D the space-time code matrix of the
possibly amplitude-modulated (according to the transmit for- radar: if fast-time coding is adopted, then D = C, with
mat 2) of Sec. II-A). Perfect Chanel State Information (CSI) C defined in situation 3). If, instead, slow-time coding is
concerning the attenuation and delay of the radar signal in undertaken, then D = G and situation 4) occurs. Denote by
its travel to the communication receiver is assumed. Thus the V the signal matrix of the communication system, composed
interference generated by the radar onto the communication of the NT spatial codewords emitted in successive epochs.
system is intermittent, and presents a large Peak-to-Average Specifically, V = [v(0), v(1), ..., v(Pr −1)] ∈ CNT ×Pr where
Power Ratio (PAPR), since it consists of pulses with large v(p) = [v1 (p), v2 (p), ..., vNT (p)]T is the spatial codeword
 Signal to noise ratio SNR 20 dB
Doppler angular frequency ωD 2ωc vr /c
Two way propagation delay τr 2r0 /c 40
7
Path loss α αji e−jωc τr
20

transmitted at epoch p. Projecting the received signal (4) and í4

10
0

(5) onto the orthonormal system {ψ(t − mTr )}P r −1

m=0 leads to:
Original Waveform

Power (dB)
NSP Waveform onto HBest í20
NSP Waveform onto HWorst
R = AD + UCR V + AI D + NR , (12) í40

(13)
í5
Y = HV + URC D + NC , 10

RMSE (degree)
í60

where A ∈ C MR ×MT
is the response of the target to be í80

detected; AI ∈ CMR ×MT is the response of the clutters; í6

10
NR is the noise at the radar RX; NC is the noise at the í100
í40 í30

communication RX; UCR ∈MR ×NT is the interfering channel

from communication TX to radar RX; URC ∈NR ×MT is the Fig. 5. Beampattern of
interfering channel from communication TX to radar RX; í7
10
0 5 10 15 20
used to calculate the nu

H ∈ CNR ×NT is the channel matrix from communication SNR

TX to radar RX. In (13) the MIMO communication system calculate null spac
Fig. 3. CRB on target direction estimation RMSE as a function of the SNR.
is assumed to have perfect channel state information - i.e. Fig. 3.H Cramer Rao Bound (CRB) on target direction estimation RMSEimportant
as approach
Best and HWorst channels are selected using Algorithms (1) and (2).
a function of the SNR, when UBest and UWorst (marked as HBest fuzzy and data. We sele
knowledge of V - to be periodically shared with the radar HWorst , respectively) channels are selected. Figure is taken from [18]. and take the corre
system through a dedicated channel. of arrival is given by equation (2) and (6) for the original equation. Thus, th
In (12), the purpose of the MIMO radar is to detect the radar waveform and the NSP radar waveform, respectively. parameter in the p
presence of a target (A = 0 for H0 and A 6= 0 for H1 ) and (SVD). We are Specifically, letting
interested in the estimation
U(n)error = ofΥ the angle
1 ΣΥ 2 , the right
H due to value of threshold t

estimate the matrix A which is related to the target parameters the NSP
singular of radar
vectors waveform. In Figure
corresponding 4, we compare
to vanishing singular performance of the
originalvalues
angles and estimated angles using ML estimation for different noticed from Figur
such as angle and velocity. An important additional degree of are collected in Ῡ for the formation of the projection matrix
2Using Algorithms (1) and (2) we can achieve original radar wave
(n)radar waveforms.
freedom is the space-time filter that can be applied to the radar PῩ almost
= Ῡsimilar H
2 (Ῡ2 ML )−1 ῩH
Ῡ2results for. original
2 The transmitted
waveform and radar signalchoose
the NSP is a larger and
2
signal R in (12). Let r̃ = vec(R) = [r(0)T , r(1)T , ..., r(Pr − thus the projection of c(p) onto the null Best
waveform which shows that by choosing H space, i.e.,
to project we value of threshold
1)T ]T with r(p) the (p+1)-th column of R. The filtered signal can cause minimum degradation in radar performance. Note smaller value corr
(n)
becomes that the ML estimate c̃(p) for
=P Ῡ2
c(p).waveform onto HWorst (15)
the NSP to our definitions
is much degraded from the original waveform and the NSP or decreasing the v
The waveform
precoderonto . magnitude of sidelo
(n)
r̄ = w̃T r̃, (14) P HBestinevitably introduces correlation among
In AlgorithmῩ(2),
2 we describe an approach to numerically it is desirable to sel
the signals emitted by the different transmit elements, thus
with w̃ ∈ CMR Pr ×1 . We recall here that the receive filter generating some performance loss for target direction esti-
is of fundamental importance in coherent MIMO radar, since mation. Note that the radar waveform is orthogonal to one
time filtering regulates the transmit beam-width, while space communication channel, but not to all. The MIMO radar
filtering controls the receive beam-pattern. selects the best interference channel, defined as 12
A possible criterion to exploit transmitter coordination for
a coherent MIMO radar co-existing with a communication UBest = U(imax ) , with imax = arg max dim[N (U(i) )], (16)
1≤i≤N̄
system is to force the radar waveforms D to live in the
null space of the interference channel URC via a spatial and avoids the worst channel, defined as
approach [18]. The MIMO structure indeed provides the
UWorst = U(imin ) , with imin = arg min dim[N (U(i) )]. (17)
degrees of freedom to suitably design the space-time code 1≤i≤N̄
matrix determining the probing signal. To illustrate further,
In general, in the fully cooperative scenario outlined in [15]
assume that the model of situation 3) of Sec. II-A is in force,
the radar can take a snapshot of the interference situation
and that the fast-time space-time code matrix C is to be de-
for each cluster, and broadcast it to allow proper user(s)
signed. To this end, we regroup the signals transmitted by the
assignment protocols. Users may then be assigned to less or
MIMO radar in the vectors c(p) = [c1 (p), c2 (p), ..., cMT (p)]T ,
more interfered base stations based on priority order.
encapsulating the spatial codeword transmitted for the p−th
In Figure 3, we compare the root-mean-square-error
sub-pulse. Consider the situation that N̄ communication RX’s
(RMSE) of the target direction estimation under different radar
exist, and let the interference channels of the communication
waveforms. Note that the estimation performance as the null-
RX’s are {U(1) , U(2) , ..., U(N̄ ) }. In [15], where the idea is
space projection (NSP) waveform onto UBest is closer to the
fully developed, these abstract “communication RX’s” are
performance of the original radar waveform in RMSE sense.
actually clusters of base-stations. The interference that would
Thus, by an appropriate selection of the interference channel,
be produced onto the n-th communication RX is U(n) c(p).
the degradation in the radar performance, due to the NSP of
At the MIMO radar, the channel state information can be
the waveform, can be reduced.
estimated using a blind null space learning algorithm [38].
A MIMO radar can operate without creating interference at
Our goal here is to assure zero interference to one of
any of the communication RXs if the number of radar transmit
the communication RXs with minimum degradation in the
antennas is greater than the sum of the requested degrees of
radar performance. Suppose we want no interference at the
freedom of all of the communication RXs [39]. Cooperation
n-th communication RX. The communication signal can be
between all of thehBSs and radar allows forming the interfer-
projected onto the null space of the channel U(n) . The i T
null space N (U(n) ) = {c ∈ CMT : U(n) c = 0} can ence matrix Ū = U(1)T , U(2)T , ..., U(N̄ )T ∈ CN̄ NR ×MT .
then be calculated based on Singular Value Decomposition Applying the previous strategy yields c̃(p) ∈ N (Ū). Other
8

(AM) techniques have been proposed and implemented in

[14], [16] through decompositions into sub-problems which
Radar RX Radar TX Comm TX Comm RX
are either convex or solvable through fractional programming
methods. In [16], for example, problem (19) has been reformu-
W f
Waveform Codebook
Channel Info. Design D Design R v lated for slow-time coding, explicitly accounting for Doppler
U CR , A I , A
shifts of both the target to be detected and the environmental
Channel Info. reverberation.
Received U RC , H
Signal R Radar Signal Waveform
Processing D Optimization III. C O - EXISTENCE VIA COGNITION
Fusion Center A. Environment sensing techniques
The idea of knowledge-based design is central for spectrum-
sharing systems [4], [7], [14], [17]–[19], [39]. The communi-
Fig. 4. Schematic structure of a coordinated design of radar and communi-
cation waveforms based on optimization. cation and/or the radar system undertake suitable “environment
sensing” phases in order to determine the transmit policies.
Inspired by cooperative methods in cognitive radio networks,
alternative strategies may rely on forcing the radar waveform [14] uses pilot signals to estimate the channels and feed back
to be designed according to an MMSE criterion (rather than the channel information between the subsystems, possibly
to the aforementioned Zero-Forcing). assigning to one of them a functional priority, like, e.g.,
More general approaches for the coordinated design of in [40], [41], where the radar is considered primary. These
radar and communication are based on optimization methods approaches rely on a centralized architecture, namely on a
(illustrated in Figure 4). We assume that the radar uses an strict coordination between the active players in order to allow
Pr ×MT space-time code matrix C: the extension to the slow- co-existence.
time coding can be undertaken by changing the time scale, More recently, approaches wherein the radar and/or the
considering the Doppler effect in the signal model of (12) and communication system are able to “learn” the environment
(13), and solving for the slow-time space-time matrix G [16]. without transmitting pilots or avoiding the need for coordi-
The space-time filter w̃ in (14) can also be optimized so as to nation have been proposed. These advanced approaches are
improve radar performance. discussed in the following two scenarios:
Assume that the SINR is the figure of merit of interest to 1) Environment sensing at the communication RX: A com-
the radar and let Q be the figure of merit chosen for the com- munication system shares its spectrum with an ensemble
munication system: they depend on D, on the symbol matrix of potential interferers, i.e., a set of radar/sensing systems.
V (or on some statistical feature thereof if random coding is The interfering waveforms from the radars lie in the
undertaken), as well as on a number of channel parameters tied subspace of a known dictionary, and impinge on the
to the reverberation, that we combine in an unspecified array communication RX with unknown, possibly time-varying
Z. A suitable figure of merit guaranteeing the performance of delays and coupling coefficients.
the communication system is the mutual information between 2) Environment sensing at the radar RX: A sparse target
the input symbol stream and the observations [3], [16]. In scene is assumed, allowing to reduce the radar sampling
particular, the mutual information averaged over Pr time slots, rate without sacrificing delay and Doppler resolution. The
assuming Gaussian interference is Xampling framework can be adopted, where the system
1
PX r −1 architecture is designed for sampling and processing of
log2 det INR L + R−1 H
, (18) analog inputs at rates far below Nyquist, whose under-


C= Cin HRv (p)H
Pr p=0
lying structure can be modeled as a union of subspaces
where Rv (p) = E v(p)v(p)
 H

is the covariance matrix of [24].
the communication codebook, and RCin ∈ C NR ×NR
is the The former situation has been described in Sec. II-C. The
covariance of interference plus noise, assumed either known communication RX must be made adaptive, in order to accom-
or perfectly estimated. The transceivers are designed so as to plish jointly the two tasks of interference estimation/removal
guarantee prescribed QoS to both systems. and data demodulation. For the latter situation, the SpeCX
A possible optimization problem can be formulated as: system (shown in Figure 5) was proposed in [4], and combines
sub-Nyquist multi-band sensing with sub-Nyquist radar so as
max SINR (D, {Rv (p)}, Z, w̃) ,

 to enable the radar to sense the communication channel at
 D,{Rv (p)},w̃


P s.t. Q (D, {R v (p)}, Z) ≥ Q 0 , QoS of Comm. Syst. very low rates. Compared to the other works, SpeCX presents

 g i (D) ≤ 0, i = 1, ..., I R , Rad. wav. constr. a complete solution that shows recovery of both the radar and
fi ({Rv (p)}) ≤ 0, i = 1, ..., IT , Comm. codes. constr. communication signal with minimal known information about


(19) the spectrum.
where SINR(·) is the SINR at the output of the radar receiver, More specifically, a sub-Nyquist cognitive radio is first
gi (·) and fi (·) are a set of constraints forced on the radar and implemented to sense the communication channel and de-
communication transmitted signals, respectively. The problem termine which bands are occupied. This can be done using
in (19) is typically non-convex, but Alternating Maximization the modulated wideband converter (MWC), a sub-Nyquist
 159

Fig.5.11.(a)(a)
Fig. SpeCX
SpeCX prototype.
prototype. The
The system
system consists
consists of of a signal
a signal generator,
generator, a CRo communication
a cognitive radio receiver analog receiver
based on including
the MWC, the MWC analog
a communication front-end
digital receiver,board,
and a
a communication
cognitive digitalcomm
radar. SpeCX receiver, a CR
system analogshowing
display and receiver.
(b) lowSpeCX comm system
rate samples display
acquired from showing
one MWC (b)channel
low rateatsamples
rate 120acquired
MHz, and from
(c)one MWC
digital channel at
reconstruction
ofrate
the120 MHz,
entire and (c)from
spectrum digital reconstruction
sub-Nyquist samples.of the entireradar
SpeCX spectrum
displayfrom sub-Nyquist
showing samples.
(d) coexisting SpeCX radar and
communication display showing
cognitive (d)(e)
radar, coexisting
cognitivecommunication
radar spectrum
and CR, (e)
compared withCRthe
spectrum
full-bandcompared with
radar, and (f) the full-band radar,
range-velocity and of
display (f)detected
range-velocity display
and true of detected
locations and trueFigure
of the targets. locations of the
is taken targets
from [4]. [40].

Once FC is identified, the communication receiver provides CRo system lies the proprietary modulated wideband converter
communication
a spectral map of receiver
occupied developed
bands tospecifically
the radar. Equipped for this with task spectral board
(MWC) map and [30]known radio environment
that implements a sub-Nyquist map (REM),
analog
which is capable of detecting sparse signals
the detected spectral map and known radio environment map at very low rates denoted as F
front-end receiver,r , the objective of the radar
which processes signals with Nyquistis to identifyratesan
[20]–[23].
(REM), theOnce the empty
objective of the bands
radar is in totheidentify
spectrum are identi-
an appropriate appropriate transmit frequency set that does
up to 6 GHz. The card first splits the wideband signal into not overlap with
transmit frequency set FR ⊂ F ∖ FC such that the radar’sa
fied, a cognitive radar receiver is employed which transmits the =union
M of Fc channels,
4 hardware and Fr , withand maximizes
an expansion thefactor of q = 5,
probability of
wideband
probabilitysignal that consists
of detection Pd is of several narrow
maximized. For aband fixedsignals,
proba- yielding M q = 20 virtual channels after digital expansion (see
correct detection. This probability increases with the SINR
in
bility of false alarm Pfa , the Pd increases with higherXampling
the vacant frequency bands [42]. Using the radar signal to whenfor
[93] themore
probability
details ofon false alarm is fixed.
the expansion). In eachTherefore,
channel,it
paradigm,
interferenceit and cannoise
be shown that high
ratio (SINR) [91]. resolution
Hence, the delay frequency and the signal is mixed with a periodic sequence pi (t), spectral
is proposed to maximize the SINR or minimize the which
Doppler can be performed from such a
selection process can, alternatively, choose to maximize the multiband wideband power
are in the undesired
truncated versions parts
of Gold of the spectrum.
Codes [94], This is achieved
generated on a
radar
SINRsignal by combining
or minimize methods
the spectral power of in sub-Nyquist
the undesired sampling
parts by using FPGA,
dedicated a structured sparsityfrequency
with periodic framework fp = [45].
20 MHz.Additional
and
of the spectrum. In order to find available bands withallows
compressed beamforming [22], [43], [44]. This least requirements of transmit
Next, the modulated energy
signal passesconstraints,
through an range
analog sidelobe
anti-
to detect targets
interference, with highsparsity
a structured resolution while using
framework [92] is a transmit
adopted levels, and minimum separation between
aliasing LPF. Finally, the low rate analog signal is sampledthe bands can also
signal
in [40].that consistsrequirements
Additional of several narrow
of transmit bands powerspread over a
constraints, by a NI ADC operating at fs = (q + 1)fp = 120 MHza
be imposed.© Once the optimal radar support is identified,
wide frequency regime. The advantage
range sidelobe levels, and minimum separation between the of such a system is suitable
(with waveform
intended code may be
oversampling), designed
leading to a over
total this support.
sampling rate
that the total bandwidth occupied is small
bands can also be imposed. At the receiver of this spectrum while still allowing of 480 MHz. The digital receiver is implemented on spectral
Another approach for waveform design is based on a NI©
for high resolution.
sharing radar, the This enables processing
sub-Nyquist transmissionmethod of an adaptive
of [31] notching that
PXIe-1065 minimizes
computer withtransmit
DC coupled energy in specific
ADC. Since the frequency
digital
radar
recoverssignal thethat can co-existmap
delay-Doppler withfroma standard
the subsetcommunication
of Fourier bands, rather than designing a waveform
processing is performed at the low rate of 120 MHz, very that is avoiding in-
channel
coefficients and defined
also leads FRlow
by to . rate, low power receivers. terference, while maintaining desirable envelope
low computational load is required in order to achieve real and sidelobe
This CR system leads to three main advantages. First, the characteristics
time recovery. The [46].prototype
A waveform is feddesigned
with RF to avoidcomposed
signals transmit-
CS reconstruction, performed as presented in [31] on the
B. Knowledge-based design of up to Nsig = 5 real communication transmissions, namely
ting in specific bands, a spectrally-disjoint waveform, must
transmitted fragmented bands, achieves the same resolution 10 spectral bands with total bandwidth occupancy of up not
be characterized using other metrics since interference is to
asIntraditional
this subsection,Nyquistweprocessing
survey knowledge-based
over a significantly radarsmaller
trans- driving
200 MHztheand design,
varying andsupport,
thus, nowith suchNyquist
SINR can ratebeofcalculated.
6 GHz.
mission
bandwidth. designs basedbyonconcentrating
Second, environmentall sensing. For example,
the available power Such metrics include average power levels in the undesired
The input transmissions then go through an RF combiner,
in
in some settings, the
the transmitted radarbands
narrow interference
rather than can overbe eliminated
a wide band- by frequency bands, peak sidelobe levels, and integrated sidelobe
resulting in a dynamic multiband input signal, that enables
forcing
width, the the CR
radarincreases
waveformsSNR. to Finally,
live in the this null space allows
technique of the levels.
fast carrier switching for each of the bands. This input is
interference
for a dynamic channel
form ofbetween the radarsignal
the transmitted transmitters
spectrum,and wherethe
specially designed to allow testing the system’s ability to
communication
only a small portion receiver
of [18]. This idea
the whole bandwidthis wellisstudied
used atineach the
rapidly sense IV. the Finput spectrum
UNCTIONAL - EXISTENCE
COand adapt to changes, as
cognitive
transmission, radioenabling
researchspectrum
community, and also
sharing with applied to spec-
communication
required by modern CRo and
A. Embedding data into radar waveforms shared spectrum standards, e.g.
trum sharing
signals, systems.inTypical
as illustrated approaches
Fig. 11(d). include exploiting
There, coexistence between
in the SSPARC program. The system’s effective sampling rate,
radar
the transmitted
spatial degrees bands in redgranted
of freedom and existing
by a MIMO communication
radar [7], A fairly natural evolution of radar and communication co-
equal to 480 MHz, is only 8% of the Nyquist rate. Support
bands[19],
[18], in white is shown.
[39] and adaptive transmit/receive strategies to test
recovery isisdigitally
existence to use performed
radar to perform
on the low communication,
rate samples. The also
the occupancy of the frequency bands [4]. known as Dual Function Radar Communication (DFRC) [47].
prototype successfully recovers the support of the communica-
C.InSpeCX
[4], [22] the bands selected by the radar are chosen to
Prototype Thistransmitted
tion approach is illustrated
bands, in Figure in6,Fig.
as demonstrated wherein radarOnce
11(b)-(c). and
optimize the radar
The SpeCX prototype, probability of detection.
presented in Fig. More specifically,
11, demonstrates communication systems are combined in the
the support is recovered, the signal itself can be reconstructed same hardware
after the communication signal support
radar and communication spectrum sharing. It is composed is identified, denoted platform,
from usually with
the sub-Nyquist the same
samples. This waveform or transmitter,
step is performed in real-
as
of a CRo receiver and a CR transceiver. At the heart ofmap
F c , the communication receiver provides a spectral the time, reconstructing the signal bands one sample performance
which should be designed so as to guarantee the at a time.
of occupied bands to the radar. Equipped with the detected of both systems. As echoed by the name itself, in these
10

communication symbol is embedded in the i-th pulse. Then

Comm user 1
the corresponding transmit signal vector can be expressed as
Target
p
Comm user 2 s̃i (t) = Pt 1MT ψk (t − iTr ), (22)
where Pt is the transmitting power, 1MT is the MT × 1 vector
of 1, and ψk (t) for k = 1, 2, ..., K are orthogonal waveforms.
2) Phase-modulation based information embedding [49]–
Comm user N [51]. Information is embedded by controlling the phase of
the signal. Assume that the k-th communication symbol b(k)
is transmitted through the phase information of the constant-
envelope vector v = [v(0), v(1), ..., v(Pr − 1)]T . Suppose
C l
Colocated
d MIMO radar
d the symbol v is in turn embedded in a single antenna radar
waveform, then the total transmit signal is given by
Fig. 6. Dual-function radar communications. PX
r −1

s(t) = v(p)c(p)ψ(t − pTr ), (23)

p=0
architectures co-existence is basically “functional” and no where the radar phase modulation c(p) enables direct control
spectrum overlap or resource negotiation takes place. This of the degree of range sidelobe modulation (RSM). RSM
philosophy relies on the strategy of “information embedding”. occurs due to the changing waveform structure during the
Consider a joint radar communication platform equipped with coherent processing interval (CPI) [52], by trading off bit
MT transmit antennas arranged as a uniform linear array error rate (BER) and/or data throughput. When not properly
(ULA). The radar receiver employs an array of MR receive addressed, RSM translates to residual clutter in the range-
antennas with an arbitrary linear configuration. Without loss Doppler response, and hence degraded target visibility [2],
of generality, a single-element communication receiver is [53]: receive filter design to mitigate RSM is addressed
assumed to be located in the direction θc , which is known for this type of information-embedding in [49], [53]. De-
to the transmitter. sign methods focus on the realization of a common filter
Let s(t) = [s1 (t), s2 (t), ..., sMT (t)]T ∈ CMT ×1 be the response and exploit the inherent commonality among the
baseband equivalent of the signal transmitted by a MIMO radar/communication waveforms. It is worth noting that phase
radar. Suppose a target is located at θ with delay τ . The modulation will also inevitably lead to spectrum alteration
received signal is then given by of the radar waveform, which may result in energy leakage
outside the assigned bandwidth [54].
r(t) = γar (θ)at (θ)T s(t − τ ) + nR (t), (20) 3) Sidelobe amplitude modulated-based communications
[55]–[57]. To embed the k-th communication symbol b(k) dur-
where at (θ) and ar (θ) are the steering vectors of the transmit ing the i-th pulse, the beamforming weight vector ck should be
and receive array, and γ is the coefficient accounting for associated with that symbol. The amplitude modulated-based
both target reflection and propogation loss. The radar needs method models the transmit signal during the i-th pulse as
to detect the presence of the target, i.e., γ = 0 for H0 and p
γ 6= 0 for H1 , and estimate the paramters θ and τ . Assuming s̃i (t) = Pt ck ψ(t − iTr ). (24)
a single-antenna communication receiver and considering a The design of ck is formulated as the following optimization
sophisticated single-pulse MIMO radar, the baseband signal problem [26]:
at the output of the communication receiver can be expressed
as min max G(θ) − |cH
k at (θ)| , θ ∈ Θ, (25)
ck θ

y(t) = uat (θc )T s(t) + nC (t) s.t. |cH H

k at (θ)| ≤ , θ ∈ Θ̄, ck at (θc ) = ∆k ,

where G(θ) is the desired transmit beam-pattern, Θ is the

X
= uat (θc )T s̃i (t) + nC (t), (21)
i spatial sector the radar keeps under surveillance, Θ̄ is the
sidelobe region for communication,  is a positive number
where u is the channel coefficient of the received signal of users choice for controlling the sidelobe levels, and ∆k is
encapsulating the propagation environment between the trans- the k-th sidelobe level toward the communication direction θc .
mit array and the communication receiver, and s̃i (t) is the Several other variations of the sidelobe modulating approach
transmitted radar signal in the i-th sub-pulse. are discussed in [58], [59].
The fine structure of the transmitted signal s(t) dictates the 4) Multi-waveform Amplitude Shift Keying-based informa-
information embedding method. Proposed strategies include: tion embedding [26]. This method uses multiple waveforms
1) Waveform diversity-based information embedding [48]. and two transmit beamforming weight vectors cH and cL . The
Here Nb bits of information per pulse are embeded by se- method requires Nb orthogonal waveforms to embed Nb bits
lecting the radar waveform on a pulse-to-pulse basis from per radar pulse. Then, Nb waveforms are transmitted simulta-
a set of K = 2Nb waveforms [47]. Assume that the k-th neously, where the total transmit energy Pt is divided equally
 11

among the Nb waveforms. Every transmitted waveform is of opportunity which can be exploited for short-range obstacle
used to deliver one information bit and the waveform ψk (t), detection, typically in automotive applications [29]. In such a
k = 1, 2, ..., Nb , is radiated either via cH for bi (k) = 0 or cL phase, the transmitted signal consists of a preamble, containing
for bi (k) = 1 [47]. The transmit signal is then concatenated complementary Golay codes, and a payload, con-
taining data. The proposed architectures rely on the presence
s̃i (t) = of a receiver, co-located with the wireless transmitter and
Nb
accessing some key information such as the timing, as well as
r
Pt X
((1 − bi (k))cH + bi (k)cL ) ψk (t − iTr ).(26) part if not all of the transmitted signal. With reference to (27),
Nb
k=1
τc = 0, u = 0 because there is no direct path, and x(t) is either
partially known, since the preamble has a fixed structure, or
B. Radar employing communication waveforms completely known, if the transmitted data are communicated
to the radar receiver.
Another evolution of functional co-existence is to exploit the Suppose there is one target in each sector. We denote by
waveforms transmitted by a communication network in order γ its unique complex scattering coefficient. A number of
to perform sensing (radar) functions. Without loss of general- receiving structures have been proposed for target detection
ity, we assume a single-element communication transmitter (or and localization in the range/Doppler domain in [29], [30],
a phased-array with an extremely directional beam-pattern). mostly based on Generalized Likelihood Ratio Test (GLRT)
The baseband signal at the communication TX is given by (3) [61] and assuming different degrees of prior knowledge and
with xi (t) and vi (p) replaced by x(t) and v(p), respectively. cooperation between the radar receiver and the communication
Suppose the radar is equipped with MR antennas and the transmitter:
communication TX is located at angle θc . There are a number
1) GLRT-1: Everything but the triplet (γ, ν, τ ) in (27) is
of scattering centers (targets), the i-th of which is with path
known;
delay τi , Doppler shift νi and angle θi . Let γi be the coefficient
2) GLRT-1, simpl.: The receiver is as GLRT-1, but only
accounting for both target reflection and propagation loss of
processes the preamble;
the i-th target. The response from the communication TX to
3) GLRT-2, SW-1: Like GLRT-1, but γ is a nuisance param-
the radar RX in (4) can be re-written as
eter, modeled as complex Gaussian;
4) GLRT-3: The payload data are not available to the radar
X
uCR
j (t) = uar,j (θc )δ(t − τc ) + γi ar,j (θi )ej2πνi t δ(t − τi ),
i receiver;
5) GLRT-4 SW-1: Like GLRT-3, but with γ a nuisance
where ar,j (θ) is the angle response of the j-th radar RX, u is
parameter;
the coefficient of the direct path between the communication
6) Preamble-det: The preamble detector of [29].
TX and radar RX, and τc is the delay of the direct path. As no
radar TX is used, the baseband equivalent We underline here that the GLRT strategy is aimed at solving
PMT signal at the radar composite hypotheses tests, namely wherein the densities
RX can be obtained from (4) with i=1 ai,j si (t − τi,j ) and
PMT I under the two alternatives contain unknown parameters. In
(a
i=1 i,j ∗ si )(t) removed:
practice, these parameters are replaced by the corresponding
r(t) = uar (θc )x(t − τc ) Maximum-Likelihood (ML) estimates, performed with the
+
X
γi ej2πνi t ar (θi )x(t − τi ) + nR (t), (27) same set of data used to make the final decision. Consequently,
i
the GLRT considers, as a by-product, an estimate of the
unknown parameters.
where ar (θ) = [ar,1 (θ), ar,2 (θ), ..., ar,MR (θ)]T ∈ CMR is the Figures 7 and 8 represent examples of what can be achieved
receive steering vector. with such opportunistic structures in terms of both detection
One option to use a communication waveform x(t) for and localization of an obstacle in short-range applications.
sensing is the opportunistic radar based on the 802.11ad
standard proposed in [29], [30]. The adoption of the 802.11ad SNRb [dB]18 7 0 -6 -10 -14 -17 -20 -22
1
standard for 5-th Generation (5G) wireless systems and the
exploitation of millimeter Waves (mmWaves) in the 28 and 0.8
60 GHz bandwidths [60] immediately raised interest towards
the exploitation for sensing applications of some key charac- 0.6

teristics of the proposed standard. Indeed, mmWaves suffer

Pd

0.4
from heavy atmospheric attenuation, resonance in the O2 GLRT-1
GLRT-1, simpl.
molecule, absorption by rain, and almost complete shadow- 0.2
GLRT-2, SW-I
GLRT-3
ing by obstacles, thus requiring Line-of-Sight (LOS) paths GLRT-4, SW-I
Preamble det.
between transmitter and receiver. This is in turn achievable 0
10 30 50 70 90 110 130 150 170 190
thanks to extremely directional beam-patterns and frequent r [m]

scanning procedures during which the surrounding space is

Fig. 7. Detection probability as a function of the target range and of the
swept in search of nodes willing to establish directional links. SNR per bit. The false alarm probability is set at Pf a = 10−4 . The Figure
As a consequence, the so-called Sector Level Sweep (SLS) is taken from [30].
phase of the beamforming training protocol provides signals
12

SNRb [dB]18
10 3
7 0 -6 -10 -14 -17 -20 -22 directly using the signal in the RC, detection and estimation
performance of such radar systems may improve [62], [63].
10 2 It is worth noting that passive radar operation is generally
RMSEr [m]
inferior to active radar operation due to non-optimal wave-
10 1
forms, spatial beampatterns, and transmit power [2]. Some
GLRT-1 recent works proposed “commensal radar” [64], [65], in which
10 0 GLRT-1, simpl.
GLRT-2, SW-I the communication signal is designed with the double purpose
10 -1
GLRT-3
GLRT-4, SW-I of transferring information and improving target localization
Preamble det.
Refined Preamble det.
(through a careful autocorrelation function shaping) for a co-
10 -2
10 30 50 70 90 110 130 150 170 190 existing passive sensing system.
r [m]
V. C ONCLUSIONS
Fig. 8. Ranging accuracy as a function of the target range and of the SNR
per bit. Figure is taken from [30]. We reviewed some of the main ideas and techniques to
allow coexistence of sensing and communication functions
sharing the same frequency spectrum. The strategies so far
Notwithstanding the encouraging results so far available, a proposed have been grouped into three major categories: the
number of problems still remain before claims can be made first one allows spectral overlap between the signal transmitted
on the feasibility of such structures. The channel models by the radar and communication systems, while the other
underlying the results of Figures 7 and 8, are very simple, two avoid mutual interference either by cognitively assigning
assuming that either a single object is present or that it absorbs disjoint sub-bands to the different services or allowing just
all of the radiation, thus shielding further obstacles. Moreover, one transmitter to be active and guaranteeing functional co-
since the range resolution is on the order of decimeters, most existence. For each of the above categories, the basic ideas
objects are typically range-spread, a situation not accounted are outlined, discussing advantages and disadvantages, and
for so far in the open literature. offering some examples to illustrate their performance. In
Passive radar is another option that exploits other transmis- the future, hardware prototypes should be built and deployed
sions (communications, broadcast, or radio navigation) rather to be tested on real data. This would permit assessing their
than having its own dedicated radar transmitter [1], [31]. It is performance in real world conditions, including different types
generally necessary to have a reference channel (RC) dedicated of noise, clutter and interference.
to acquiring the direct path signal as the reference waveform
for matched filtering, and surveillance channels (SCs) from R EFERENCES
which the target reflections are acquired. For communication [1] H. Griffiths, L. Cohen, S. Watts, E. Mokole, C. Baker, M. Wicks, and
TX with known position, θc in (27) can be obtained [62]. The S. Blunt, “Radar spectrum engineering and management: Technical and
regulatory issues,” Proceedings of the IEEE, vol. 103, no. 1, pp. 85–102,
signal in the RC is given by 2015.
[2] S. Blunt and E. Perrins, “Radar & communication spectrum sharing,”
zRC (t) = ar (θc )H z(t) = ux(t − τc ) + nRC (t), (28) 2018.
[3] B. Li, A. P. Petropulu, and W. Trappe, “Optimum co-design for spectrum
where nRC (t) = ar (θc )H ( i γi ar (θi )x(t − τi ) + nR (t)).
P
sharing between matrix completion based MIMO radars and a MIMO
The SC signal is obtained via beamforming on direction θ̃: communication system,” IEEE Transactions on Signal Processing,
vol. 64, no. 17, pp. 4562–4575, 2016.
zSC (t) = ar (θ̃)H z(t) [4] D. Cohen, K. V. Mishra, and Y. C. Eldar, “Spectrum sharing radar:
X Coexistence via Xampling,” IEEE Trans. Aerosp. Electron. Syst., vol. 54,
= uζc x(t − τc ) + γi ej2πνi t ζi x(t − τi ) + nSC (t), no. 3, pp. 1279–1296, 2018.
i [5] J. B. Evans, “Shared spectrum access for radar and commu-
(29) nications (SSPARC),” DARPA, Press Release.[Online]. Available:
http://www. darpa. mil/program/shared-spectrum-access-for-radar-and-
where ζc = ar (θ̃)H ar (θc ), ζi = ar (θ̃)H ar (θi ), and nSC (t) = communications, 2016.
[6] S. M. Patole, M. Torlak, D. Wang, and M. Ali, “Automotive radars:
ar (θ̃)H nR (t). To detect the target at delay τ and Doppler ν, A review of signal processing techniques,” IEEE Signal Processing
the signal is “match-filtered” via [63] Magazine, vol. 34, no. 2, pp. 22–35, 2017.
Z [7] H. Deng and B. Himed, “Interference mitigation processing for
spectrum-sharing between radar and wireless communications systems,”
r(τ ) = zSC (t)e−j2πνt zRC ∗
(t − τ + τc )dt. (30) IEEE Transactions on Aerospace and Electronic Systems, vol. 49, no. 3,
pp. 1911–1919, 2013.
The surveillance signal zSC (t) contains the signal from the [8] F. H. Sanders, R. L. Sole, J. E. Carroll, G. S. Secrest, and T. L. Allmon,
direct path, which causes strong interference. Another issue is Analysis and resolution of RF interference to radars operating in the
band 2700-2900 MHz from broadband communication transmitters. US
that the RC is not very clean in many practical cases, and the Department of Commerce, National Telecommunications and Informa-
performance of the radar is significantly degraded when there tion Administration, 2012.
is lots of interference, clutter and noise. [9] A. Aubry, A. De Maio, M. Piezzo, and A. Farina, “Radar waveform
design in a spectrally crowded environment via nonconvex quadratic
To improve the performance of passive radar, one can make optimization,” IEEE Transactions on Aerospace and Electronic Systems,
use of structural information of the underlying communica- vol. 50, no. 2, pp. 1138–1152, 2014.
tion signal. In particular, since the type of modulation is [10] A. Aubry, A. De Maio, Y. Huang, M. Piezzo, and A. Farina, “A new
radar waveform design algorithm with improved feasibility for spectral
typically known, we can first estimate the data symbols by coexistence,” IEEE Transactions on Aerospace and Electronic Systems,
demodulation. As demodulation provides better accuracy than vol. 51, no. 2, pp. 1029–1038, 2015.
 13

[11] L. Zheng, M. Lops, and X. Wang, “Adaptive interference removal [34] A. Aubry, A. DeMaio, A. Farina, and M. Wicks, “Knowledge-aided
for uncoordinated radar/communication coexistence,” IEEE Journal of (potentially cognitive) transmit signal and receive filter design in signal-
Selected Topics in Signal Processing, vol. 12, no. 1, pp. 45–60, 2018. dependent clutter,” IEEE Transactions on Aerospace and Electronic
[12] F. Liu, C. Masouros, A. Li, T. Ratnarajah, and J. Zhou, “MIMO radar and Systems, vol. 49, no. 1, pp. 93–117, 2013.
cellular coexistence: A power-efficient approach enabled by interference [35] L. Zheng, M. Lops, and X. Wang, “Adaptive Interference Removal for
exploitation,” IEEE Transactions on Signal Processing, 2018. Un-coordinated Radar/Communication Co-existence,” IEEE Journal of
[13] N. Nartasilpa, A. Salim, D. Tuninetti, and N. Devroye, “Communications Selected Topics in Signal Processing, vol. 12, no. 1, pp. 1–16, 2018.
system performance and design in the presence of radar interference,” [36] M. Piezzo, A. De Maio, A. Aubry, and A. Farina, “Cognitive radar wave-
IEEE Transactions on Communications, vol. XX, no. X, p. XXX, 2018. form design for spectral coexistence,” in 2013 IEEE Radar Conference.
[14] B. Li and A. P. Petropulu, “Joint transmit designs for coexistence of IEEE, 2013.
MIMO wireless communications and sparse sensing radars in clutter,” [37] A. Aubry, A. De Maio, M. Piezzo, M. Naghsh, M. Soltananian, and
IEEE Transactions on Aerospace and Electronic Systems, vol. 53, no. 6, S. Petre, “Cognitive radar waveform design for spectral coexistence in
pp. 2846–2864, 2017. signal-dependent interference,” in 2014 IEEE Radar Conference. IEEE,
[15] J. Mahal, A. Khawar, A. Abdelhadi, and C. Clancy, “Spectral coexistence 2014, pp. 474–478.
of MIMO radar and MIMO cellular systems,” IEEE Trans. Aerospace [38] Y. Noam and A. J. Goldsmith, “Blind null-space learning for MIMO
and Electronic Systems, vol. 53, no. 2, pp. 655–668, 2017. underlay cognitive radio with primary user interference adaptation,”
IEEE Transactions on Wireless Communications, vol. 12, no. 4, pp.
[16] J. Qian, M. Lops, L. Zheng, X. Wang, and Z. He, “Joint system design
1722–1734, 2013.
for co-existence of MIMO radar and MIMO communication,” IEEE
[39] A. Babaei, W. H. Tranter, and T. Bose, “A practical precoding ap-
Transactions on Signal Processing, 2018.
proach for radar/communications spectrum sharing,” in Cognitive Radio
[17] L. Zheng, M. Lops, X. Wang, and E. Grossi, “Joint design of overlaid
Oriented Wireless Networks (CROWNCOM), 2013 8th International
communication systems and pulsed radars,” IEEE Transactions on
Conference on. IEEE, 2013, pp. 13–18.
Signal Processing, vol. 66, no. 1, pp. 139–154, 2018.
[40] R. Zhang and Y.-C. Liang, “Exploiting multi-antennas for opportunistic
[18] A. Khawar, A. Abdel-Hadi, and T. C. Clancy, “Spectrum sharing spectrum sharing in cognitive radio networks,” IEEE Journal of selected
between S-band radar and LTE cellular system: A spatial approach,” topics in signal processing, vol. 2, no. 1, pp. 88–102, 2008.
in Dynamic Spectrum Access Networks (DYSPAN), 2014 IEEE Interna- [41] R. Zhang, Y.-C. Liang, and S. Cui, “Dynamic resource allocation in
tional Symposium on. IEEE, 2014, pp. 7–14. cognitive radio networks,” IEEE Signal Processing Magazine, vol. 27,
[19] A. Puglielli, A. Townley, G. LaCaille, V. Milovanović, P. Lu, K. Trot- no. 3, pp. 102–114, 2010.
skovsky, A. Whitcombe, N. Narevsky, G. Wright, T. Courtade et al., [42] D. Cohen, A. Dikopoltsev, R. Ifraimov, and Y. C. Eldar, “Towards sub-
“Design of energy-and cost-efficient massive MIMO arrays,” Proceed- Nyquist cognitive radar,” in Radar Conference (RadarConf), 2016 IEEE.
ings of the IEEE, vol. 104, no. 3, pp. 586–606, 2016. IEEE, 2016, pp. 1–4.
[20] M. Mishali, Y. C. Eldar, O. Dounaevsky, and E. Shoshan, “Xampling: [43] O. Bar-Ilan and Y. C. Eldar, “Sub-Nyquist radar via Doppler focusing,”
Analog to digital at sub-Nyquist rates,” IET circuits, devices & systems, IEEE Transactions on Signal Processing, vol. 62, no. 7, pp. 1796–1811,
vol. 5, no. 1, pp. 8–20, 2011. 2014.
[21] M. Mishali and Y. C. Eldar, “From theory to practice: Sub-Nyquist [44] E. Baransky, G. Itzhak, N. Wagner, I. Shmuel, E. Shoshan, and Y. Eldar,
sampling of sparse wideband analog signals,” IEEE Journal of Selected “Sub-Nyquist radar prototype: Hardware and algorithm,” IEEE Transac-
Topics in Signal Processing, vol. 4, no. 2, pp. 375–391, 2010. tions on Aerospace and Electronic Systems, vol. 50, no. 2, pp. 809–822,
[22] Y. C. Eldar, Sampling theory: Beyond bandlimited systems. Cambridge 2014.
University Press, 2015. [45] J. Huang, T. Zhang, and D. Metaxas, “Learning with structured sparsity,”
[23] D. Cohen, S. Tsiper, and Y. C. Eldar, “Analog-to-Digital cognitive radio: Journal of Machine Learning Research, vol. 12, no. Nov, pp. 3371–3412,
Sampling, detection, and hardware,” IEEE Signal Processing Magazine, 2011.
vol. 35, no. 1, pp. 137–166, 2018. [46] S. W. Frost and B. Rigling, “Sidelobe predictions for spectrally-disjoint
[24] D. Cohen and Y. C. Eldar, “Sub-nyquist radar systems: Temporal, radar waveforms,” in Radar Conference (RADAR), 2012 IEEE. IEEE,
spectral, and spatial compression,” IEEE Signal Processing Magazine, 2012, pp. 0247–0252.
vol. 35, no. 6, pp. 35–58, 2018. [47] A. Hassanien, M. G. Amin, Y. D. Zhang, and F. Ahmad, “Signaling
[25] S. D. Blunt, P. Yathan, and J. Stiles, “Intrapulse radar-embedded com- strategies for dual-function radar communications: an overview,” IEEE
munications,” IEEE Transactions on Aerospace and Electronic Systems, Transactions on Aerospace and Electronic Systems Magazine, vol. 31,
vol. 46, no. 3, pp. 1185–1200, 2010. no. 10, pp. 36–45, 2016.
[26] A. Hassanien, M. G. Amin, Y. D. Zhang, and F. Ahmad, “Dual-Function [48] S. D. Blunt, M. R. Cook, and J. Stiles, “Embedding information into
Radar-Communications: Information Embedding Using Sidelobe Con- radar emissions via waveform implementation,” in Waveform Diversity
trol and Waveform Diversity.” IEEE Trans. Sig. Proc, vol. 64, no. 8, pp. and Design Conference (WDD), 2010 International. IEEE, 2010, pp.
2168–2181, 2016. 000 195–000 199.
[49] C. Sahin, J. Jakabosky, P. M. McCormick, J. G. Metcalf, and S. D.
[27] Z. Geng, R. Xu, H. Deng, and B. Himed, “Fusion of radar sensing and
Blunt, “A novel approach for embedding communication symbols into
wireless communications by embedding communication signals into the
physical radar waveforms,” in Radar Conference (RadarConf), 2017
radar transmit waveform,” IET Radar, Sonar, Navigation, vol. 12, no. 6,
IEEE. IEEE, 2017, pp. 1498–1503.
pp. 632–630, 2018.
[50] C. Sahin, J. G. Metcalf, A. Kordik, T. Kendo, and T. Corigliano, “Ex-
[28] F. Liu, L. Zhou, C. Masouros, A. Li, W. Luo, and A. Petropulu, “To-
perimental validation of phase-attached radar/communication (PARC)
ward dual-functional radar-communication systems: Optimal waveform
waveforms: Radar performance,” 2018.
design,” IEEE Transactions on Signal Processing, vol. 66, no. 16, pp.
[51] M. J. Nowak, Z. Zhang, L. LoMonte, M. Wicks, and Z. Wu, “Mixed-
4264–4279, Aug 2018.
modulated linear frequency modulated radar-communications,” IET
[29] P. Kumari, N. Gonzalez-Prelcic, and R. W. Heath, “Investigating the Radar, Sonar & Navigation, vol. 11, no. 2, pp. 313–320, 2016.
IEEE 802.11ad standard for millimeter wave automotive radar,” in [52] J. Tian, W. Cui, and S. Wu, “A novel method for parameter estimation
Vehicular Technology Conference (VTC Fall), 2015 IEEE 82nd. IEEE, of space moving targets,” IEEE Geoscience and Remote Sensing Letters,
2015, pp. 1–5. vol. 11, no. 2, pp. 389–393, 2014.
[30] E. Grossi, M. Lops, L. Venturino, and A. Zappone, “Opportunistic radar [53] C. Sahin, J. G. Metcalf, and S. D. Blunt, “Filter design to address range
in 802.11ad networks,” IEEE Transactions on Signal Processing, vol. 66, sidelobe modulation in transmit-encoded radar-embedded communica-
no. 9, pp. 2441–2454, 2018. tions,” in Radar Conference (RadarConf), 2017 IEEE. IEEE, 2017, pp.
[31] H. Griffiths, “Passive bistatic radar and waveform diversity,” Defence 1509–1514.
Academy of the United Kingdom Shrivenham (United Kingdom), Tech. [54] X. Chen, Z. Liu, Y. Liu, and Z. Wang, “Energy leakage analysis of the
Rep., 2009. radar and communication integrated waveform,” IET Signal Processing,
[32] S. D. Blunt and E. L. Mokole, “Overview of radar waveform diversity,” vol. 12, no. 3, pp. 375–382, 2018.
IEEE Aerospace and Electronic Systems Magazine, vol. 31, no. 11, pp. [55] J. Euziere, R. Guinvarc’h, M. Lesturgie, B. Uguen, and R. Gillard,
2–42, 2016. “Dual function radar communication time-modulated array,” in Radar
[33] A. De Maio and M. Lops, “Design principles of MIMO radar detectors,” Conference (Radar), 2014 International. IEEE, 2014, pp. 1–4.
IEEE Transactions on Aerospace and Electronic Systems, vol. 43, no. 3, [56] P. M. McCormick, S. D. Blunt, and J. G. Metcalf, “Simultaneous radar
pp. 886–898, 2007. and communications emissions from a common aperture, Part I: Theory,”
14

in Radar Conference (RadarConf), 2017 IEEE. IEEE, 2017, pp. 1685– Yonina C. Eldar is a Professor in the Department of Electrical Engineering
1690. at the Technion - Israel Institute of Technology, Haifa, Israel, where she holds
[57] P. M. McCormick, B. Ravenscroft, S. D. Blunt, A. J. Duly, and the Edwards Chair in Engineering. She is also an Adjunct Professor at Duke
J. G. Metcalf, “Simultaneous radar and communication emissions from University, a Research Affiliate with the Research Laboratory of Electronics at
a common aperture, Part II: Experimentation,” in Radar Conference MIT and was a Visiting Professor at Stanford University, Stanford, CA. She is
(RadarConf), 2017 IEEE. IEEE, 2017, pp. 1697–1702. a member of the Israel Academy of Sciences and Humanities, an IEEE Fellow
[58] A. Hassanien, M. G. Amin, Y. D. Zhang, and F. Ahmad, “Efficient and a EURASIP Fellow. She has received many awards for excellence in
sidelobe ASK based dual-function radar-communications,” in Radar research and teaching, including the IEEE Signal Processing Society Technical
Sensor Technology XX, vol. 9829. International Society for Optics Achievement Award, the IEEE/AESS Fred Nathanson Memorial Radar Award,
and Photonics, 2016, p. 98290K. the IEEE Kiyo Tomiyasu Award, the Michael Bruno Memorial Award from
[59] ——, “Phase-modulation based dual-function radar-communications,” the Rothschild Foundation, the Weizmann Prize for Exact Sciences, and the
IET Radar, Sonar & Navigation, vol. 10, no. 8, pp. 1411–1421, 2016. Wolf Foundation Krill Prize for Excellence in Scientific Research. She is the
[60] T. S. Rappaport, S. Sun, R. Mayzus, H. Zhao, Y. Azar, K. Wang, G. N. Editor in Chief of Foundations and Trends in Signal Processing, and serves
Wong, G. N. Schulz, Samimi, and F. Gutierrez, “Millimeter wave mobile the IEEE on several technical and award committees.
communications for 5g cellular: It will work!” IEEE Access, vol. 1, pp.
335–349, 2013.
[61] S. Kay, “Fundamentals of statistical signal processing, volume II:
Detection theory,” 1993.
[62] L. Zheng and X. Wang, “Super-resolution delay-doppler estimation for
OFDM passive radar,” IEEE Transactions on Signal Processing, vol. 65,
no. 9, pp. 2197–2210, 2017.
[63] C. R. Berger, B. Demissie, J. Heckenbach, P. Willett, and S. Zhou,
“Signal processing for passive radar using OFDM waveforms,” IEEE
Journal of Selected Topics in Signal Processing, vol. 4, no. 1, pp. 226–
238, 2010.
[64] H. Griffiths, I. Darwazeh, and M. Inggs, “Waveform design for com-
mensal radar,” in Radar Conference (RadarCon), 2015 IEEE. IEEE,
2015, pp. 1456–1460.
[65] B. Ravenscroft, P. M. McCormick, S. D. Blunt, J. Jakabosky, and J. G.
Metcalf, “Tandem-hopped OFDM communications in spectral gaps of
FM noise radar,” in Radar Conference (RadarConf), 2017 IEEE. IEEE,
2017, pp. 1262–1267.

Xiaodong Wang (S’98-M’98-SM’04-F’08) received the Ph.D degree in

Electrical Engineering from Princeton University. He is a Professor of
Electrical Engineering at Columbia University in New York. Dr. Wang’s
research interests fall in the general areas of computing, signal processing
Le Zheng (M’17) received the B.S. degree in Communication Engineering and communications,and has published extensively in these areas. Among his
from Northwestern Polytechnical University (NWPU), Xi’an, China, in 2009 publications is a book entitled “Wireless Communication Systems: Advanced
and the Ph.D. degree in Target Detection and Recognition from the Beijing Techniques for Signal Reception”, published by Prentice Hall in 2003. His
Institute of Technology (BIT), Beijing, China, in 2015. From Feb, 2015 to current research interests include wireless communications, statistical signal
Feb, 2018, he was a Postdoctoral Researcher in the Electrical Engineering processing, and genomic signal processing. Dr. Wang received the 1999 NSF
Department of Columbia University, New York, USA. Now he works at CAREER Award, the 2001 IEEE Communications Society and Information
Aptiv (formerly Delphi), CA, USA as a Principal Radar Systems Engineer. Theory Society Joint Paper Award, and the 2011 IEEE Communication
His research interests lie in the areas of radar system, statistical signal pro- Society Award for Outstanding Paper on New Communication Topics. He has
cessing, wireless communication and high-performance hardware for signal served as an Associate Editor for the IEEE Transactions on Communications,
processing. the IEEE Transactions on Wireless Communications, the IEEE Transactions
on Signal Processing, and the IEEE Transactions on Information Theory. He
is a Fellow of the IEEE and listed as an ISI Highly-cited Author.

Marco Lops is a professor with the Department of Electrical Engineering

and Information Technologies at the University ”Federico II” of Naples. He
obtained his “Laurea” and his Ph.D. degrees from “Federico II” University
(Naples), where he was assistant (1989-1991) and associate (1991-2000)
professor. From March 2000 to November 2018 he was a professor at
University of Cassino and Southern Lazio and, in 2009-2011, he was also with
ENSEEIHT (Toulouse), first as full professor and then as visiting professor. In
fall 2008 he was a visiting professor at University of Minnesota and in spring
2009 at Columbia University. He was selected to serve as a Distinguished
Lecturer for the Signal processing Society during 2018-2020. His research
interests are in detection and estimation, with emphasis on communications
and radar signal processing.