4562 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 64, NO.

17, SEPTEMBER 1, 2016

Optimum Co-Design for Spectrum Sharing between

Matrix Completion Based MIMO Radars and a
MIMO Communication System
Bo Li, Student Member, IEEE, Athina P. Petropulu, Fellow, IEEE, and Wade Trappe, Fellow, IEEE

Abstract—Spectrum sharing enables radar and communication In this paper we study spectrum sharing between a special
systems to share the spectrum efficiently by minimizing mutual class of collocated MIMO radars and a MIMO communication
interference. Recently proposed multiple-input multiple-output system. The rationale behind considering a MIMO-type radar
radars based on sparse sensing and matrix completion (MIMO-
MC), in addition to reducing communication bandwidth and power system is the high resolution which such systems can achieve
as compared with MIMO radars, offer a significant advantage for with a relatively small number of transmit (TX) and receive (RX)
spectrum sharing. The advantage stems from the way the sampling antennas [15]–[18]. A MIMO radar system lends itself to a net-
scheme at the radar receivers modulates the interference channel worked implementation, which is very desirable in both military
from the communication system transmitters, rendering it symbol and civilian applications. A networked radar is a configuration
dependent and reducing its row space. This makes it easier for
the communication system to design its waveforms in an adaptive of TX and RX antennas. The TX antennas transmit probing
fashion so that it minimizes the interference to the radar subject waveforms, and target information is extracted by jointly pro-
to meeting rate and power constraints. Two methods are proposed. cessing the measurements of all RX antennas. This processing
First, based on the knowledge of the radar sampling scheme, the can be done at a fusion center, i.e., a network node endowed with
communication system transmit covariance matrix is designed to more computational power than the rest of the nodes. Reliable
minimize the effective interference power (EIP) at the radar re-
ceiver, while maintaining certain average capacity and transmit surveillance requires collection, communication and fusion of
power for the communication system. Second, a joint design of the vast amounts of data from various antennas. This is a power and
communication transmit covariance matrix and the MIMO-MC bandwidth consuming task, which can be especially taxing in
radar sampling scheme is proposed, which achieves even further scenarios in which the antennas are on battery operated devices
EIP reduction. and are connected to the fusion center via a wireless link. Re-
Index Terms—Collocated MIMO radar, matrix completion, cently, MIMO radars using compressive sensing (MIMO-CS)
spectrum sharing. [19]–[22], and MIMO radars via matrix completion (MIMO-
MC) [23]–[26] have been proposed to save power and band-
I. INTRODUCTION width on the link between the receivers and the fusion center,
thus facilitating the network implementation of MIMO radars.
HE operating frequency bands of communication and
T radar systems often overlap, causing one system to ex-
ert interference to the other. For example, the high UHF radar
MIMO-MC radars transmit orthogonal waveforms from their
multiple TX antennas. Each RX antenna samples the target re-
turns in a pseudo-random sub-Nyquist fashion and forwards the
systems overlap with GSM communication systems, and the samples to the fusion center, along with the seed of the ran-
S-band radar systems partially overlap with Long Term Evo- dom sampling sequence. By collecting the samples of all RX
lution (LTE) and WiMax systems [2]–[5]. Spectrum sharing antennas, and based on knowledge of each antenna’s sampling
is an emerging technology that can be applied to enable radar scheme, the fusion center constructs a matrix, refereed to as
and communication systems to share the spectrum efficiently by the data matrix (see [24] Scheme I), in which only the entries
minimizing mutual interference [4]–[14]. corresponding to sampled times contain non-zero values. Sub-
sequently, the missing entries, corresponding to non-sampled
Manuscript received October 29, 2015; revised March 11, 2016; accepted times, are provably recovered via MC techniques. In MIMO-
May 07, 2016. Date of publication May 17, 2016; date of current version July MC radars the interference is confined to the sampled entries of
22, 2016. The associate editor coordinating the review of this manuscript and
approving it for publication was Prof. Amir Asif. This work was supported by the data matrix, while after matrix completion the target echo
NSF under Grant ECCS-1408437. Parts of this work have been presented at the power is preserved. Unlike MIMO-CS, MIMO-MC does not
IEEE International Conference on Acoustics, Speech, and Signal Processing require discretization of the target space, thus does not suffer
(ICASSP) 2015.
B. Li and A. P. Petropulu are with Department of Electrical and Computer from grid mismatch issues [27].
Engineering, Rutgers, The State University of New Jersey, Piscataway NJ 08854 Spectrum sharing between a MIMO radar and a communica-
USA (e-mail: [email protected]; [email protected]). tion system has been considered in [5]–[9], where the radar inter-
W. Trappe is with the Department of Electrical and Computer Engineering,
Rutgers University, Piscataway, NJ 08854 USA, and also with WINLAB, ference is eliminated by projecting the radar waveforms onto the
Rutgers University, North Brunswick, NJ 08901 USA (e-mail: trappe@ null space of the interference channel between the MIMO radar
winlab.rutgers.edu). transmitters and the communication system. In [10], a radar re-
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. ceive filter was proposed to mitigate the interference from the
Digital Object Identifier 10.1109/TSP.2016.2569479 communication systems. However, null space projection-type

1053-587X © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
Authorized licensed use limited to:
SeeIndian Institute of Technology Hyderabad.
http://www.ieee.org/publications Downloaded on October 18,2024 at for
standards/publications/rights/index.html 04:57:41 UTC from IEEE Xplore. Restrictions apply.
more information.
LI et al.: OPTIMUM CO-DESIGN FOR SPECTRUM SHARING BETWEEN MATRIX COMPLETION BASED MIMO RADARS 4563

or spatial filtering-type techniques might miss targets aligned Σ. | · |, Tr( · ),  · ∗ and  · F denote the matrix determi-
with the interference channel. In general, the existing literature nant, trace, nuclear norm and Frobenius norm, respectively. The
on MIMO radar-communication systems spectrum sharing ad- set NL+ is defined as {1, . . . , L}. N (A) and R(A) denote the
dresses interference mitigation for either solely the communica- null and row spaces of matrix A, respectively. Ai· and A·j re-
tion system [5]–[9] or solely the radar [10]. While joint design of spectively, denote the i-th row and j-th column of matrix A.
traditional radar and communication systems for spectrum shar- [A]i,j denotes the element on the i-th row and j-th column of
ing has been considered in [4], [13], [14], co-design of MIMO matrix A. x+ is defined as max(0, x).
radar and MIMO communication systems for spectrum sharing
has not been addressed before, with the exception of our prelim-
inary results in [1], [28]. In practice, however, the two systems II. BACKGROUND ON MIMO-MC RADARS
are often aware of the existence of each other, and they could Consider a collocated MIMO radar system with Mt,R TX
share information, which could be exploited for co-design. Re- antennas and Mr,R RX antennas. The targets are in the far-field
cent developments in cognitive radios [29] and cognitive radars of the antennas and are assumed to fall in the same range bin.
[30] could provide the tools for information sharing and chan- The radar operates in two phases; in the first phase the TX
nel feedback, thus facilitating the cooperation between radar antennas transmit waveforms and the RX antennas receive
and communication systems. target returns, while in the second phase, the RX antennas
Motivated by the cooperative methods in cognitive radio net- forward their measurements to a fusion center. In each pulse,
works [31]–[33], we propose ways via which a MIMO-MC the m-th, m ∈ NM +
t,R
, antenna transmits a coded waveform
radar and a MIMO communication system, in a cooperative containing L symbols {sm (1), . . . , sm (L)} of duration TR
fashion, negotiate spectrum use in order to mitigate mutual in- each. Each RX antenna samples the target returns every TR
terference. In addition to reducing communication bandwidth seconds, i.e., samples each symbol exactly once. Following the
and power, MIMO-MC radars offer a significant advantage for model of [23]–[25], the data matrix at the fusion center can be
spectrum sharing. The advantage stems from the way the sam- formulated as
pling scheme at the radar receivers modulates the interference
channel from the communication system transmitters, rendering YR = γρDS + WR , (1)
it symbol dependent and reducing its row space. This makes it
easier for the communication system to design its waveforms where the m-th row of YR ∈ C M r , R ×L contains the L samples
in an adaptive fashion to minimize its interference to the radar forwarded by the m-th antenna; γ and ρ respectively denote
subject to meeting rate and power constraints. Two methods are the path loss corresponding to the range bin of interest, and
proposed. The first method is a cooperative design; for a fixed the radar transmit power; D ∈ C M r , R ×M t , R denotes the target
radar sampling scheme, which is known to the communication response matrix, which depends on the target reflectivity, angle
system, the communication system optimally selects its precod- of arrival and target speed (details can be found in [24]);
ing matrix to minimize the interference to the radar. The second S = [s(1), . . . , s(L)], with s(l) = [s1 (l), . . . , sM t , R (l)]T being
method is a joint design, whereby the radar sampling scheme as the l-th snapshot across the transmit antennas. The transmit
well as the communication system precoding matrix are opti- waveforms are assumed to be orthogonal, i.e., it holds that
mally selected to minimize the interference to the radar. For the SSH = I [24]; WR denotes additive noise. After matched
first method, an efficient algorithm for solving the correspond- filtering at the fusion center, target estimation can be performed
ing optimization problem is proposed based on the Lagrangian based on YR via standard array processing schemes [35].
dual decomposition (see Algorithm 1). For the second method, If the number of targets is smaller than Mr,R and L, matrix
alternating optimization is employed to solve the corresponding DS is low-rank and can be provably recovered based on a subset
optimization problem. The candidate sampling scheme needs to of its entries [24], [26]. This observation gave rise to MIMO-
be such that the resulting data matrix can be completed. Recent MC radars [23]–[26], where each RX antenna sub-samples the
work [34] showed that for matrix completion, the sampling lo- target returns and forwards the samples to the fusion center.
cations should correspond to a binary matrix with large spectral The sampling scheme could be a pseudo-random sequence of
gap. Since the spectral gap of a matrix is not affected by column integers in [1, L], with the fusion center knowing the random
and row permutations, we propose to search for the optimum seed of each RX antenna. In MIMO-MC radars, the partially
sampling matrix among matrices which are row and column filled data matrix at the fusion center can be mathematically
permutations of an initial sampling matrix with large spectral expressed as follows (see [24] Scheme I)
gap.
The paper is organized as follows. Section III introduces the Ω ◦ YR = Ω ◦ (γρDS + WR ), (2)
signal model when the MIMO-MC radar and communication
systems coexist. The problem of a MIMO communication sys- where ◦ denotes Hadamard product and Ω is a matrix containing
tem sharing the spectrum with a MIMO-MC radar system is 0’s and 1’s; the 1’s in the m-th row correspond to the sampled
studied in Section IV. Numerical results, discussions and con- symbols of the m-th RX antenna. The sub-sampling rate, p,
clusions are provided in Section V–Section VII. equals  Ω 0 /(LMr,R ). When p = 1, the Ω matrix is filled
Notation: CN (μ, Σ) denotes the circularly symmetric com- with 1’s, and the MIMO-MC radar is identical to the traditional
plex Gaussian distribution with mean μ and covariance matrix MIMO radar. At the fusion center, the completion of γρDS is

Authorized licensed use limited to: Indian Institute of Technology Hyderabad. Downloaded on October 18,2024 at 04:57:41 UTC from IEEE Xplore. Restrictions apply.
 4564 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 64, NO. 17, SEPTEMBER 1, 2016

Communication receiver :
YC = 
HX + ρG1 SΛ1 + WC , (4b)
   
signal interference noise

where
1) YR , ρ, D, S, WR , and Ω are defined in Section II.
2) X  [x(1), . . . , x(L)]; x(l) ∈ C M t , C ×1 denotes the
transmit vector by the communication TX antennas during
the l-th symbol duration. The rows of X are codewords
Fig. 1. MIMO communication system sharing spectrum with a colocated from the code-book of the communication system.
MIMO radar system. 3) WC and WR denote the additive noise; their elements
are assumed to be independent identically distributed as
formulated as the following problem [36] CN (0, σC2 ) and CN (0, σR2 ), respectively.
4) H ∈ C M r , C ×M t , C denotes the communication channel,
min  M ∗ s.t.  Ω ◦ M − Ω ◦ YR F ≤ δ, (3) where Mr,C and Mt,C denote respectively the number of
M
RX and TX antennas of the communication system; G1 ∈
where δ > 0 is a parameter related to the noise over the sam- C M r , C ×M t , R denotes the interference channel from the
pled noise matrix entries, i.e., Ω ◦ WR . On denoting by M  the radar TX antennas to the communication system RX an-
solution of (3), the recovery error  M − γρDS  is deter- tennas; G2 ∈ C M r , R ×M t , C denotes the interference chan-
F
mined by the noise power in Ω ◦ WR , i.e., the noise enters only nel from the communication TX antennas to the radar RX
through the sampled entries of the data matrix. It is important antennas. All channels are assumed to be flat fading and
to note that, assuming that the reconstruction error is small, the remain the same over L symbol intervals [5], [6], [8], [31].
 has the same received target echo power as 5) Λ1 and Λ2 are diagonal matrices. The l-th diagonal entry
reconstructed M
of Λ1 , i.e., ej α 1 l , denotes the random phase offset between
γρDS of (1).
the MIMO-MC radar carrier and the communication re-
Early studies on matrix completion theory suggested that the
ceiver reference carrier at the l-th sampling time. The l-th
low-rank matrix reconstruction requires that the entries are sam-
diagonal entry of Λ2 , i.e., ej α 2 l , denotes the random phase
pled uniformly at random. However, recent works [34] showed
offset between the communication transmitter carrier and
that non-uniform sampling would still work, as long as the sam-
the MIMO-MC radar reference carrier at the l-th sampling
pling matrix has large spectral gap (i.e., large gap between the
time. The phase offsets arise due to random phase jitter of
largest and second largest singular values).
the radar oscillator and the oscillator at the communication
receiver Phase-Locked Loops. In the literature [37]–[39],
III. SYSTEM MODEL the phase jitter of oscillator α(t) is modeled as a zero-
Consider a MIMO communication system which coexists mean Gaussian process. In this paper, we model {α1l }Ll=1
with a MIMO-MC radar system as shown in Fig. 1, sharing the as a sequence of zero-mean Gaussian random variables
same carrier frequency. The MIMO-MC radar operates in two with variance σα2 . Modern CMOS oscillators exhibit very
phases, i.e., in Phase 1 the RX antennas obtain measurements of low phase noise, e.g., −94 dB below the carrier power per
the target returns, and in Phase 2, the RX antennas forward the Hz (i.e., −94 dBc/Hz) at an offset of 2π × 1 MHz, which
obtained samples to a fusion center. The communication system yields phase jitter variance σα2 ≈ 2.5 × 10−3 [40].
interferes with the radar system during both phases. In the fol- The following assumptions are made:
lowing, we will address spectrum sharing during the first phase 1) About the synchronization of sampling times—We as-
only. The interference during the second phase can be viewed as sume that the radar receivers and the communica-
the interference between two communication systems; address- tion system sample in a time synchronous man-
ing this problem has been covered in the literature [31], [32]. ner. Although this assumption is later relaxed in
Suppose that the two systems have the same symbol rate and Section V, we next provide an example of radar and com-
are synchronized in terms of sampling times (see Section V for munication parameter settings suggesting that the afore-
the mismatched case). We do not assume perfect carrier phase mentioned assumption is applicable in real world systems.
synchronization between the two systems. The data matrix at the The typical range resolution for an S-band search and ac-
radar fusion center, and the received matrix at the communica- quisition radar is between 100 m and 600 m [41], [42].
tion RX antennas during L symbol durations can be respectively Thus, for range resolution of cTb /2 = 300 m, where c =
expressed as 3 × 108 m/s denotes the speed of light, the radar sub-pulse
duration is Tb = 2 μs. In order to have identical symbol
Radar fusion center : rate for two systems, the communication symbol duration
should be 2 μs, which corresponds to signal bandwidth of
Ω ◦ YR = Ω ◦ (γρDS) + Ω ◦ (G2 XΛ2 ) + Ω ◦ WR , (4a)
         0.5 MHz. This symbol interval value falls in the typical
signal interference noise
range of symbol interval values in LTE systems [43].

Authorized licensed use limited to: Indian Institute of Technology Hyderabad. Downloaded on October 18,2024 at 04:57:41 UTC from IEEE Xplore. Restrictions apply.
LI et al.: OPTIMUM CO-DESIGN FOR SPECTRUM SHARING BETWEEN MATRIX COMPLETION BASED MIMO RADARS 4565

2) About channel fading- We assume that H, G1 and G2 communication system transmission, i.e., Ω ◦ (G2 XΛ2 ) will
are flat fading, which is valid when the channel coherence degrade the completion of the data matrix and as a result
bandwidth is larger than the signal bandwidth [44]–[46], the target detection/estimation. The application of traditional
i.e., when the transmitted signals are narrowband. Con- spatial filtering on Ω ◦ YR for eliminating the communication
sider the symbol interval value 2 μs and signal bandwidth system interference is not straightforward and to the best of our
0.5 MHz given above. In a LTE macro-cell, the coher- knowledge has not been previously addressed. For the case with
ence bandwidth is in the order of 1 MHz [43], [47]. The complete samples, the optimal detector to maximize the SINR
typical values of LTE channel coherence bandwidth are is matched filtering following a whitening filter. However,
much larger than the signal bandwidth of 0.5 MHz, thus in the case of partially sampled YR , i.e., Ω ◦ YR , S cannot
making the flat fading channel assumption valid. Since be fully matched due to the sub-sampling operator. Also, the
the radar and communication systems use the same car- interference plus noise at the whitening filter output would not
rier and signal bandwidth, the flat fading assumption is be white anymore. Further, note that the recovery of γρDS
valid for all H, G1 and G2 . In the radar-communication via matrix completion in (3) is based on (Ω ◦ YR ). Even if we
system coexistence literature [5]–[9], the flat fading as- somehow find the spatial filter W that maximizes the SINR,
sumption is quite typical. If the narrowband assumption the filter output W(Ω ◦ YR ) cannot be used by the matrix
is not valid then perhaps one could consider an OFDM completion formulation in (3), which follows the formulation in
scenario, where the flat fading model would apply on the MIMO-MC radar [23]–[26] and general matrix completion
each carrier [13], [14]. literature [36]. The extension of the matrix completion working
3) About channel information feedback—The channels H with the additional filtering matrix is out the scope of this paper.
and G2 are also assumed to be perfectly known at the Of course, one could apply filtering on the recovered data
communication TX antennas. In practice, such channel matrix DS as post processing. However, such post-filtering
information can be obtained at the radar RX antennas would first need the matrix completion to be successful.
through the transmission of pilot signals [5], [48]. View- The approach that we propose here for addressing the radar
ing the radar system as the primary user of a cognitive ra- and comm systems interference is a design for the commu-
dio system and the MIMO communication system as the nication TX signals, or a co-design of the communication TX
secondary user, techniques similar to those of [31]–[33], signals and the radar sub-sampling scheme, so that we minimize
[49] can be used to estimate and feed back the channel the interference at the radar RX antennas for successful matrix
information between the primary and secondary systems. completion, while satisfying certain communication system rate
Let Sobs  ρG1 S be the radar interference as viewed by the requirements.
communication system. This can be obtained during times that
the communication system does not transmit. Since the radar
IV. SPECTRUM SHARING BETWEEN MIMO-MC RADAR AND A
transmission power ρ is very high, ρG1 S can be estimated with
MIMO COMMUNICATION SYSTEM
high accuracy. Based on Sobs , the communication receivers
can eliminate some of the interference via direct subtraction. First, let us provide expressions for the communication TX
However, due to the high power of the radar [3] and the unknown power and channel capacity, and the interference power at the
phase offset, there will always be residual interference, i.e., MIMO-MC radar receiver. The total transmit power of the com-
munication TX antennas equals
ρG1 S (Λ1 − I) ≈ ρG1 SΛα ≡ Sobs Λα , (5)

where Λα = diag(jα11 , . . . , jα1L ), and the approximation is  
L 
L
H H
E Tr XX = E Tr x (l) x (l) = Tr (Rx l ) ,
based on the fact that {α1l }Ll=1 are small. In the above, we l= 1 l= 1
assume that the radar waveforms have not changed between the
time the interference is estimated and used in (5). The signal where Rxl  E{x(l)xH (l)}.
at the communication receiver after interference cancellation Due to the sampling performed at the MIMO-MC radar re-
equals ceiver, the effective interference power (EIP) at the radar RX
 C = HX + Sobs Λα + WC .
Y (6) nodes can be expressed as:
  
This residual interference is not circularly symmetric, and EIP  E Tr Ω ◦ (G2 XΛ2 ) (Ω ◦ (G2 XΛ2 ))H
thus the communication channel capacity is achieved by

non-circularly symmetric Gaussian codewords, whose covari- = E Tr [G21 x (1) . . . G2L x (L)]Λ2 ΛH 2
ance and complementary covariance matrix would need to be 
H
designed simultaneously [50]. Here, we consider circularly × [G21 x (1) . . . G2L x (L)]
symmetric complex Gaussian codewords x(l) ∼ CN (0, Rxl ), 
which achieve a lower bound of the channel capacity [50], [51]. L
H H
This reduces the complexity of the design since we only need = E Tr G2l x (l) x (l) G2l ]
to design the transmit covariance matrix Rxl . l=1

Unless special measures are taken, the interference from 

L 
L
the radar transmissions, i.e., Sobs Λα , will reduce the com- = Tr G2l Rxl GH
2l = Tr Δl G2 Rxl GH
2 , (7)
munication system capacity, and the interference from the l=1 l=1

Authorized licensed use limited to: Indian Institute of Technology Hyderabad. Downloaded on October 18,2024 at 04:57:41 UTC from IEEE Xplore. Restrictions apply.
 4566 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 64, NO. 17, SEPTEMBER 1, 2016

where G2l  Δl G2 , with Δl = diag(Ω·l ). We note that the A. Cooperative Spectrum Sharing
EIP at sampling time l contains the interference corresponding In the cooperative approach, the MIMO-MC radar shares
only to 1’s in Ω·l . Thus, the effective interference channel dur- its sampling scheme Ω with the communication system. The
ing the l-th symbol duration is G2l . In the following, the EIP is spectrum sharing problem can be formulated as
used as the figure of merit for MIMO-MC radars as it affects the
performance of matrix completion and further target estimation 
L

(see simulation results in Section VI.A). Before matrix comple- (P1 ) min EIP ({Rxl }) s.t. Tr (Rxl ) ≤ Pt (9a)
{R x l } 0
l=1
tion and any target estimation, the EIP should be minimized.
From another perspective, the EIP is a reasonable surrogate of Cavg ({Rxl }) ≥ C, (9b)
the radar SINR, which is widely used as figure of merit in the
literature [52], [53], as in this paper we do not assume any prior where the constraint of (9a) restricts the total transmit power at
information on target parameters. the communication TX antennas to be no larger than Pt . The
In the coexistence model of (4a) and (6), both the effective in- constraint of (9b) restricts the communication average capacity
terference channel G2l , and the interference covariance matrix during L symbol durations to be at least C, in order to provide
at the communication receiver, i.e., Rintl  σα2 Sobs (l)SH reliable communication and avoid service outage. {Rxl } 0
obs (l),
vary between sampling times. Thus, the optimum scheme for the imposes the positive semi-definiteness on the solution.
communication transmitter would be adaptive/dynamic trans- Problem (P1 ) is convex and involves multiple matrix vari-
mission. A symbol dependent covariance matrix, i.e., Rxl , ables, the joint optimization with respect to which requires high
would need to be designed for each symbol duration in order to computational complexity. Fortunately, we observe that both the
match the variation of G2l and Rintl . objective and constraints are separable functions of {Rxl }. An
The channel G2l can be equivalently viewed as a fast fad- efficient algorithm for solving the above problem can be im-
ing channel with perfect channel state information at both the plemented based on the Lagrangian dual decomposition [55] as
transmitter and receiver [45], [54]. Similar to the definition of follows.
ergodic capacity [54], the achieved capacity is the average over 1) An Efficient Algorithm Based on Dual Decomposition:
L symbols, i.e., The Lagrangian of (P1 ) can be written as
L ({Rxl } , λ1 , λ2 ) = EIP ({Rxl }) + λ2 (C − Cavg ({Rxl }))

1  
L
Cavg ({Rxl })  log2 I + R−1 H L
w l HRxl H , (8) + λ1 Tr (Rxl ) − Pt ,
L
l=1
l=1

where λ1 ≥ 0 is the dual variable associated with the transmit

where {Rxl } denotes the set of all Rxl ’s and Rw l  Rintl + power constraint, and λ2 ≥ 0 is the dual variable associated with
σC2 I for all l ∈ NL+ . the average capacity constraint. The dual problem of (P1 ) is
The adaptive transmission could be implemented using the
V-BLAST transmitter architecture ([45], Chapter 7), where the (P1 − D) max g (λ1 , λ2 ) ,
1/2 λ1 ,λ2 ≥0
precoding matrix for symbol index l is set to Rxl . This idea
is also used in the transceiver architecture for achieving the where g(λ1 , λ2 ) is the dual function defined as
capacity of a fast fading MIMO channel with full channel state
g (λ1 , λ2 ) = inf L ({Rxl } , λ1 , λ2 ) .
information ([45], Chapter 8.2.3), and is also discussed in ([46], {R x l } 0
Chapter 9). The adaptive transmission in response to highly
mobile, fast fading channels requires the transmitter to vary the The domain of the dual function, i.e., dom g, is λ1 , λ2 ≥ 0 such
rate, power and even the coding strategy. The main bottleneck that g(λ1 , λ2 ) > −∞. It is also called dual feasible if (λ1 , λ2 ) ∈
of the system is not due to the complexity of designing and dom g. It is interesting to note that g(λ1 , λ2 ) can be obtained
implementing the variable transmission parameters, but rather by solving L independent subproblems, each of which can be
due to the feedback delay of the fast fading channel. In our paper, written as follows
the latter issue is not relevant because the channel variations are (P1 -sub) min Tr 2 Δl G2 + λ1 I Rxl
GH
introduced by the MC technique and radar waveforms, which Rx l 0
 
are available at the communication transmitter. − λ2 log2 I + R−1
w l HRxl H
H
. (10)
In this section, spectrum sharing between the communica-
tion system and the MIMO-MC radar is achieved by min- Before giving the solution of (P1 -sub), let us first state some
imizing the interference power at the MIMO-MC radar RX observations.
node, while satisfying the communication rate and TX power Observation 1: The average capacity constraint should be
constraints of the communication system. The design vari- active at the optimal point. This means that the achieved capacity
ables are the communication TX covariance matrices and/or is always C and λ2 > 0. To show this, let us assume that the
the radar sub-sampling scheme. In the following we will con- optimal point {R∗xl } achieves Cavg ({R∗xl }) > C. Then we can
sider two approaches, namely, a cooperative and a joint design always shrink {R∗xl } until the average capacity reduces to C,
approach. which would also reduce the objective. Thus, we end up with a

Authorized licensed use limited to: Indian Institute of Technology Hyderabad. Downloaded on October 18,2024 at 04:57:41 UTC from IEEE Xplore. Restrictions apply.
LI et al.: OPTIMUM CO-DESIGN FOR SPECTRUM SHARING BETWEEN MATRIX COMPLETION BASED MIMO RADARS 4567

Algorithm 1: Cooperative Spectrum Sharing (P1 ). channel, i.e., {H l }L . This justifies the definition of average
l=1

1: Input: H, G1 , G2 , Ω, Pt , C, σC2 capacity in (8). Lemma 1 shows that the communication trans-
2: Initialization: λ1 ≥ 0, λ2 ≥ 0 mitter will allocate more power to directions determined by the
3: repeat left singular vectors of H corresponding to larger eigenvalues
4: Calculate R∗xl (λ1 , λ2 ) according to (11) with the and by the eigenvectors of Φl corresponding to smaller eigen-
given λ1 and λ2 ; values. In other words, the communication will transmit more
5: Compute the subgradient of g(λ1 , λ2 ) as power in directions that convey larger signal at the commu-
L ∗ nication receivers and smaller interferences to the MIMO-MC
l=1 Tr(Rxl (λ1 , λ2 )) − Pt and radars.
C − Cavg ({R∗xl (λ1 , λ2 )}) respectively for λ1
2) Spectrum Sharing Without Knowledge of the Radar’s
and λ2 ;
Sampling Scheme: If the MIMO-MC radar does not share Ω
6: Update λ1 and λ2 accordingly based on the ellipsoid
with the communication system, the expression of EIP of (7) is
method [56];
also not available at the communication system. In this case, the
7: until λ1 and λ2 converge to a prescribed accuracy.
communication system can design its covariance assuming that
8: Output: R∗xl = R∗xl (λ1 , λ2 )
Ω is full of 1’s, i.e., for the worst case of interference


L
contradiction. By complementary slackness, the corresponding (P0 ) min Tr G2 Rxl GH
2
{R x l } 0
dual variable is positive, i.e., λ2 > 0. l=1
Observation 2: (GH 2 Δl G2 + λ1 I) is positive definite for all 
L
l ∈ NL+ . This can be shown by contradiction. Suppose that there s.t. Tr (Rxl ) ≤ Pt , Cavg ({Rxl }) ≥ C. (13)
exists l such that GH 2 Δl G2 + λ1 I is singular. Then it must hold l=1
that GH 2 Δ G
l 2 is singular and λ1 = 0. Therefore, we can always
find a nonzero vector v lying in the null space of GH The same design would hold for the case in which a traditional
2 Δ l G2 .
−1/2 MIMO radar is used instead of a MIMO-MC radar. Problem
At the same time, it holds that Rw l Hv = 0 with very high
(P0 ) is also convex and has exactly the same constraints as
probability, because H is a realization of the random chan-
(P1 ). The efficient algorithm based on the dual decomposition
nel. If we choose Rxl = αvvH and α → ∞, the Lagrangian
technique in Algorithm 1 could also be applied to solve (P0 ).
L({Rxl }, 0, λ2 ) will be unbounded from below, which indi-
The following theorem compares the minimum EIP achieved
cates that λ1 = 0 is not dual feasible. This means that λ1 is
by the cooperative approaches of (P0 ) and (P1 ) under the same
strictly larger than 0 if GH 2 Δl G2 is singular for any l. Thus,
communication constraints.
the claim is proven.
Theorem 1: For any Pt and C, the EIP achieved by the co-
Based on the above observations, we have the following
operative approaches of (P1 ) is less or equal than that achieved
lemma.
by the approach of (P0 ).
Lemma 1 ([32], [33]): For given feasible dual variables
Proof: Let {R∗0 ∗1
xl } and {Rxl } denote the solution of (P0 ) and
λ1 , λ2 ≥ 0, the optimal solution of (P1 -sub) is given by
(P1 ), respectively. We know that {R∗0 xl } satisfies the constraints
R∗xl (λ1 , λ2 ) = Φl
−1/2
Ul Σl UH
−1/2 in (P1 ), which means that {R∗0 } is a feasible point of (P1 ).
l Φl , (11) xl
The optimal {R∗1 xl } achieves an objective value no larger than
where Φl  GH 2 Δl G2 + λ1 I; Ul is the right singular matrix any feasible point, including {R∗0xl }. It holds that EIP({R∗1xl }) ≤
 −1/2
of Hl  Rw l HΦl
−1/2
; Σl = diag(βl1 , . . . , βlr ) with βli = EIP({R∗0 xl }), which proves the claim. 
+ There are certain scenarios in which (P1 ) outperforms (P0 )
(λ2 − 1/σli2 ) , r and σli , i = 1, . . . , r, respectively being the
rank and the positive singular vales of H  l . It also holds that significantly in terms of EIP. Let us denote by φ1 the intersec-
1/2
tion of null space N (G2l ) and range space R(Rw l H), and
   r
by φ2 the intersection of null space N (G2 ) and range space
log2 I + R−1 H
∗ +
wl HRxl H = log λ2 σli2 . (12) 1/2
i=1
R(Rw l H). It holds that φ2 ⊆ φ1 . Consider the case where
φ1 is nonempty while φ2 is empty. This happens with high
Based on Lemma 1, the solution of (P1 ) can be obtained probability when Mr,R ≥ Mt,C but pMr,R is much smaller
by finding the optimal dual variables λ∗1 , λ∗2 . The cooperative than Mt,C . Problem (P1 ) will guide the communication sys-
spectrum sharing problem (P1 ) can be solved via the procedure tem to focus its transmission power along the directions in φ1
outlined in Algorithm 1. The convergence of Algorithm 1 is to satisfy both communication system constraints, while intro-
guaranteed by the convergence of the ellipsoid method [56]. ducing zero EIP to the radar system. On the other hand, since
Based on Lemma 1, the coexistence model can be equivalently φ2 is empty, Problem (P0 ) will guide the communication sys-
viewed as a fast fading MIMO channel H  l . The covariance of
tem transmit power along directions that introduce nonzero EIP.
  1/2
the waveforms transmitted on Hl is Rxl  Φl Rxl Φl . It
1/2
In other words, the sub-sampling procedure in the MIMO-MC
is well-known that the optimum R  xl equals Ul Σl UH with radar essentially modulates the interference channel from the
l
power allocation obtained by the water-filling algorithm [54]. communication transmitter to the radar receiver by multiplying
The achieved capacity is the average over all realization of the {Δl }. Compared to the original interference channel G2 , the

Authorized licensed use limited to: Indian Institute of Technology Hyderabad. Downloaded on October 18,2024 at 04:57:41 UTC from IEEE Xplore. Restrictions apply.
 4568 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 64, NO. 17, SEPTEMBER 1, 2016

dimension of the row space of modulated channel G2l may be it is a uniformly random sampling matrix, or it has large spectral
greatly reduced. The cooperative approach allows the communi- gap. However, it is difficult to incorporate such conditions in the
cation system to optimally design the communication precoding above optimization problem.
matrices with respect to the effective interference channel G2l . Noticing that row and column permutation of the sampling
Therefore, it is expected that the cooperative approach based on matrix would not affect its singular values and thus the spectral
the knowledge of Ω, i.e., (P1 ), introduces smaller EIP than its gap, we propose to optimize the sampling scheme by permuting
counterpart approach without knowledge of Ω, i.e., (P0 ), does the rows and columns of an initial sampling matrix Ω0 , i.e.,
under the same the communication constraints.
Ωn = arg min Tr ΩT Qn s.t. Ω ∈ ℘ Ω0 , (16)
Ω
B. Joint Communication and Radar System Design for
Spectrum Sharing where ℘(Ω0 ) denotes the set of matrices obtained by arbitrary
In the above described spectrum sharing strategies, the row and/or column permutations. The Ω0 is generated with bi-
MIMO-MC radar operates with a predetermined pseudo ran- nary entries and pLMr,R  ones. One good candidate for Ω0
dom sampling scheme. In this section, we consider a joint design would be a uniformly random sampling matrix, as such matrix
of the communication system transmit covariance matrices and exhibit large spectral gap with high probability [34]. Brute-force
the MIMO-MC radar random sampling scheme, i.e., Ω. The search can be used to find the optimal Ω. However, the complex-
candidate sampling scheme needs to ensure that the resulting ity is very high since |℘(Ω0 )| = Θ(Mr,R !L!). By alternately
data matrix can be completed. This means that Ω is either a optimizing w.r.t. row permutation and column permutation on
uniformly random sub-sampling matrix [36], or a matrix with a Ω0 , we can solve (16) using a sequence of linear assignment
large spectral gap [34]. problems [57].
 To optimize w.r.t. column permutation, we need to find the
Recall that EIP = Ll=1 Tr(Δl G2 Rxl GH 2 ). The joint design
scheme is formulated as best one-to-one match between the columns of Ω0 and the
columns of Qn . We construct a cost matrix Cc ∈ RL ×L with

L T
[Cc ]m l  (Ω0·m ) Qn·l . The problem turns out to be a linear as-
(P2 ) min Tr Δl G2 Rxl GH
2
{R x l } 0,Ω
l=1
signment problem with cost matrix Cc , which can be solved
in polynomial time using the Hungarian algorithm [57]. Let

L
Ωc denote the column-permutated sampling matrix after the
s.t. Tr (Rxl ) ≤ Pt , Cavg ({Rxl }) ≥ C,
above step. Then, we permute the rows of Ωc to optimally
l=1
match the rows of Qn . Similarly, we construct a cost ma-
Δl = diag (Ω·l ), Ω is proper. trix Cr ∈ RM r , R ×M r , R with [Cr ]m l  Ωcm · (Qnl· )T . Again, the
Hungarian algorithm can be used to solve the row assignment
The above problem is not convex. A solution can be obtained
problem. The above column and row permutation steps are alter-
via alternating optimization. Let ({Rnxl }, Ωn ) be the variables
nately repeated until Tr(ΩT Qn ) becomes smaller than a certain
at the n-th iteration. We alternatively solve the following two
predefined threshold δ1 .
problems:
The complete joint-design spectrum sharing algorithm is

L summarized in Algorithm 2. The proposed algorithm stops when
{Rnxl } = arg min Tr Δnl −1 G2 Rxl GH
2 , the value of EIP changes between two iterations drops below a
{R x l } 0 l=1
certain threshold δ2 . It is easy to show that the objective function,

L i.e., EIP, is nonincreasing during the alternating iterations be-
s.t. Tr (Rxl ) ≤ Pt , Cavg ({Rxl }) ≥ C, (14a) tween (14a) and (14b), and is lower bounded by zero. According
l=1 to the monotone convergence theorem [58], the alternating op-
timization is guaranteed to converge. The proposed joint-design

L
Ωn = arg min Tr Δl G2 Rnxl GH spectrum sharing strategy is expected to further reduce the EIP
2 ,
Ω
l=1 at the MIMO-MC radar RX node compared to the cooperative
methods in Section IV.A. However, (P2 ) has higher computa-
s.t. Δl = diag (Ω·l ) , Ω is proper. (14b) tional complexity than (P1 ) and (P0 ) (see detailed complexity
The problem of (14a) is convex and can be solved efficiently. By analysis in Section IV.C). (P1 ) and (P0 ) could be preferable in
simple algebraic manipulation, the EIP can be reformulated as cases of limited computing resources.
EIP = Tr(ΩT Q), where the l-th column of Q contains the di-
agonal entries of G2 Rxl GH2 . Based on the above reformulation C. Complexity
of EIP, we can rewrite (14b) as
The adaptive communication transmission in the proposed
Ωn = arg min Tr ΩT Qn s.t. Ω is proper, (15) spectrum sharing methods involves high complexity. A natu-
Ω
ral question would be how much would one lose by using a
where the l-th column of Qn contains the diagonal entries of sub-optimal transmission approach of constant rate, i.e., Rxl =
G2 Rnxl GH
2 . Therefore, the EIP can be reduced by carefully · · · = RxL ≡ Rx , which has lower implementation complex-
choosing Ω. Recall that the sampling matrix Ω is proper either if ity. In such case, the spectrum sharing problem (P1 ) can be

Authorized licensed use limited to: Indian Institute of Technology Hyderabad. Downloaded on October 18,2024 at 04:57:41 UTC from IEEE Xplore. Restrictions apply.
LI et al.: OPTIMUM CO-DESIGN FOR SPECTRUM SHARING BETWEEN MATRIX COMPLETION BASED MIMO RADARS 4569

To solve (P2 ), several iterations of solving problems in (14a)

Algorithm 2: Joint Design Based Spectrum Sharing
and (14b) are required. The computational complexity of (14a) is
Between MIMO-MC Radar and a MIMO Communication
identical to that of (P1 ), which has been considered previously.
System.
Problem (14b) is in turn solved via several iterations of linear
1: Input: H, G1 , G2 , Pt , C, σC2 , δ1 , δ2 assignment problem, whose complexity cubically scales with
2: Initialization: Ω0 is a uniformly random sampling L. Simulations show that the numbers of both inner and outer
matrix iterations in Algorithm 2 are relative small. In summary, the
3: repeat computational complexity of (P2 ) is the sum of L times of a
4: {Rnxl } ← Solve problem (14a) using Algorithm 1 polynomial of Mt,C 2
and O(L3 ).
while fixing Ωn −1
5: Ωpr ev ← Ωn −1 V. MISMATCHED SYSTEMS
6: loop
7: Ωc ← Find the best column permutation of Ωpr ev In Section III, the waveform symbol duration of the radar
by solving the linear assignment problem with system is assumed to match that of the communication system.
cost matrix Cc In this section, we consider the mismatched case, and show
8: Ωr ← Find the best row permutation of Ωc by that the proposed techniques presented in the previous sections
solving the linear assignment problem with cost can still be applied. Let fsR = 1/TR and fsC denote the radar
matrix Cr waveform symbol rate and the communication symbol rate, re-
9: if |Tr((Ωr )T Qn ) − Tr((Ωpr ev )T Qn )| < δ1 spectively. Also, let the length of radar waveforms be denoted
then by LR . The number of communication symbols transmitted in
10: Break the duration of LR /fsR is LC  LR fsC fsR . The communica-
11: end if tion average capacity and transmit power can be expressed in
12: Ωpr ev ← Ωr terms of {Rxl }Ll=1C
as in Section IV. In the following, we will
13: end loop only focus on the effective interference to the MIMO-MC radar
14: Ωn ← Ωr ; n ← n + 1 receiver.
15: until |EIPn − EIPn −1 | < δ2 If fsR < fsC , the interference arrived at the radar receiver
16: Output: {Rxl } = {Rnxl }, Ω = Ωn will be down-sampled. Let I1 ⊂ NL+C be the set of indices of
communication symbols that are sampled by the radar in as-
cending order. It holds that |I1 | = LR . Following the derivation
reformulated as in previous sections, we have the following interference power
expression:
(P1 ) min EIP (Rx ) s.t. LTr (Rx ) ≤ Pt , Cavg (Rx ) ≥ C, 
Rx 0 EIP = Tr Δl  G2 Rxl GH 2 ,
(17) l∈I1
where EIP(Rx )  Tr(ΔG2 Rx GH 2 ) and Δ is diagonal and
with each entry equal to the sum of the entries in the corre- where l ∈ NL+R is the index of l in ordered set I1 . We observe
sponding row of Ω. We can see that (P1 ) is much easier to that the communication symbols indexed by NL+C \I1 , which
solve because there is only one matrix variable. However, as it are not sampled by the radar receiver, would introduce zero
will be seen in the simulations of Section VI.B, the constant interference power to the radar system.
rate transmission based on solving (9) is inferior to the adaptive If fsR > fsC , the interference at the radar receiver will be
transmission based on solving (17). over-sampled. One individual communication symbol will in-
It is clear that (P0 ) and (P1 ) have the same computational troduce interference to the radar system in fsR /fsC  consecu-
complexity, because the objectives and the constraints are simi- tive symbol durations. Let Il be the set of radar sampling time
lar. If an interior-point method [55] is used directly to the prob- instances during the period of the l-th communication symbol.
lems, the complexity is polynomial (cubic or slightly higher Note that Il is with cardinality fsR /fsC , and the collection of
orders) in the number of real variables in each problem. For sets I1 , . . . , IL C is a partition of NL+R . The effective interference
both (P0 ) and (P1 ), the semidefinite matrix variables {Rxl } power for both schemes of MIMO-MC radar is respectively
have LMt,C 2
real scalar variables. For the sub-optimal (P1 ),
there is only one semidefinite matrix variable Rx , which results 
LC  
2 EIP = Tr Δ l G2 Rxl GH ,
in Mt,C real scalar variables. Therefore, the computational costs 2
l=1
of (P0 ) and (P1 ) are at least L3 times of that of (P1 ), which 
are prohibitive if L is large. Fortunately, when (P0 ) and (P1 ) where Δ l = l Δl  . We observe that each individual com-
l  ∈I
are solved using Algorithm 1 based on dual decomposition, the munication transmit covariance matrix will be weighted by the
computation complexity is greatly reduced and scales linearly sum of interference channels for fsR /fsC  radar symbol dura-
with L. Furthermore, the overall computation time of (P0 ) and tions instead of one single interference channel.
(P1 ) using dual decomposition even becomes independent of L We conclude that in the above mismatched cases, the EIP
and thus equal to that of (P1 ) if all L sub-problems (P1 -sub) are expressions have the same form as those in the matched case
solved simultaneously in parallel using the same computational except the diagonal matrix Δl . To calculate the corresponding
routine [31]. diagonal matrices, the communication system only needs to
Authorized licensed use limited to: Indian Institute of Technology Hyderabad. Downloaded on October 18,2024 at 04:57:41 UTC from IEEE Xplore. Restrictions apply.
 4570 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 64, NO. 17, SEPTEMBER 1, 2016

know the sampling time of the radar system. Therefore, the

spectrum sharing problems in such cases can still be solved
using the proposed algorithms of Section IV.

VI. NUMERICAL RESULTS

For the simulations, we set the number of symbols to L = 32
and the noise variance to σC2 = 0.01. The MIMO radar sys-
tem consists of colocated TX and RX antennas forming half-
wavelength uniform linear arrays, and transmitting Gaussian
orthogonal waveforms [23]. The channel H is taken to have
independent entries, distributed as CN (0, 1). The interference
channels G1 and G2 are generated with independent entries, dis-
tributed as CN (0, σ12 ) and CN (0, σ22 ), respectively. The chan-
nels are Rayleigh fading and are consistent with a flat fad-
ing model assumption [5]–[9], [31], [32], [44], [45]. We fix
σ12 = σ22 = 0.01 unless otherwise stated. The maximum com-
munication transmit power is set to Pt = L (the power is nor-
malized w.r.t. the power of radar waveforms). The propagation
path from the radar TX antennas to the radar RX antennas via
the far-field target introduces a much more severe loss of power,
γ 2 , which is set to −30 dB in the simulations. The transmit
power of the radar antennas is fixed to ρ2 = ρ0  1000L/Mt,R
unless otherwise stated, and noise in the received signal is Fig. 2. MC relative recovery and target angle estimation success rate un-
der different levels of EIP for the MIMO-MC radar. M t , R = 16, M r, R =
added at SNR = 25 dB. The phase jitter variance is taken to be 32, M t , C = 4, M r, C = 4.
σα2 = 10−3 . The same uniformly random sampling scheme Ω0
is adopted by the radar in both the cooperative spectrum sharing
targets at angle 30◦ and 32.5◦ w.r.t. the radar arrays, with target
(SS) methods of (P0 ) and (P1 ). The joint-design spectrum shar-
reflection coefficients equal to 0.2 + 0.1j. The sub-sampling
ing method uses the same sampling matrix as its initial sampling
rate of MIMO-MC radar is fixed to 0.5. We simulate different
matrix. Recall that (P0 ) is the cooperative spectrum sharing
levels of EIP by setting the communication TX covariance ma-
method when Ω is not shared with the communication system.
trices equal to identity matrix and varying a scaling parameter.
In (P0 ), the communication system designs its waveforms by
In Fig. 2, we show the MC relative recovery errors and target
assuming Ω as the all 1’s matrix. Based on the obtained commu-
angle estimation success rates under different levels of EIP. The
nication waveforms, an EIP value is calculated for (P0 ) using
angle estimation is achieved by the MUSIC method based on
the true Ω for the ease of comparison. In the following figures,
5 pulses [24]. A success occurs if the angle estimation error is
we denote the cooperative spectrum sharing method of (P0 )
smaller than 0.25◦ . The results are calculated based on 200 in-
without knowledge of Ω by “cooperative SS w/ Ω unknown”.
dependent trials. One can see that the EIP indeed greatly affects
We denote the cooperative spectrum sharing method of (P1 ) by
the matrix completion accuracy and further the target angle es-
“cooperative SS”; and denote the joint-design spectrum sharing
timation. In particular, a 0.5 unit increase of EIP causes a sharp
method of (P2 ) by “joint-design SS”. The TFOCUS package
30% drop of the target angle estimation success rate. Therefore,
[59] is used for low-rank matrix completion at the radar fu-
in order to guarantee the function of the MIMO-MC radar, the
sion center. The communication covariance matrix is optimized
EIP has to be maintained at a small level.
according to the criteria of Section IV. The obtained Rxl is
1/2
used to generate x(l) = Rxl randn(Mt,C , 1). We use the EIP
and MC relative recovery error as the performance metrics. The B. Spectrum Sharing Based on Adaptive Transmission and
 / DS  , Constant Rate Transmission
relative recovery error is defined as  DS − DS F F

where DS is the completed result of DS. For comparison, we In this simulation, we compare the performance of the coop-
also implement a “selfish communication” scenario, where the erative scheme of (P1 ) based on adaptive transmission and the
communication system minimizes the transmit power to achieve constant rate transmission scheme of (P1 ). We also implement
certain average capacity without any concern about the interfer- the selfish communication scenario using constant rate trans-
ences it exerts to the radar system. mission. We take Mt,R = 4, Mr,R = Mt,C = 8, Mr,C = 4, and
one far-field stationary target at angle 30◦ w.r.t. the radar ar-
rays, with target reflection coefficient equal to 0.2 + 0.1j. For
A. The Impact of EIP on Matrix Completion and Target Angle the communication capacity constraint, we consider C = 12
Estimation
bits/symbol. Fig. 3 shows the EIP and MC relative recovery
In the following we provide simulation results in support error as functions of the sub-sampling rate at the MIMO-MC
of the use of EIP as a design objective. We take Mt,R = radar. We observe that the cooperative scheme of (P1 ) (labeled
16, Mr,R = 32, Mt,C = 4, Mr,C = 4. We consider two far-field as “Cooperative SS + Adaptive”) achieves much smaller EIP and
Authorized licensed use limited to: Indian Institute of Technology Hyderabad. Downloaded on October 18,2024 at 04:57:41 UTC from IEEE Xplore. Restrictions apply.
LI et al.: OPTIMUM CO-DESIGN FOR SPECTRUM SHARING BETWEEN MATRIX COMPLETION BASED MIMO RADARS 4571

Fig. 3. Spectrum sharing based on adaptive transmission and constant Fig. 5. Spectrum sharing with the MIMO-MC radar under different sub-
rate transmission for the MIMO-MC radar. M t , R = 4, M r, R = M t , C = sampling rates. M t , R = 4, M r, R = M t , C = 8, M r, C = 4. Dashed curves
8, M r, C = 4. correspond to EIP results using different realization of Ω 0 .

times are independent of L; 2) the running times of the adaptive

rate algorithms, both selfish and cooperative ones, scale lin-
early with L; 3) the joint-design spectrum sharing method takes
about 2–3 times of the cooperative spectrum sharing methods’
running time. These observations match our complexity analysis
in Section IV.C.

C. Spectrum Sharing Between a MIMO-MC Radar and a

MIMO Communication System
1) Performance Under Different Sub-Sampling Rates:
Fig. 4. CPU time comparison for various spectrum sharing algorithms under There is a far-field stationary target at angle 30◦ w.r.t. the radar
different values of waveform length L.
arrays, with target reflection coefficient equal to 0.2 + 0.1j. For
the communication capacity constraint, we consider C = 12
MC errors than the constant rate transmission scheme of (P1 ) bits/symbol. The sub-sampling rate of MIMO-MC radar varies
(labeled as “Cooperative SS + Const. Rate”) does. It can also from 0.2 to 1. The following two scenarios are considered.
be seen that the constant rate transmission scheme is inferior In the first scenario, we take Mt,R = 4, Mr,R = Mt,C =
even to the adaptive transmission based selfish communication 8, Mr,C = 4. In Fig. 5(a) we plot the EIP results for 4 different
scheme. This implies that the adaptive transmission technique realizations of Ω0 . For better visualization, Fig. 5(b) shows the
plays an important role in reducing the EIP and MC errors. In relative recovery errors averaged over all 4 realizations of Ω0 .
the following, the performance of adaptive transmission based The cooperative scheme (see P1 ) outperforms its counterpart
schemes is evaluated in more detail. As we already mentioned without knowledge of Ω (see P0 ) in terms of both the EIP
in Section II, when the sub-sampling rate p equals 1, the MIMO- and the MC relative recovery error. As discussed in Section IV,
MC radar becomes the traditional MIMO radar. Therefore, the the EIP is significantly reduced by the cooperative method when
above comparison between the adaptive and the constant rate p < 0.6, i.e., when pMr,R is much smaller than Mt,C . The joint-
transmission scheme for MIMO-MC radars also holds for tra- design scheme in this scenario performs the same as the coop-
ditional MIMO radars. erative scheme, possibly because the row dimension of Ω is too
To get an idea of the complexity involved in the aforemen- small to generate sufficient difference in EIP among the various
tioned simulations, we recorded the CPU times for the various permutations of Ω.
spectrum sharing algorithms executed on a laptop with Intel In the second scenario, we take Mt,R = 16, Mr,R =
Core i7 CPU and 8 GB memory. Fig. 4 shows the CPU times 32, Mt,C = 4, Mr,C = 4. In Fig. 6(a), we plot the EIP corre-
under different values of waveform length. One can observe that sponding to 4 different realizations of Ω0 , taken as uniformly
1) the constant rate algorithms are the fastest and their running random sampling matrices. Again, Fig. 6(b) shows the relative

Authorized licensed use limited to: Indian Institute of Technology Hyderabad. Downloaded on October 18,2024 at 04:57:41 UTC from IEEE Xplore. Restrictions apply.
 4572 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 64, NO. 17, SEPTEMBER 1, 2016

Fig. 6. Spectrum sharing with the MIMO-MC radar under different sub- Fig. 8. Spectrum sharing with the MIMO-MC radar under different capacity
sampling rates. M t , R = 16, M r, R = 32, M t , C = M r, C = 4. Dashed curves constraints C . M t , R = 16, M r, R = 32, M t , C = M r, C = 4. Dashed curves
correspond to EIP results using different realization of Ω 0 . correspond to EIP results using different realization of Ω 0 .

Fig. 9. Spectrum sharing with the MIMO-MC radar when multiple targets
present. M t , R = 16, M r, R = 32, M t , C = M r, C = 4p = 0.5 and C = 12
bits/symbol.

In the above scenarios, we would like p ≥ 0.5 for a small rel-

ative recovery error during matrix completion. However, values
of p > 0.7 require more samples while achieving little or even no
improvement on the recovery accuracy. Therefore, the optimal
range of p is [0.5, 0.7], where the proposed joint-design scheme
significantly outperforms the “selfish communication method”
Fig. 7. Spectrum sharing with the MIMO-MC radar under different capac- and the “SS method w/o knowledge of Ω”. We conclude that
ity constraints C . M t , R = 4, M r, R = M t , C = 8, M r, C = 4. Dashed curves
the proposed co-design based spectrum sharing methods utilize
correspond to EIP results using different realization of Ω 0 .
the sub-sampling procedure in the MIMO-MC radar to achieve
small EIP and high matrix recovery accuracy.
recovery errors averaged over all 4 realizations of Ω0 . The co- Interestingly, for the joint-design based spectrum sharing
operative scheme outperforms the cooperative scheme without method, the relative recovery error achieved by the sub-sampling
knowledge of Ω only marginally. This is due to the fact that rate p ∈ [0.5, 0.9] is even smaller than that by full samp
both G2 and G2l are full rank. The joint-design scheme (see ling p = 1. This indicates that due to the achieved small EIP, the
Section IV.B) optimizes Ω starting from the same sampling ma- full signal matrix is accurately completed. The completion pro-
trix used by the other three methods. In this case, the joint-design cess smooths out the noise, and as result, the completed matrix
scheme achieves smaller EIP and relative recovery errors than enjoys higher SIR that the initial full signal matrix. Therefore,
the other three methods. the sub-sampling procedure in MIMO-MC radar is beneficial

Authorized licensed use limited to: Indian Institute of Technology Hyderabad. Downloaded on October 18,2024 at 04:57:41 UTC from IEEE Xplore. Restrictions apply.
LI et al.: OPTIMUM CO-DESIGN FOR SPECTRUM SHARING BETWEEN MATRIX COMPLETION BASED MIMO RADARS 4573

Mt,R = 4, Mr,R = Mt,C = 8, Mr,C = 4. For the “selfish com-

munication” scheme and the cooperative scheme without knowl-
edge of Ω, the EIP and relative recovery errors increase as the
communication capacity increases. In contrast, the cooperative
and joint-design schemes achieve significantly smaller EIP and
relative recovery errors under all values of C. This indicates
that the latter two spectrum sharing methods successfully allo-
cate the communication transmit power in directions that result
in high communication rate, but small EIP to the MIMO-MC
radar.
The results for Mt,R = 16, Mr,R = 32, Mt,C = Mr,C = 4
are shown in Fig. 8. Since Mr,R is much larger than Mt,C ,
the cooperative scheme with the knowledge of Ω outperforms
its counterpart without knowledge of Ω only marginally. Mean-
while, the joint-design scheme can effectively further reduce the
EIP and relative recovery errors.
3) Performance Under Different Number of Targets: In this
simulation, we fix p = 0.5 and C = 12 and evaluate the perfor-
mance when multiple targets are present. The target reflection
coefficients are designed such that the target returns have fixed
Fig. 10. Spectrum sharing with the MIMO-MC radar under different levels of power, independent of the number of targets. We observe that
radar TX power. M t , R = 16, M r, R = 32, M t , C = M r, C = 4. the EIPs of different methods remain constant for different num-
ber of targets. This is because the design of the communication
waveforms is not affected by the target number. Fig. 9 shows
the results of the relative recovery error, which increases as the
number of targets increases. All methods have large recovery er-
ror for large number of targets, because the retained samples are
not sufficient for reliable matrix completion under any level of
noise. The proposed joint-design scheme can work effectively
for the MIMO-MC radar when a moderate number of targets
are present.
4) Performance Under Different Levels of Radar TX Power:
In this simulation, we evaluate the effect of radar TX power
ρ2 , while fixing p = 0.5, C = 12 and the target number to be
1. Fig. 10 shows the results of EIP and relative recovery errors
for Mt,R = 16, Mr,R = 32, Mt,C = Mr,C = 4. Again, we see
that the joint-design scheme performs the best, followed by
the cooperative scheme with the knowledge of Ω and then the
cooperative scheme without knowledge of Ω. When the radar
TX power increases, the EIP increases but with a much slower
rate. Therefore, increasing the radar TX power improves the
relative recovery errors.
Fig. 11. Spectrum sharing with the MIMO-MC radar under different chan- 5) Performance Under Different Interference Channel
nel variance σ 12 for the interference channel G 1 . M t , R = 16, M r, R = Strength: In this simulation, we evaluate the effect the inter-
32, M t , C = M r, C = 4.
ference channel G1 with different σ12 , while fixing p = 0.5,
C = 12 and the target number to be 1. As the communica-
for radar-communication spectrum sharing in terms of improv- tion RX gets closer to the radar TX antennas, σ12 gets larger.
ing radar SINR as well as reducing the amount of data to be sent Fig. 11 shows the results of EIP and relative recovery errors for
to the fusion center. Mt,R = 16, Mr,R = 32, Mt,C = Mr,C = 4. For all the spec-
In addition, simulations indicate that the communication av- trum sharing methods, when the interference channel G1 gets
erage capacity constraint holds with equality in both scenarios, stronger, the communication TX increases its transmit power
confirming observation (1) of Section IV.A. in order to satisfy the capacity constraint. Therefore, the EIP
2) Performance Under Different Capacity Constraints: In and the relative recovery errors increases with the variance
this simulation, the constant C in the communication capac- σ12 . We also observe that the joint-design scheme performs
ity constraint of (9b) varies from 6 to 14 bits/symbol, while the best, followed by the cooperative scheme with the knowl-
the sub-sampling rate p is fixed to 0.5. Four different real- edge of Ω and then the cooperative scheme without knowledge
izations of Ω0 are considered. Fig. 7 shows the results for of Ω.

Authorized licensed use limited to: Indian Institute of Technology Hyderabad. Downloaded on October 18,2024 at 04:57:41 UTC from IEEE Xplore. Restrictions apply.
 4574 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 64, NO. 17, SEPTEMBER 1, 2016

VII. CONCLUSION [5] S. Sodagari, A. Khawar, T. C. Clancy, and R. McGwier, “A projection

based approach for radar and telecommunication systems coexistence,” in
This paper has considered spectrum sharing between a MIMO Proc. IEEE Global Telecommun. Conf., Dec. 2012, pp. 5010–5014.
communication system and a MIMO-MC radar system. In or- [6] A. Babaei, W. H. Tranter, and T. Bose, “A practical precoding approach for
radar/communications spectrum sharing,” in Proc. 8th Int. Conf. Cognit.
der to reduce the effective interference power (EIP) at radar Radio Orient. Wireless Netw., Jul. 2013, pp. 13–18.
RX antennas, we have first considered the cooperative spec- [7] S. Amuru, R. M. Buehrer, R. Tandon, and S. Sodagari, “MIMO radar
trum sharing method, which designs the communication trans- waveform design to support spectrum sharing,” in Proc. IEEE Military
Commun. Conf., Nov. 2013, pp. 1535–1540.
mit covariance matrix based on the knowledge of the radar [8] A. Khawar, A. Abdel-Hadi, and T. C. Clancy, “Spectrum sharing between
sampling scheme. We have also formulated the spectrum shar- S-band radar and LTE cellular system: A spatial approach,” in Proc. IEEE
ing method for the case where the radar sampling scheme is Int. Symp. Dynam. Spectrum Access Netw., Apr. 2014, pp. 7–14.
[9] C. Shahriar, A. Abdelhadi, and T. C. Clancy, “Overlapped-MIMO radar
not shared with the communication system. Our theoretical re- waveform design for coexistence with communication systems,” in Proc.
sults guarantee that the cooperative approach can effectively IEEE Wireless Commun. Netw. Conf., 2015, pp. 223–228.
reduce the EIP to a larger extent as compared to the spectrum [10] H. Deng and B. Himed, “Interference mitigation processing for
spectrum-sharing between radar and wireless communications systems,”
sharing method without the knowledge of the radar sampling IEEE Trans. Aerosp. Electron. Syst., vol. 49, no. 3, pp. 1911–1919,
scheme. Second, we have proposed a joint design of the com- Jul. 2013.
munication transmit covariance matrix and the radar sampling [11] A. Aubry, A. De Maio, M. Piezzo, and A. Farina, “Radar waveform design
in a spectrally crowded environment via nonconvex quadratic optimiza-
scheme to further reduce the EIP. The EIP reduction and the tion,” IEEE Trans. Aerosp. Electron. Syst., vol. 50, no. 2, pp. 1138–1152,
matrix completion recovery errors have been evaluated under 2014.
various system parameters. We have shown that the MIMO-MC [12] A. Aubry, A. De Maio, Y. Huang, M. Piezzo, and A. Farina, “A new
radar waveform design algorithm with improved feasibility for spectral
radars enjoy reduced interference by the communication system coexistence,” IEEE Trans. Aerosp. Electron. Syst., vol. 51, no. 2, pp. 1029–
when the proposed spectrum sharing methods are considered. 1038, Apr. 2015.
In particular, the sparse sampling at the radar RX antennas can [13] S. C. Surender, R. M. Narayanan, and C. R. Das, “Performance analysis
of communications & radar coexistence in a covert UWB OSA system,”
reduce the rank of the interference channel. Our simulations in Proc. IEEE Global Telecommun. Conf., 2010, pp. 1–5.
have confirmed that significant EIP reduction is achieved by [14] A. Turlapaty and Y. Jin, “A joint design of transmit waveforms for radar
the cooperative approach; this is because in that approach, the and communications systems in coexistence,” in Proc. IEEE Radar Conf.,
2014, pp. 0315–0319.
communication power is allocated to directions in the null space [15] E. Fishler, A. Haimovich, R. Blum, D. Chizhik, L. Cimini, and
of the effective interference channel. Our simulations have sug- R. Valenzuela, “MIMO radar: An idea whose time has come,” in Proc.
gested that the joint-design scheme can achieve much smaller IEEE Radar Conf., Apr. 2004, pp. 71–78.
[16] J. Li and P. Stoica, “MIMO radar with colocated antennas,” IEEE Signal
EIP and relative recovery errors than other methods when the Process. Mag., vol. 24, no. 5, pp. 106–114, 2007.
number of radar TX and RX antennas is moderately large. [17] J. Li, P. Stoica, L. Xu, and W. Roberts, “On parameter identifiability of
The adaptive communication transmission has been shown to MIMO radar,” IEEE Signal Process. Lett., vol. 14, no. 12, pp. 968–971,
Dec. 2007.
be the optimal scheme for the considered spectrum sharing sce- [18] C. Chen and P. P. Vaidyanathan, “MIMO radar space time adaptive pro-
nario. Compared to the constant rate transmission, the adaptive cessing using prolate spheroidal wave functions,” IEEE Trans. Signal
transmission requires higher computational and implementation Process., vol. 56, no. 2, pp. 623–635, Feb. 2008.
[19] C. Chen and P. P. Vaidyanathan, “MIMO radar ambiguity properties and
complexity. To reduce the computation complexity, efficient al- optimization using frequency-hopping waveforms,” IEEE Trans. Signal
gorithms have been provided based on the Lagrangian dual Process., vol. 56, no. 12, pp. 5926–5936, 2008.
decomposition. As more and more powerful digital signal pro- [20] M. A. Herman and T. Strohmer, “High-resolution radar via compressed
sensing,” IEEE Trans. Signal Process., vol. 57, no. 6, pp. 2275–2284, Jun.
cessors are used in modern communication terminals, advanced 2009.
adaptive transmission approaches ought to weigh heavily due to [21] Y. Yu, A. P. Petropulu, and H. V. Poor, “MIMO radar using compressive
the increasing demand on high spectral efficiency. Nevertheless, sampling,” IEEE J. Sel. Topics Signal Process., vol. 4, no. 1, pp. 146–163,
Feb. 2010.
the adaptive transmission approach considered in the paper pro- [22] M. Rossi, A. M. Haimovich, and Y. C. Eldar, “Spatial compressive sensing
vides useful insights on the optimal design of the MIMO com- for MIMO radar,” IEEE Trans. Signal Process., vol. 62, no. 2, pp. 419–430,
munication system coexisting with MIMO-MC radars, which Jan. 2014.
[23] S. Sun, A. P. Petropulu, and W. U. Bajwa, “Target estimation in colocated
deserves research attention despite the computational and im- MIMO radar via matrix completion,” in Proc. IEEE Int. Conf. Acoust.,
plementation complexity. Speech, Signal Process., May 2013, pp. 4144–4148.
[24] S. Sun, W. Bajwa, and A. P. Petropulu, “MIMO-MC radar: A MIMO radar
approach based on matrix completion,” IEEE Trans. Aerosp. Electron.
REFERENCES Syst., vol. 51, no. 3, pp. 1839–1852, Jul. 2015.
[1] B. Li and A. P. Petropulu, “Spectrum sharing between matrix completion [25] D. S. Kalogerias and A. P. Petropulu, “Matrix completion in colocated
based MIMO radars and a MIMO communication system,” in Proc. IEEE MIMO radar: Recoverability, bounds and theoretical guarantees,” IEEE
Int. Conf. Acoust., Speech, Signal Process., Apr. 2015, pp. 2444–2448. Trans. Signal Process., vol. 62, no. 2, pp. 309–321, Jan. 2014.
[2] Radar Spectrum Regulatory Overview, 2013, [Online]. Available: http:// [26] S. Sun and A. P. Petropulu, “Waveform design for MIMO radars with
www.darpa.mil/WorkArea/DownloadAsset.aspx?id=2147486331 matrix completion,” IEEE J. Sel. Topics Signal Process., vol. 9, no. 8,
[3] F. H. Sanders, R. L. Sole, J. E. Carroll, G. S. Secrest, and T. L. Allmon, pp. 1400–1414, Dec. 2015.
“Analysis and resolution of RF interference to radars operating in the [27] Y. Chi, L. L. Scharf, A. Pezeshki, and A. R. Calderbank, “Sensitivity
band 2700–2900 MHz from broadband communication transmitters,” U.S. to basis mismatch in compressed sensing,” IEEE Trans. Signal Process.,
Dept. of Commerce, Tech. Rep. Tech. Rep. NTIA Tech. Rep. TR-13-490, vol. 59, no. 5, pp. 2182–2195, May 2011.
2012. [28] B. Li and A. P. Petropulu, “Radar precoding for spectrum sharing between
[4] A. Lackpour, M. Luddy, and J. Winters, “Overview of interference mit- matrix completion based MIMO radars and a MIMO communication
igation techniques between WiMAX networks and ground based radar,” system,” in Proc. IEEE Global Conf. Signal Inf. Process., Dec. 2015,
in Proc. 20th Annu. Wireless Opt. Commun. Conf., Apr. 2011, pp. 1–5. pp. 737–741.

Authorized licensed use limited to: Indian Institute of Technology Hyderabad. Downloaded on October 18,2024 at 04:57:41 UTC from IEEE Xplore. Restrictions apply.
LI et al.: OPTIMUM CO-DESIGN FOR SPECTRUM SHARING BETWEEN MATRIX COMPLETION BASED MIMO RADARS 4575

[29] K. Letaief and W. Zhang, “Cooperative communications for cognitive [59] S. R. Becker, E. J. Candès, and M. C. Grant, “Templates for convex cone
radio networks,” Proc. IEEE, vol. 97, no. 5, pp. 878–893, May 2009. problems with applications to sparse signal recovery,” Math. Programm.
[30] S. Haykin, “Cognitive radar: A way of the future,” IEEE Signal Process. Comput., vol. 3, no. 3, pp. 165–218, 2011.
Mag., vol. 23, no. 1, pp. 30–40, Jan. 2006.
[31] R. Zhang and Y. Liang, “Exploiting multi-antennas for opportunistic spec-
trum sharing in cognitive radio networks,” IEEE J. Sel. Topics Signal Bo Li (S’13) received his B.E. degree in communi-
Process., vol. 2, no. 1, pp. 88–102, Feb. 2008. cation engineering from Lanzhou University, China,
[32] R. Zhang, Y. Liang, and S. Cui, “Dynamic resource allocation in cognitive in 2009, and the M.S. degree in electrical engineer-
radio networks,” IEEE Signal Process. Mag., vol. 27, no. 3, pp. 102–114, ing from Peking University, China, in 2012. Since
May 2010. September 2012, he has been working towards the
[33] S. J. Kim and G. B. Giannakis, “Optimal resource allocation for MIMO Ph.D. degree at the Department of Electrical & Com-
Ad Hoc cognitive radio networks,” IEEE Trans.Inf. Theory, vol. 57, no. 5, puter Engineering, Rutgers, The State University of
pp. 3117–3131, May 2011. New Jersey, and he is a member of the Signal Process-
[34] S. Bhojanapalli and P. Jain, “Universal matrix completion,” in Proc. 31st ing and Communications Laboratory (CSPL). His re-
Int. Conf. Mach. Learn., 2014, pp. 1881–1889. search interests are in signal processing for MIMO
[35] H. Krim and M. Viberg, “Two decades of array signal processing research: radar and communication systems, including com-
The parametric approach,” IEEE Signal Process. Mag., vol. 13, no. 4, pressive sensing and matrix completion based MIMO radar, spectrum sharing
pp. 67–94, 1996. in cooperative radar, and communication systems.
[36] E. J. Candes and Y. Plan, “Matrix completion with noise,” Proc. IEEE,
vol. 98, no. 6, pp. 925–936, Jun. 2010.
[37] F. M. Gardner, Phaselock Techniques. New York, NY, USA: Wiley, 2005.
[38] R. Poore, “Overview on phase noise and jitter,” Keysight Technologies, Athina P. Petropulu (F’08) received her undergrad-
Santa Rosa, CA, USA, Tech. Overview, 2014. [Online]. Available: http:// uate degree from the National Technical University
literature.cdn.keysight.com/litweb/pdf/5990-3108EN.pdf?id=827513 of Athens, Greece, and the M.Sc. and Ph.D. degrees
[39] R. Mudumbai, G. Barriac, and U. Madhow, “On the feasibility of dis- from Northeastern University, Boston MA, all in elec-
tributed beamforming in wireless networks,” IEEE Trans. Wireless Com- trical and computer engineering. Since 2010, she is
mun., vol. 6, no. 5, pp. 1754–1763, 2007. Professor of the Electrical and Computer Engineering
[40] B. Razavi, “A study of phase noise in CMOS oscillators,” IEEE J. Solid- (ECE) Department at Rutgers, having served as chair
State Circuits, vol. 31, no. 3, pp. 331–343, 1996. of the department during 2010–2016. Before that she
[41] C. Kopp, “Search and acquisition radars (S-band, X-band),” Tech. Rep. was faculty at Drexel University. Dr. Petropulu’s re-
Tech. Rep. APA-TR-2009-0101, 2009. [Online]. Available: http://www. search interests span the area of statistical signal pro-
ausairpower.net/APA-Acquisition-GCI.html cessing, wireless communications, signal processing
[42] Radar Performance, Radtec Engineering Inc., 2015. [Online]. Available: in networking, physical layer security, and radar signal processing. Her research
http://www.radar-sales.com/PDFs/Performance_RDR%26TDR.pdf has been funded by various government industry sponsors including the Na-
[43] “LTE in a nutshell: The physical layer,” (White paper), Telesystem In- tional Science Foundation, the Office of Naval research, the U.S. Army, the
novations, Markham, ON, Canada, 2010. [Online]. Available: https:// National Institute of Health, the Whitaker Foundation, Lockheed Martin.
www.scribd.com/doc/297688375/LTE-in-a-Nutshell-Physical-Layer Dr. Petropulu is recipient of the 1995 Presidential Faculty Fellow Award given
[44] T. S. Rappaport et al., Wireless Communications: Principles and Practice. by NSF and the White House. She has served as Editor-in-Chief of the IEEE
Englewood Cliffs, NJ, USA: Prentice-Hall PTR, 2001, vol. 2. TRANSACTIONS ON SIGNAL PROCESSING (2009–2011), IEEE Signal Processing
[45] D. Tse and P. Viswanath, Fundamentals of Wireless Communication. Society Vice President—Conferences (2006–2008), and member-at-large of the
Cambridge, U.K.: Cambridge Univ. Press, 2005. IEEE Signal Processing Board of Governors. She was the General Chair of
[46] A. Goldsmith, Wireless Communications. Cambridge, U.K.: Cambridge the 2005 International Conference on Acoustics Speech and Signal Processing
Univ. Press, 2005. (ICASSP-05), Philadelphia PA. In 2005 she received the IEEE Signal Pro-
[47] R. P. Jover, LTE PHY Fundamentals, 2015. [Online]. Available: cessing Magazine Best Paper Award, and in 2012 the IEEE Signal Processing
http://www.ee.columbia.edu/roger/LTE_PHY_fundamentals.pdf Society Meritorious Service Award. More information on her work can be found
[48] M. Filo, A. Hossain, A. R. Biswas, and R. Piesiewicz, “Cognitive pilot at www.ece.rutgers.edu/∼cspl
channel: Enabler for radio systems coexistence,” in Proc. 2nd Int. Work-
shop Cognit. Radio Adv. Spectrum Manag., May 2009, pp. 17–23.
[49] L. Lu, X. Zhou, U. Onunkwo, and G. Y. Li, “Ten years of research in Wade Trappe (F’14) is a Professor with the Depart-
spectrum sensing and sharing in cognitive radio.,” EURASIP J. Wireless ment of Electrical and Computer Engineering, Rut-
Commun Netw., vol. 2012, p. 28, 2012. gers University, New Brunswick, NJ, USA, and an
[50] G. Taubock, “Complex-valued random vectors and channels: Entropy, di- Associate Director of the Wireless Information Net-
vergence, and capacity,” IEEE Trans. Inf. Theory, vol. 58, no. 5, pp. 2729– work Laboratory (WINLAB), where he directs WIN-
2744, 2012. LABs research in wireless security. He has led several
[51] S. N. Diggavi and T. M. Cover, “The worst additive noise under a covari- federally funded projects in the area of cybersecu-
ance constraint,” IEEE Trans. Inf. Theory, vol. 47, no. 7, pp. 3072–3081, rity and communication systems, projects involving
Nov. 2001. security and privacy for sensor networks, physical
[52] C. Chen and P. P. Vaidyanathan, “MIMO radar waveform optimization layer security for wireless systems, a security frame-
with prior information of the extended target and clutter,” IEEE Trans. work for cognitive radios, the development of wire-
Signal Process., vol. 57, no. 9, pp. 3533–3544, 2009. less testbed resources (the ORBIT testbed), and new RFID technologies. His
[53] G. Cui, H. Li, and M. Rangaswamy, “MIMO radar waveform design with experience in network security and wireless spans over 15 years, and he has
constant modulus and similarity constraints,” IEEE Trans. Signal Process., coauthored a popular textbook in security, Introduction to Cryptography with
vol. 62, no. 2, pp. 343–353, 2014. Coding Theory, as well as several monographs on wireless security, including
[54] A. Goldsmith, S. A. Jafar, N. Jindal, and S. Vishwanath, “Capacity limits Securing Wireless Communications at the Physical Layer and Securing Emerg-
of MIMO channels,” IEEE J. Sel. Areas Commun., vol. 21, no. 5, pp. 684– ing Wireless Systems: Lower-layer Approaches. He served as an Editor for the
702, 2003. IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY (TIFS), the
[55] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.: IEEE Signal Processing Magazine (SPM), and the IEEE TRANSACTIONS ON
Cambridge Univ. Press, 2004. MOBILE COMPUTING (TMC). He served as the Lead Guest Editor for Septem-
[56] R. G. Bland, D. Goldfarb, and M. J. Todd, “The ellipsoid method: A ber 2011 IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY
survey,” Oper. Res., vol. 29, no. 6, pp. 1039–1091, 1981. Special Issue on Using the Physical Layer for Securing the Next Generation
[57] H. W. Kuhn, “The Hungarian method for the assignment problem,” Naval of Communication Systems and served as the IEEE Signal Processing Society
Res. Logist. Quart., vol. 2, no. 1–2, pp. 83–97, 1955. representative to the governing board of IEEE TMC. He is currently the IEEE
[58] J. Yeh, “Real analysis,” Theory of Measure and Integration. Singapore: SPS Regional Director for Regions 1–6.
World Scientific, 2006.

Authorized licensed use limited to: Indian Institute of Technology Hyderabad. Downloaded on October 18,2024 at 04:57:41 UTC from IEEE Xplore. Restrictions apply.