Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LSP.2020.3007033, IEEE Signal
Processing Letters 1

Efficient Joint Localization and Synchronization in


Distributed MIMO Radars
Seyed Amir Reza Kazemi, Rouhollah Amiri and Fereidoon Behnia

Abstract—This letter addresses the problem of joint localization reaches too high. Furthermore, the problem of joint localization
and clock synchronization in distributed multiple-input multiple- and synchronization remains unsolved.
output (MIMO) radar systems. While the well-known two-stage
This letter is considered as an improvement of the estimation
weighted least squares (WLS) method provides an acceptable es-
timate of target position when a rough approximation of antennas method proposed in [25]. By using a simple WLS estimator to
clock parameters, the drifts and offsets, is available, its performance further enhance the position estimate while refining the uncertain
can degrade quickly if the level of uncertainty in these values in- values of clock parameters, the presented method offers an effi-
creases. The proposed method offers a solution for synchronizing the cient solution for simultaneous localization and synchronization
clocks while simultaneously improving the target position estimate.
problem. Needless to say, the uncertainty in antennas locations
The uncertainty in positions of antennas is also taken into account.
The presented method is shown to be approximately unbiased and is also taken into account. The proposed estimator achieves
able to attain the Cramer-Rao lower bound (CRLB) under small the optimum performance as shown by theoretical analysis and
Gaussian error assumption. Numerical simulations are employed to corroborated by numerical simulations.
validate the theoretical findings.
Index Terms—Target localization, synchronization, multiple-input
multiple-output (MIMO) radar, time delay measurement, Cramer- II. P ROBLEM F ORMULATION
Rao lower bound (CRLB), weighted least squares (WLS) estimation. We consider the same localization scenario as in [25] where a
MIMO radar system consisting of Nt transmit antennas situated
at toi ∈ R3 (i = 1, . . . , Nt ) and Nr receive antennas at soj ∈ R3
I. I NTRODUCTION (j = 1, . . . , Nr ) is employed to estimate the unknown position of
a single target located at uo ∈ R3 . The exact values of antennas
T HE topic of target localization in distributed multiple-input
multiple-output (MIMO) radar systems has received substan-
tial attention over the past decade owing to its extensive range of
positions are not known and only the erroneous versions that
are corrupted by random error, are available. Considering such
application in surveillance and tracking [1]. Generally speaking, uncertainties for position of each antenna, we define the total
localization methods in MIMO radars can be based on different faulty position vector p as
types of measurements, such as time delay (TD) [2]–[10], Doppler
shift [11]–[20] and angle of arrival [21]–[24]; however, TD-based p = po + ∆p (1)
methods are the most prevalent one due to their high accuracy ∆
and simplicity. where po = [soT oT oT oT T
1 ,. . .,sNr , t1 ,. . .,tNt ] and ∆p is the error vector

Although there exist numerous works in the literature that assumed to be zero-mean Gaussian with covariance matrix Qp .
propose positioning methods based on TD measurements (e.g., The exact values of transmitters clock drifts εot and offsets η ot ,
[2]–[9]), they rarely address the practical issues of localization which model clock imperfections, are also corrupted by random
such as non-ideal clocks and uncertainty in positions of antennas. error such that the faulty values are
The assumption that the transmit and receive antennas have
εt = εot + ∆εt , η t = η ot + ∆η t . (2)
ideal clocks and are fully synchronized is highly unreasonable,
particularly when the nonzero drifts and offsets with respect to Similarly, for the receive antennas we have
the global clock cannot be neglected. In our recent work [25], we
proposed a closed-form method for target localization in MIMO εs = εos + ∆εs , η s = η os + ∆η s . (3)
radar systems in the presence of such practical issues as antenna
position uncertainties and clock parameter imperfections. This By collecting (2) and (3), we form the clock parameters vector
method makes use of consecutive weighted least squares (WLS) χ as
T
estimations to arrive at a satisfactory estimate for coordinates of χ = εT , η T = χo + ∆χ

(4)
the target, and is shown to reach the Cramer-Rao lower bound
(CRLB) for low levels of measurement error. The problem is where ε = [εTt , εTs ]T and η = [η Tt , η Ts ]T are the drift and offset
 T
that since the equations involved in WLS estimation are highly vectors, respectively. The error vector ∆χ = ∆εT , ∆η T
nonlinear with respect to clock parameters, the performance is assumed to be a zero-mean Gaussian random vector with
of the estimator can deteriorate rapidly if no compensation is covariance matrix Qχ .
considered when the level of uncertainty in clock parameters The localization task is carried out via utilizing the TD mea-
surements extracted by the receivers. The delay associated with
The authors are with Department of Electrical Engineering, Sharif Uni-
versity of Technology, Tehran 1136511155, Iran. (e-mail: behnia@sharif.edu; a specific transmitter-receiver pair, and multiplied by the wave
arezrakazemi@yahoo.com; amiri rouhollah@ee.sharif.edu). propagation speed c is supposed to indicate the bistatic range

1070-9908 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Indian Institute of Technology - BHUBANESWAR. Downloaded on July 11,2020 at 04:58:08 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LSP.2020.3007033, IEEE Signal
Processing Letters 2


corresponding to that exact pair; however, owing to the non-ideal where ρa1 ,a2 = (a1 − a2 )/ ka1 − a2 k. Note that ωi,j and ψi,j
clock of antennas, we have contain the estimation parameters and the bothersome error terms,
kuo − toi k + kuo − soj k + cτio + ni,j respectively.
ri,j = + ηsoj (5) Next, we rewrite (9) as λi,j − R̂sj ' R̂ti + ψi,j − ωi,j and
1 + εosj
square both sides to get
where ri,j /c denotes the time measured by the jth receiver when di,j + 2R̂ti ωi,j ' 2R̂ti ψi,j (10)
the signal from the ith transmitter is received, and ni,j is the
inherent additive zero-mean Gaussian TD measurement noise. Let where
us collect all ni,j ’s in a vector n with covariance matrix Qr . 2 2
di,j = λ2i,j + ksj k − kti k − 2(sj − ti )T û − 2λi,j R̂sj
Moreover, in the above equation, τio is the transmission time, the
value of which is sent by the ith transmitter. Since this value is and only the first-order error terms are kept. Additionally, we
measured with respect to the transmitter’s non-ideal clock, it is make use of the statistical information hidden in ∆χ to write
subject to its inherent imperfections, and thus the actual reported [26]
transmission time by the ith transmitter is in fact equal to 02(Nt +Nr ) + ∆χ = ∆χ. (11)
τio By stacking the two sets of equations in (10) and (11), we
τi = + ηtoi /c. (6) ∆
1 + εoti obtain the following linear matrix equation with respect to φo =
T T T
Using (5) and (6) to eliminate τio , yields the following equation [∆u , ∆χ ]
b − Mφo = ξ (12)
λoi,j = kuo − toi k + kuo − soj k + ni,j (7)
where b = [dT , 0T2(Nt +Nr ) ]T , [d]k = di,j and
where we have defined  
M1 M2

λoi,j = (1 + εosj )(ri,j − ηsoj ) + (1 + εoti )(ηtoi − cτi ). (8) M=
O2(Nt +Nr )×3 −I2(Nt +Nr )
T T
The above equation is deeply involved in development of the [M1 ]k,: = −2R̂ti (ρû,sj
+ ρû,ti
)
proposed joint localization and synchronization algorithm, as we [M2 ]k,: = 2R̂ti [0Ti−1 , ηti − cτi , 0TNt −i+j−1 , ri,j − ηsj ,
shall demonstrate in the next section.
0TNr −j+i−1 , 1 + εti , 0TNt −i+j−1 , −1 − εsj , 0TNr −j ]
III. P ROPOSED A LGORITHM where the index k in the above equations is k = (j − 1)Nt + i for
I) First Stage: The first stage of the algorithm is identical to the i = 1, 2, . . . , Nt and j = 1, 2, . . . , Nr . Also the error term ξ is
localization method proposed in our previous work [25], which expressed as ξ = [(A1 n + B1 ∆p)T , ∆χT ]T , where the matrices
we will summarize briefly. By manipulating (7) and introducing A1 and B1 are defined in [25, Sec. III]. The optimum WLS
some auxiliary variables, also called nuisance parameters, the estimator for the given model in (12) is [27]
aforementioned method establishes a set of linear equations, φ̂ = (MT WM) MT Wb
−1
(13)
whose estimated solution is obtained via a WLS estimator. This  T −1
solution contains a rough estimate of uo and auxiliary variables. where W = E ξξ is the weighting matrix equal to
The estimated nuisance parameter values are subsequently ex-
W ' blkdiag(Σ−1 , Q−1
χ ) (14)
ploited to further enhance the accuracy of the estimated target
location, which we denote by û. For more details, please refer to where Σ = A1 Qr AT1
+ B1 Qp BT1 .
Accordingly, the estimated
[25, Sec. III]. correction terms are simply obtained as ∆û = φ̂(1:3) and ∆χ̂ =
II) Second Stage: In this stage, we aim to enhance the initial φ̂(4:3+2(Nt +Nr )) . Finally, the refined estimates of target position
estimation of the target position û while simultaneously achieving and antennas clock parameters, respectively, are acquired as
an improved estimate for antennas clock parameters. By consid-
ering the correction terms ∆u = û − uo and ∆χ = χ − χo , ū = û − ∆û
we try to determine these values such that subtracting them from χ̄ = χ − ∆χ̂. (15)
û and χ, respectively, provides better target position and clock
parameters estimates. IV. P ERFORMANCE A NALYSIS
We start by expressing the true values uo , po and χo in terms of In estimation-related studies, including localization, it is usually
their erroneous values û−∆u, p−∆p and χ−∆χ, respectively, desired that the optimality properties of the proposed estimator,
and substituting them in (7). Doing so while ignoring the second namely unbiasedness and efficiency, be investigated. We shall es-
and higher-order noise terms results in tablish this by means of theoretical analysis. We first point out that
in our previous work, we showed that under small measurement
λi,j ' R̂ti + R̂sj + ψi,j − ωi,j (9)
error conditions, the proposed two-stage WLS method is unbiased
where λi,j is the noisy version of λoi,j in (8), R̂ti = kû − ti k, and able to reach the CRLB. These conditions are [25, Sec. IV-C]
R̂sj = kû − sj k and C1)k∆sj k  Rsoj , C2)k∆ti k  Rtoi ,
T T
ψi,j = ni,j + ρû,ti
∆ti + ρû,sj
∆sj C3)|∆ηti |  |ηtoi |, C4)|∆εti |  |εoti |,
T
ωi,j = (ρû,sj
T
+ ρû,ti
)∆u − (ηti − cτi )∆εti C5)|∆ηsj |  |ηsoj |, C6)|∆εsj |  |εosj |,
− (1 + εti )∆ηti − (ri,j − ηsj )∆εsj + (1 + εsj )∆ηsj C7)|ni,j | Rtoi + Rsoj , i = 1, . . . , Nt , j = 1, . . . , Nr

1070-9908 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Indian Institute of Technology - BHUBANESWAR. Downloaded on July 11,2020 at 04:58:08 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LSP.2020.3007033, IEEE Signal
Processing Letters 3

∆ ∆
where Rtoi = kuo − toi k and Rsoj = kuo − soj k. Thus, to prove where we have defined K1 = A−1 −1
1 M1 and K2 = A1 M2 . By
performance optimality, the statistical properties of the refinement paying attention to the Woodbury matrix inversion lemma [28],
stage estimate φ̂ must be analyzed. To do so, we first derive the we notice that the right-hand side
−1 of the above equation is equal
covariance matrix of φ̂. By substituting b from (12) into (13), to KT1 (S + K2 Qχ KT2 )−1 K1 , which results in
we can obtain the second-stage estimation error term ∆φ as −1
cov(ū) ' KT1 (Qr + K3 Qp KT3 + K2 Qχ KT2 )−1 K1 (23)
∆ −1
∆φ = φ̂ − φo = (MT WM) MT Wξ. (16) ∆
where K3 = A−1 o
1 B1 . Furthermore, CRLB(u ) is explicitly found
Taking the expected value of (16) while considering the first-order from applying the block matrix inversion formula [28] to (21), as
error analysis produces
T −1
CRLB (uo ) = J11 − J12 J−1

−1 22 J12 . (24)
E{∆φ} ' (MT WM) MT WE{∆ξ} ' 0. (17)
By substituting the values from (22) into (24), and utilizing the
In light of this result and the expressions in (15), we conclude Woodbury matrix inversion lemma (similar to the procedure of
that E{ū} ' uo and E{χ̄} ' χo . Using similar justifications, deriving (23)), we arrive at
the covariance matrix of φ̂ can be expressed as
CRLB(uo ) = ∇Tuo (Qr + ∇po Qp ∇Tpo
cov(φ̂) = E{∆φ∆φT } ' (MT WM)−1 (18) −1
+ ∇χo Qχ ∇Tχo )−1 ∇uo . (25)
where W is given in (14). In order to write covariance matrices
We observe by comparing (25) with (23) that they have similar
of the estimated target position and clock parameters separately,
forms. In fact, it can be easily verified after some algebraic
we note that rewriting W and M in their partitioned form and
manipulations that we have K1 ' ∇uo , K2 ' ∇χo and
applying multiplication, yield
  K3 ' ∇po , from which cov(ū) ' CRLB(uo ) results. The proof
Z1 Z2 of CRLB(χo ) ' cov(χ̄) can be done in an analogous manner. 
MT WM = (19)
ZT2 Z3 As a final note, we contend that by using the refinement stage
∆ ∆
presented in Sec. III, enhancement in clock parameters accuracy
where we have defined Z1 = MT1 Σ−1 M1 , Z2 = MT1 Σ−1 M2 is guaranteed, as summarized by the following proposition.

and Z3 = MT2 Σ−1 M2 +Q−1 χ . By employing the block matrix in- Proposition 2: The incorporation of the refinement stage in
version lemma [28], the desired covariance matrices are obtained the proposed method leads to an increased accuracy in clock
as parameters estimation.
T −1 Proof: The covariance matrix in (20b) equals
cov(ū) ' Z1 − Z2 Z−1

3 Z2 (20a)
T −1
T −1
cov(χ̄) ' Z3 − Z2 Z1 Z2
−1
. (20b) cov(χ̄) ' Q−1
χ + M2 Σ M2
−1
− MT2 Σ−1 M1 (MT1 Σ−1 M1 )−1 MT1 Σ−1 M2 . (26)
The CRLB based on the measurement model presented in Sec.
II is equal to the inverse of Fisher information matrix given by The right-hand side of (26) seems quite complex; nonetheless,
[25, Sec. IV-A]   by inserting Σ = A1 SAT1 into (26) and subsequently replacing
FIM = 11
J J12
(21) A−T
1 S
−1 −1
A1 with its Cholesky decomposition LLT , (26) can
where JT12 J22 be written in the following succinct form
−1
J11 = ∇Tα−β,uo Q−1
r ∇α−β,uo
cov(χ̄) ' Q−1 T
χ + M2 LVL M2
T
(27)
J12 = ∇Tα−β,uo Q−1
r ∇α−β,γ o where V = I − LT M1 (MT1 LLT M1 )−1 MT1 L. It is readily
−1 confirmed that V is a symmetric idempotent matrix, and thus
J22 = ∇Tα−β,γ o Q−1
r ∇α−β,γ o + blkdiag (Qp , Qχ ) (22)
we can write V = VT V. Doing so produces
where ∇α−β,uo and ∇α−β,γ o are Jacobian matrices whose val- −1
ues are derived in [25, Appendix A]. For brevity, we refer to these cov(χ̄) ' Q−1
χ +U U
T
(28)
matrices using shortened notations ∇uo and ∇γ o , respectively. where U = VLT M2 . Applying the Woodbury matrix inversion
We also partition ∇γ o as [∇po , ∇χo ] where ∇po contains the lemma to (28) gives
first 3(Nt + Nr ) columns of ∇γ o and ∇χo contains the rest. The
CRLB(uo ) and CRLB(χo ) are the 3 × 3 upper-left block and cov(χ̄) ' Qχ − Qχ UT (I + UQχ UT )−1 UQχ . (29)
2(Nt +Nr )×2(Nt +Nr ) lower-right block of FIM−1 , respectively.
It is evident that the second term in right-hand side of (29) is
Having obtained the CRLB and covariance matrices, the fol- a positive semidefinite matrix, from which Qχ − cov(χ̄)  0
lowing proposition establishes the efficiency of the proposed follows, hence completing the proof. 
estimator.
Proposition 1: The proposed two-stage method for estimating
V. S IMULATION R ESULTS
both target position and antennas clock parameters reaches the
CRLB under conditions C1-C7. In this section, we intend to evaluate the performance of the
Proof: We first rewrite Σ−1 as Σ−1 = (A1 SAT1 )−1 where proposed method using numerical simulations, and confirm the
S = Qr + (A−1 −1 T theoretical findings established in the previous section. Specifi-
1 B1 )Qp (A1 B1 ) . From (20a), we have
 −1 cally, we aim to showcase the efficiency of the proposed estimator
cov(ū) ' KT1 S−1 − S−1 K2 (KT2 S−1 K2 + Q−1 χ )
−1
K1 in both localization and synchronization, and the improvement

1070-9908 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Indian Institute of Technology - BHUBANESWAR. Downloaded on July 11,2020 at 04:58:08 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LSP.2020.3007033, IEEE Signal
Processing Letters 4

TABLE I: Transmitters and Receivers True Positions (in Meters).


Section IV which confirms the efficiency of the proposed method
Tx no. i xoti ytoi ztoi Rx no. j xosj ysoj zsoj as well as the improvement made by incorporating the refinement
1 -200 -300 200 1 -450 -450 200 stage within the method.
2 -200 300 100 2 450 450 100
3 200 300 80 3 0 600 200
4 200 -300 120 4 600 0 100 25
Proposed Method
5 - - - 5 -600 0 150 2S-WLS
20 Root CRLB
6 - - - 6 0 -600 100
15

10 log(Position RMSE(m))
10

regarding the estimation of clock parameters. To this end, we 5


attempt to localize a target situated at uo = [400, 800, 300]T m
0
by employing a MIMO radar with Nt = 4 transmitters and
Nr = 6 receivers, whose true positions in 3-D space is tabulated -5

in Table I. The erroneous versions of these positions are generated -10

by adding a zero-mean Gaussian noise with covariance matrix -15


-30 -25 -20 -15 -10 -5
Qp = σp2 I3(Nt +Nr ) to the exact values, as represented in (1). 10 log( )

The true clock drifts and offsets of each antenna are selected Fig. 1: RMSE(ū) as a function of measurement error level, comparing performance of the
randomly according to uniform distribution over [−0.002, 0.002] proposed method with the CRLB.
and [10, 50] m, respectively. Also, the transmission times are
uniformly adopted over [10, 100]/c s, where c is the wave
propagation speed. To produce noisy measurements, similar to -20
Proposed method
Available measurement
the sensors positions, the noisy values of TD measurements and -25 Root CRLB

clock parameters are generated by corrupting the actual values

10 log(Average Drift RMSE)


-30
with additive zero-mean Gaussian noise with covariance matrices
Qr = σr2 INt Nr and Qχ = blkdiag(σε2 INt +Nr , ση2 INt +Nr ), where -35

σr2 , σε2 and ση2 denote variance of the TD measurements, clock -40

drifts and offsets, respectively. Note that the proposed method


-45
is designed for the general forms of the covariance matrices,
however, the aforementioned diagonal forms are considered for -50

ease of analysis. -55


-30 -25 -20 -15 -10 -5
The accuracy of the estimator is evaluated via root mean square 10 log( )

error (RMSE) criterion which can be computed by utilizing an Fig. 2: The average drift RMSE of clocks as a function of measurement error level, before
L-trial Monte-Carlo simulation method as and after applying the refinement stage.
r
XL
x(l) − xo 2 /L

RMSE (x) = (30)
l=1
15
Proposed method
where x can indicate the position of target or the clock parame- 10
Available measurement
Root CRLB

ters, and x(l) is the estimated value of xo at the lth iteration of the
10 log(Average Offset RMSE(m))

5
simulation. For our setup, we pick the number of Monte-Carlo
ensemble runs as L = 5000. 0

By introducing the scaling factor ρ, we set the noise variances -5

as σr = 0.5ρ m, σε = 0.01ρ, ση = 20ρ m and σp = 0.05ρ -10

m. Consequently, the RMSE can now be calculated as a function -15

of noise level. The simulation results are summarized in Figs.


-20
1-3. As illustrated in Fig. 1, the estimated position is able to
-25
attain the CRLB for small levels of measurement noise, which -30 -25 -20 -15 -10 -5
10 log( )
is consistent with theoretical results presented in Sec. IV. We
acknowledge that the deviation from the bound for higher noise Fig. 3: The average offset RMSE of clocks as a function of measurement error level, before
and after applying the refinement stage.
levels is attributed to a common phenomenon in WLS estimation
known as the threshold effect, that occurs on account of neglecting
the higher-order error terms in the estimator design (see, e.g., [7], VI. C ONCLUSION
[25], [29]). Furthermore, the proposed method outperforms our In this letter, we presented an efficient solution for the problem
previous work in [25]. The improvement in position estimation of joint target localization and clock synchronization in a dis-
may seem small, however, the emphasis of our work here is to tributed MIMO radar system where the uncertainty in positions
solve the problem of joint localization and synchronization which of antennas is also taken into consideration. The proposed method
is not provided in previous works such as [25]. simply employs a WLS estimator to refine the uncertain values
Figs. 2 and 3 depict the average RMSE of estimated clock of available clock parameters while simultaneously improving the
parameters before and after applying the refinement stage, and target position estimate. Under Gaussian noise model assumption,
compare these values with their corresponding CRLBs. These we demonstrated, both analytically and numerically, that the
results are clearly consistent with our theoretical analysis in proposed estimator can reach the CRLB for small noise levels.

1070-9908 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Indian Institute of Technology - BHUBANESWAR. Downloaded on July 11,2020 at 04:58:08 UTC from IEEE Xplore. Restrictions apply.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LSP.2020.3007033, IEEE Signal
Processing Letters 5

R EFERENCES [15] F. Zhang, Y. Sun, J. Zou, D. Zhang, and Q. Wan, “Closed-form localization
method for moving target in passive multistatic radar network,” IEEE
[1] H. Godrich, A. M. Haimovich, and R. S. Blum, “Target localization accuracy
Sensors Journal, vol. 20, no. 2, pp. 980–990, 2019.
gain in MIMO radar-based systems,” IEEE Trans. Inf. Theory, vol. 56, no. 6,
[16] R. Amiri, F. Behnia, and H. Zamani, “Closed-form positioning in MIMO
pp. 2783–2803, 2010.
radars with antenna location uncertainties,” IET Radar, Sonar & Navigation,
[2] M. Einemo and H. C. So, “Weighted least squares algorithm for target
2019.
localization in distributed MIMO radar,” Signal Process., vol. 115, pp. 144–
[17] A. Noroozi, M. A. Sebt, and A. H. Oveis, “Efficient weighted least squares
150, 2015.
estimator for moving target localization in distributed MIMO radar with
[3] R. Amiri and F. Behnia, “An efficient weighted least squares estimator for
location uncertainties,” IEEE Systems Journal, vol. 13, no. 4, pp. 4454–
elliptic localization in distributed MIMO radars,” IEEE Signal Process. Lett.,
4463, 2019.
vol. 24, no. 6, pp. 902–906, 2017.
[18] R. Amiri, F. Behnia, and A. Noroozi, “Efficient joint moving target and
[4] R. Amiri, F. Behnia, and H. Zamani, “Asymptotically efficient target
antenna localization in distributed MIMO radars,” IEEE Trans. Wireless
localization from bistatic range measurements in distributed MIMO radars,”
Commun., vol. 18, no. 9, pp. 4425–4435, 2019.
IEEE Signal Process. Lett., vol. 24, no. 3, pp. 299–303, 2017.
[19] H. Song, G. Wen, and L. Zhu, “An approximately efficient estimator for
[5] A. Noroozi, A. H. Oveis, and M. A. Sebt, “Iterative target localization in
moving target localization in distributed MIMO radar systems in presence
distributed MIMO radar from bistatic range measurements,” IEEE Signal
of sensor location errors,” IEEE Sensors Journal, vol. 20, no. 2, pp. 931–938,
Process. Lett., vol. 24, no. 11, pp. 1709–1713, 2017.
2019.
[6] C. H. Park and J. H. Chang, “Closed-form localization for distributed
[20] A. Noroozi, R. Amiri, M. M. Nayebi, and A. Farina, “Efficient closed-
MIMO radar systems using time delay measurements,” IEEE Trans. Wireless
form solution for moving target localization in MIMO radars with minimum
Commun., vol. 15, no. 2, pp. 1480–1490, 2016.
number of antennas,” IEEE Trans. Signal Process., 2020.
[7] R. Amiri, F. Behnia, and M. A. M. Sadr, “Exact solution for elliptic local-
[21] R. Amiri, H. Zamani, F. Behnia, and F. Marvasti, “Sparsity-aware target
ization in distributed MIMO radar systems,” IEEE Trans. Vehic. Technol.,
localization using TDOA/AOA measurements in distributed MIMO radars,”
2018.
ICT Express, vol. 2, no. 1, pp. 23–27, 2016.
[8] A. Noroozi, M. A. Sebt, S. M. Hosseini, R. Amiri, and M. M.
[22] R. Amiri, F. Behnia, and H. Zamani, “Efficient 3-D positioning using time-
Nayebi, “Closed-form solution for elliptic localization in distributed MIMO
delay and AOA measurements in MIMO radar systems,” IEEE Commun.
radar systems with minimum number of sensors,” IEEE Transactions on
Lett., 2017.
Aerospace and Electronic Systems, 2020.
[23] A. Noroozi and M. A. Sebt, “Algebraic solution for three-dimensional
[9] R. Amiri, F. Behnia, and A. Noroozi, “Efficient algebraic solution for
TDOA/AOA localisation in multiple-input–multiple-output passive radar,”
elliptic target localisation and antenna position refinement in multiple-input–
IET Radar, Sonar Navigation, vol. 12, no. 1, pp. 21–29, 2018.
multiple-output radars,” IET Radar, Sonar & Navigation, vol. 13, no. 11, pp.
[24] S. A. R. Kazemi, R. Amiri, and F. Behnia, “Efficient convex solution for 3-
2046–2054, 2019.
D localization in MIMO radars using delay and angle measurements,” IEEE
[10] H. Zamani, R. Amiri, and F. Behnia, “Compressive sensing for elliptic
Communications Letters, vol. 23, no. 12, pp. 2219–2223, 2019.
localization in MIMO radars,” in 24th Iranian Conf. Electrical Eng. (ICEE), [25] R. Amiri, S. A. R. Kazemi, F. Behnia, and A. Noroozi, “Efficient elliptic
2016, pp. 525–528. localization in the presence of antenna position uncertainties and clock
[11] H. Yang and J. Chun, “An improved algebraic solution for moving target parameter imperfections,” IEEE Transactions on Vehicular Technology,
localization in noncoherent MIMO radar systems,” IEEE Trans. Signal vol. 68, no. 10, pp. 9797–9805, 2019.
Process., vol. 64, no. 1, pp. 258–270, 2016. [26] H. W. Sorenson, Parameter Estimation: Principles and Problems. Marcel
[12] R. Amiri, F. Behnia, and M. A. M. Sadr, “Efficient positioning in MIMO Dekker, 1980.
radars with widely separated antennas,” IEEE Commun. Lett., vol. 21, no. 7, [27] S. M. Kay, Fundamentals of statistical signal processing, Estimation theory.
pp. 1569–1572, 2017. Prentice Hall, 1993.
[13] H. Song, G. Wen, L. Zhu, and D. Li, “A novel TSWLS method for [28] R. A. Horn and C. R. Johnson, Matrix Analysis. Cambridge University
moving target localization in distributed MIMO radar systems,” IEEE Press, 2013.
Communications Letters, vol. 23, no. 12, pp. 2210–2214, 2019. [29] K. C. Ho and W. Xu, “An accurate algebraic solution for moving source
[14] R. Amiri, F. Behnia, and M. A. M. Sadr, “Positioning in MIMO radars location using TDOA and FDOA measurements,” IEEE Trans. Signal
based on constrained least squares estimation,” IEEE Commun. Lett., vol. 21, Process., vol. 52, no. 9, pp. 2453–2463, 2004.
no. 10, pp. 2222–2225, 2017.

1070-9908 (c) 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Indian Institute of Technology - BHUBANESWAR. Downloaded on July 11,2020 at 04:58:08 UTC from IEEE Xplore. Restrictions apply.

You might also like