Received 3 December 2022, accepted 19 December 2022, date of publication 29 December 2022, date of current version 5 January 2023.

Digital Object Identifier 10.1109/ACCESS.2022.3233103

Waveform Design and DoA-DoD Estimation of

OFDM-LFM Signal Based on
SDFnT for MIMO Radar
JINGQI WANG , (Member, IEEE), PINGPING WANG, FAN LUO,
AND WEN WU , (Senior Member, IEEE)
Ministerial Key Laboratory of JGMT, Nanjing University of Science and Technology, Nanjing 210094, China
Corresponding author: Jingqi Wang ([email protected])

ABSTRACT In comparison to orthogonal frequency division multiplexing (OFDM), the OFDM linear
frequency modulation (OFDM-LFM) signal has similar time-bandwidth product, lower peak-to-average
power ratio, and higher sensitivity to Doppler frequency shift, making it a promising candidate waveform
for multiple-in multiple-out (MIMO) radar applications. However, the OFDM-LFM signal is unsatisfactory
in practical applications due to the challenges in waveform design and its non-stationary property, which
significantly degrades the performance of current high-resolution subspace techniques for direction of
departure (DoD) and direction of arrival (DoA) angle estimation. In this paper, we first proposed a novel
OFDM-LFM waveform design method based on the scale discrete Fresnel transform (SDFnT), which
efficiently generates orthogonal LFM signals with unlimited number of subcarriers and can be convieniently
implemented using the fast Fourier transform (FFT). Then, for angle estimation in OFDM-LFM MIMO
radars, we presented a new scale discrete Fresnel transform multiple signal classification (SDFnT-MUSIC)
method. By converting the steering vector into the chirp domain and making it non-time-vary, high resolution
angle estimation for OFDM-LFM wideband signals can be achieved. Simulation results shows that the
proposed waveform design method requires less computational time than conventional method based based
on the fractional Fourier transform (FrFT), whereas the proposed DoA-DoD estimation method outperform
the FrFT-MUSIC method in terms of computational complexity and target angle estimate performance.

INDEX TERMS MIMO radar, MUSIC, DoA, DoD, SDFnT, OFDM-LFM.

I. INTRODUCTION (LFM) signals, the OFDM-LFM signal consists of a group

In recent years, multiple-input multiple-output (MIMO) radar of mutually orthogonal LFM signals in each subcarrier that
systems have been prosperously developed due to their high can be directly fed into different antennas of the MIMO
power efficiency, great flexibility, powerful anti-interference radar system, and therefore provides the same benefits for
capability, and effective weak target detection [1], [2], [3], MIMO applications as OFDM signals. Furthermore, the
[4], [5], [6]. Orthogonal frequency division multiplexing OFDM-LFM signal outperforms OFDM signals in terms
(OFDM) signal, which is commonly used in communica- of peak-to-average power ratio (PAPR) and Doppler toler-
tion systems, has become a widely accepted waveform for ance [11], [12], making it a promising candidate waveform
MIMO radars because of its subcarriers’ mutual orthogo- for MIMO radar.
nality broad bandwidth, high range resolution, adaptability The primary challenge in waveform design for
and ease of generation of transmit signals [7], [8], [9], [10]. OFDM-LFM based MIMO radar is how to efficiently gener-
As a combination of OFDM and linear frequency modulation ate numerous orthogonal LFM signals. J. H. Kim suggested
to produce OFDM-LFM signals by inserting extra zeros
The associate editor coordinating the review of this manuscript and between the subcarriers [13], [14], which leads to a drasti-
approving it for publication was Guolong Cui . cally increase in pulse duration as the number of subcarriers

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
1348 VOLUME 11, 2023
J. Wang et al.: Waveform Design and DoA-DoD Estimation of OFDM-LFM Signal Based on SDFnT for MIMO Radar

grows. S. J. Cheng enhanced the number of OFDM-LFM propose a novel DoA-DoD (direction of departure) estimation
subcarriers [15] by combining the OFDM LFM scheme and method for OFDM-LFM-based MIMO system by combining
direct sequence spread spectrum approach, but the quantity the SDFnT method with the high-resolution multiple signal
of generated orthogonal waveforms remains insufficient. classification algorithm (MUSIC) algorithm. Since the pro-
Y. Q. Wang proposed utilizing a random matrix to modu- posed SDFnT is discrete Fourier transform (DFT) compati-
late OFDM-LFM waveforms [16] which can produce more ble, the proposed SDFnT-based waveform design and angle
than three orthogonal waveforms simultaneously but suffers estimation methods have lower computational complexity
from high sidelobes of the ambiguity function and com- than FrFT-based methods and can be efficiently implemented
plicated waveform production. A technique for generating in the existing OFDM MIMO systems. The main contribu-
an OFDM-LFM orthogonal waveform using the fractional tions of this article are summarized below.
Fourier transform (FrFT) method is described in the liter- 1) A fast OFDM-LFM waveform design method is pro-
atures [17], [18], [19]. This approach has the advantage of posed for MIMO radar applications. It is derived from
producing any desired number of subcarriers at the same time the SDFnT, a deformation of the Fresnel transform, and
while satisfying orthogonality and having good correlation can be easily implemented by multipying the scaled
characteristics, but it has the disadvantage of being computa- phase matrice and the fast Fourier transform. The effec-
tionally complex. tiveness of the generated signals is validated by ultiliz-
Another major current focus in OFDM-LFM based MIMO ing ambiguity function.
radar research is the high resolution angle estimation with 2) A novel angle estimation for OFDM-LFM wideband
LFM non-stationary signal. MIMO radars can virtually signals is presented. The SDFnT-based method com-
extend the antenna array and enlarge the aperture, allowing pensates the time-varying steering vector and converts
it to achieve precise angle estimation that requires extensive it to a non-time-varying steering vector. The MUSIC
spatial sampling. High resolution angle estimation methods method is combined further to achieve high resolution
for OFDM-based MIMO radars have generated considerable angle estimation.
recent research interest [20], [21], [22], [23]. Unfortunately, 3) A DoA-DoD estimation for bistatic OFDM-LFM-
when applied to OFDM-LFM-based MIMO radars, the per- based MIMO radar is porposed based on the aforemen-
formance of these traditional approaches degrades signifi- tioned method.
cantly. This is due to the fact that the subcarrier frequency 4) The Cramer-Rao Bound (CRB) of angle estimation of
of OFDM-LFM waveform changes with time rather than broadband signal under the uniform linear array (ULA)
being constant, and the steering vector of the OFDM-LFM- model is derived.
based MIMO radar varies with not only the incident angle 5) Numerical simulations are provided to evaluate the
but also the time. Time-frequency (TF) analysis methods, superiority of the proposed methods in terms of estima-
including wavelet transform (WT), Wigner-Ville Distribution tion accuracy and computational complexity compared
(WVD) method, short-time Fourier transform (STFT) and with previous ones.
FrFT, etc, can explore the joint time-frequency properties
of non-stationary signals by transforming the signals to the The division of this article is as follows. Section II of this
time-frequency domain [24], [25], [26], [27], and can thus be article covers the design of the OFDM-LFM-based MIMO
combined with traditional array signal processing methods to orthogonal waveform, along with its computational com-
achieve high resolution angle estimation with LFM-like non- plexity and ambiguity function analysis.Section III discusses
stationary signals. Hurstman et al. proposed a time-frequency about the invalidation of the conventional MUSIC algorithm
coherent signal subspace technique based on the WVD for of non-stationary OFDM-LFM signals, and proposes a novel
LFM-like nonstationary signals [28], whereas the cross-term high-resolution angle estimation method by combining the
interference severely degrades the performance of direction SDFnT method with the MUSIC algorithm. Section IV gives
of arrival (DoA) estimate for multiple targets. While the the simulation results. Finally, section V outlines the conclu-
STFT can avoid the cross-term interference, improper win- sion of the study.
dow selection can readily result in a reduction in estimation Throughout this article, the boldface italic letters are used
accuracy. Due to its flexible time-frequency focusing prop- to represent vectors, matrices, and tensors. The Kronecker
erty for non-stationary signals and high robustness against product of A and B is set as A ⊗ B. The notation diag (·)
cross-term interference, FrFT was used for DoA estimation of represhents a diagonal matrix. I L is a L × L identity matrix,
multi-component LFM signals in [30], however, it still suffers E [·] denotes the expectation operator. mod means take the
from its computational cost. remainder, (·)∗ means conjugation, (·)T stands for transpose,
In this paper, we first propose an efficient OFDM-LFM tr(·) represents the trace of the matrix.
waveform design method based on a novel scale discrete
Fresnel transform (SDFnT), whose kernel function is a set II. MIMO ORTHOGONAL WAVEFORM DESIGN
of orthogonal LFM with adjustable chirp rate. In order In this section, an efficient OFDM-LFM waveform design
to achieve high-precision estimation of transceiver angle method is presented. Firstly, a scale discrete Fresnel trans-
for MIMO radars using non-stationary signals, we further form developed from the discrete Fresnel transform (DFnT),

VOLUME 11, 2023 1349

 J. Wang et al.: Waveform Design and DoA-DoD Estimation of OFDM-LFM Signal Based on SDFnT for MIMO Radar

an integral transformation comes from classical optics [33],

[34], [35], [36], is proposed, and then a novel OFDM-LFM
waveform design method based on it is presented for MIMO
radar.

A. SDFnT-BASED OFDM-LFM WAVEFORM DESIGN

The kernel functions of the proposed SDFnT are a set of
orthogonal chirp signals. Unlike the conventional DFnT with
fixed chirp rate, the chirp rate of the SDFnT kernel fucntion is
adjustable and controlled by a non-zero scale transform fac-
tor. By varying the scale transformation factor, more types of
orthogonal waveforms as well as improved interception resis-
tance and radar detection performance can be developed.The
n-th chirp signal of the proposed OFDM-LFM waveform is FIGURE 1. α = 0.25j , 3D ambiguity fuction of OFDM-LFM.
defined as
 2
N T
j π4 −jπ T 2 t·α−n N α
ψn (t) = e e (1)
where T is the duration of the chirp basis signal, N is the
number of chirps, namely the number of subcarriers, and α is
the scale transformation factor.
The discrete expression of OFDM-LFM is as follows:
1 π XN −1
s(n) = √ ej 4 x (k)
N k=0

 e−j Nπ ( mα −nα )
2

N ≡ 0 (mod2)
× π m

1
2 (2)
−j −nα+
e N α 2
N ≡ 1 (mod2)


in this equation, m and n stand for the numbers of sample

FIGURE 2. α = 4j , 3D ambiguity fuction of OFDM-LFM.
points in the time and frequency domains, As can be easily
seen from the aforementioned equation, the time-frequency
characteristics of this OFDM-LFM signal consist of a col-
lection of diagonal lines that are mutually parallel and have
a controlled chirp rate. Let the bandwidth of th signal be B,
then the chirp spacing is 1f = B/N , the initial frequency
and tuning slope of the n-th chirp is f0 = n1f and B α 2 T


respectively. This collection of chirp signals has varied initial

frequencies that are proportional to the number of subcar-
riers, but they all have the same tuning slope. It behaves
as a series of mutually parallel diagonal lines in the time-
frequency plane, similar to LFM signals with step-frequency,
thus increasing spectrum efficiency. The chirp rate of this
collection of chirp signals varies with the transformation
factor α.
As shown by the following, the chirp signals in equation 1 FIGURE 3. α = 0.25j , top view of OFDM-LFM ambiguity fuction.
are mutually orthogonal,
Z
ψp∗ (t)ψq (t)dt We define the (m, n)th element of the N by N SDFnT
Z T  2  2 matrix as
jπ N t − pT α −jπ N t − pT α
= e T2 α N e T2 α N dt
 ej N ( α −nα )
 π m 2
π
(0 π 2 2
2 e−j 4 N ≡ 0 (mod2)
Tej N ρ −q α p = q 8α (m, n) = √ × π m 1
 2 (4)
N j + −nα
N ≡ 1 (mod2)
= (3)  N α 2
e
0 p 6= q
where p and q represent the p-th sub-carrier and the q-th sub- In order to comply with the practical OFDM system, N
carrier, respectively. is set to be an even number, and we can thus simplify the

1350 VOLUME 11, 2023

J. Wang et al.: Waveform Design and DoA-DoD Estimation of OFDM-LFM Signal Based on SDFnT for MIMO Radar

Clearly, 8α (m, n) can be represented in matrix form as

8α = 21 F22 (7)

where

21 = diag(θ 1 (m)) ∈ C N ×N , m = 0, 1, . . . , N − 1 (8)

22 = diag(θ 2 (n)) ∈ C N ×N , n = 0, 1, . . . , N − 1 (9)

where θ 1 (m) = ej 4 ej N ( α ) , m = 1, 2, · · · M − 1, θ 2 (n) =

π π m 2

π 2
ej N nα , n = 1, 2, · · · N − 1.
Equation 7 makes it apparent that the SDFnT is
DFT-compatible because it can be implemented using
DFT and phase matrix multiplications. as the result,the
OFDM-LFM waveform developed in this paper is com-
FIGURE 4. α = 4j , top view of OFDM-LFM ambiguity fuction.
patible with current OFDM systems while having a lower
peak-to-average ratio than the OFDM waveform, making
the SDFnT-based waveform design method presented in this
paper more practical in waveform synthesis than the method
proposed in the literatures [17], [18], [19]. Additionally,
because the SDFnT has an inverse transform, it is suitable
for modulation and demodulation in communication systems,
as well as joint radar and communication (JRC) systems.

B. ANALYSIS OF COMPUTATIONAL
In this subsection, we compare the computational complexity
of the SDFnT and FrFT in terms of time complexity, which
is the trend of code execution time as data size changes
and typically represented by a large O representation. The
execution time of algorithm is given as
FIGURE 5. Comparison of velocity ambiguity fuctions of different
orthogonal waveforms. T [n] = O(f (n)) (10)

where T [n] denotes the code execution time, n is the size of

the data scale, and f (n) denotes the total of the number of
times each line of code is run.
As is well known, the N points discrete Fourier transform
(DFT) algorithm requires N 2 complex multiplication opera-
tions and N (N − 1) complex additional operations, therefore
the time complexity of the DFT algorithm is O(N 2 ). In con-
trast, the fast Fourier transform (FFT) algorithm, an effective
and practical DFT algorithm with the basic operation mod-
ules as the butterfly operation with radix-2, requires N2 ×logN2
complex multiplication operations and N × logN 2 complex
addition operations, respectively [31]. As a result, the FFT
algorithm’s time complexity is reduced to O(N × logN 2 ).
FIGURE 6. Comparison of range ambiguity fuctions of different
As described in Section II-A, the proposed SDFnT algorithm
orthogonal waveforms. is DFT-compatible, then it can be effectively implemented
by FFT operation and two additional phase rotations. For
definition of SDFnT matrix 8α ∈ CN × N as the length of N , the time complexity of FrFT algorithm is
1 T (N ) = O(N 2 ) [32], whereas the time complexity of SDFnT
8α (m, n) = √ e−j 4 × ej N ( α ) e−j N mn ej N (nα)
π π m 2 2π π 2
(5)
N is T (N ) = O(N × logN 2 ). Evidently, SDFnT has a lower
The expression of the classical Fourier transform is computational complexity than FrFT.
1 2π
F (m, n) = √ e−j N mn (6) C. ANALYSIS OF RADAR AMBIGUITY FUCTION
N One of the crucial performance indicators for radar waveform
F is the discrete Fourier transform matrix. analysis is the ambiguity function (AF), which can accurately

VOLUME 11, 2023 1351

 J. Wang et al.: Waveform Design and DoA-DoD Estimation of OFDM-LFM Signal Based on SDFnT for MIMO Radar

assess the autocorrelation performance of radar signals, para-

metric traits, as well as Doppler shift and distance ambiguity.
The performance analysis of the proposed OFDM-LFM sig-
nal is provided in this part from the viewpoint of the radar
ambiguity fuction.
Figure 1 and Figure 3 show the ambiguity function of the
SDFnT-based OFDM-LFM signal with the scale transforma-
tion factor value of 0.25j in 3D view and in top view respec-
tively, while Figure 2 and Figure 4 demostrate the AF with
the value of 4j. These four figures can be compared to show
that as the value of the scale transformation factor increases,
the modulation slope of OFDM-LFM declines, the velocity
FIGURE 7. Bistatic MIMO radar model.
resolution of the OFDM-LFM radar remains constant, and the
range resolution decreases.
Figure 5 compares the zero Doppler cuts of the ambi- radar with Mt transmitting antenna and Mr receiving antenna.
guity functions of OFDM, OCDM [33], and the proposed The array model satisfies the following conditions:
OFDM-LFM signal with different scale transformation fac- (1) the sources meet the requirement for a far-field, they are
tors, while Figure 6 compares the zero time delay cuts. uncorrelated with one another, and more virtual array
Here, the OCDM is analyzed as a special example of the members are present than desired;
proposed SDFnT-based OFDM-LFM signal with the value of (2) the receiving array is an ULA with a less than half-
scale transformation factor set to 1. The velocity ambiguity wavelength array spacing;
functions of the four signals are nearly identical, as shown in (3) zero-mean Gaussian white noise (AWGN) makes up
Figure 5, indicating that they have similar velocity resolution. the noise signal;
The range ambiguity function in Figure 6 reveals that the The array signal’s received data model can be obtained as
main lobe of an OFDM signal is narrower than that of an
OCDM signal. Furthermore, when |α| < 1, as in the curve X (t) = AS (t) + N (t) (11)
corresponding to α = 0.25j, the main lobe of the proposed
where A = [a (θ1 ) , a (θ2 ) , · · · , a (θK )]T is M ×K dimension
signal is wider than OCDM and the side lobe is higher; when
steering vector, s(t) if K × 1 dimension vector of wideband
|α| > 1, as in the curve corresponding to α = 4j, the main
nonstationary source waveforms, N(t) is M × 1 dimension
lobe becomes the narrowest and the side lobe becomes the
vector of zero-mean Gaussian white noise. M is the number
lowest of the four simulated signals.
of receiving virtual array elements, while K is the number of
snapshots. θK is the incident angle of the K -th signal. a (θk ) =
III. ANGLE ESTIMATION BASED ON SDFnT T
This section’s major topic is the bistatic MIMO radar multi- 1, e−jϕ2k , · · · , e−jϕMk , k = 1, 2, · · · , K is the steering
target estimatin problem for the DoD and DoA. A detailed vector for the k-th signal. Under the condition of narrowband
description of the conventional MUSIC algorithm, a high- stationary, the phase term ϕmk = 2π (m − 1) d sin θk /λ in
resolution DoA estimation approach [37], as well as the the steering vector is only related to the angle θk , and is
issues that can arise when using the MUSIC algorithm to independent of time t, λ stands for wavelength and d is the
estimate DoA for non-stationary signals is presented. We first receiving array element spacing.
convert the received wideband OFDM-LFM signals from The core process in the MUSIC algorithm is to build the
the time domain into the chirp domain by using the SDFnT signal’s covariance matrix and perform eigen decomposition.
algorithm. Next, we establish the steering vector in the chirp The eigen vector is then split into signal and noise subspaces.
domain, which won’t change with the frequency and will Finally, the MUSIC spectrum is built and the high resolution
have the same characteristics as the steering vector of the DoA estimation is achieved by spectral peak searching. How-
narrowband signal if the appropriate scale factor is applied. ever, the classic MUSIC algorithm only works with narrow-
As a result, the conventional high resolution algorithm can band signals and is ineffective with non-stationary signals like
be used to estimate the transmitting and receiving angles for OFDM-LFM signals.
non-stationary OFDM-LFM signals. Assuming that there are k non-stationary signals incident
on the above uniform linear array in the far field, the output
A. MIMO RADAR SYSTEM MODEL signal of the m-th array element is
The bistatic MIMO radar model is depicted in Figure 7. Each K
X
transmitting array element transmits a single OFDM-LFM x m (t) = sk (t − τkm ) + nm (t) (12)
subcarrier, and each subcarrier, which resembles a series k=1
of frequency-stepped LFM signals, has a distinct beginning  
j2π k·1f ·t+ 1 µ t 2
2 k
frequency and the same tuning slope. We can synthesize a where sk (t) = e is the k-th incident non-
virtual uniform linear array of Mt ∗ Mr using bistatic MIMO stationary signal, fk = k·1f is the initial frequency of the k-th

1352 VOLUME 11, 2023

J. Wang et al.: Waveform Design and DoA-DoD Estimation of OFDM-LFM Signal Based on SDFnT for MIMO Radar

subcarrier of OFDM-LFM signal and 1f is the frequency α and the spectral peak position u of Skm (α, u), as shown in
separation between the two signals, µk is the frequency mod- (18) and (19), respectively:
ulation slope, nm (t) is the noise signal of the m-th element. s
Substituting the above equation into X (t) = AS (t) + N (t) 1
and simplifying, the time-varying direction vector of the k-th α = αkm = − (18)
µ · tb2 · L
incident signal on M receiving elements can be obtained as
u = umk = fk − µτk · L · tb
m


follows: (19)


1 2 M 2
ak (t) = 1, e−j2πfk tτk ejπ µk τk , · · · , e−j2πfk tτk ejπ µk τk
1 M


As illustrated in (15) and (18), the scale transform factor is
fixed while the spectral peak location changes for each array.
(13) By substituting (16) into (19) and simplifying it as follows,
one can determine the relationship between the spectral peak
where τkm = (m − 1) d sin θk /c is the delay of the k-th signal
location of the m-th receiving array element and that of the
on the m-th receiving array element, c is the speed in the
reference array element:
medium where the received signal is located. It is evident
from (13) that this direction vector is related to both time τkm
1
and the angle of the incident signal; hence, the classic DoA um
k = uk + (20)
α · tb
estimation method cannot be used.
Then, compare the chirp domain projections of the refer-
B. SIGNAL MODEL ence array element and the m-th array element:
In this subsection, we proposed a novel SDFnT-based DoA
π jπ ( k)
2
u1
estimation method for OFDM-LFM signal. The chirp domain  
signal after SDFnT transformation can be expressed as S1k αk , u1k = e−j 4 e L α2 T (21)

π (uk )
m 2
N −1 −j π4 m j L α2
Sm α , m
τ


j π4 −j Nπ ( m
)
2 u = e B e T (22)
X
X α (n) = x(n)e e α −nα k k k k k

n=0
N −1
Substituting (20) into (22), we get
2 2
jπn2 µ− αN
 
πm
j π4 jϕ −j N α 2 j2πn(f0 + Nm )
X
= Ae e e e e (14) τkm
!2
u1k + α·tb
n=0
−j π4

jπ
k αk , uk = e
Sm Bk τkm e L
m


α2 T (23)
The frequency modulation slope of the n-th subcarrier is
µ = B/α 2 T , the bandwidth is B, and the center frequency
is f0 = n1f . Firstly, the k-th LFM signal s1k (t) received Then, substituting (16) into the above formulas and simpli-
on the reference array element (zero delay) is discretized fying it, we get
and SDFnT is performed to obtain the projection on the
k αk , uk
Sm m


chirp domain Sk1 (α, u). Once the scale transformation factor
α approaches the proper value, an energy-concentrated peak τk m
jπ k
(τ m )2 π jπ (u1k )
2

Bk τkm ej2πfk α e L (α·tb)2 e−j 4 e L

= α2 T
would appear in the chirp domain. The relationship between
the scale transformation factor α and the frequency modula-
 
Ak τkm · S1k αk , u1k


= (24)
tion slope µ is

Since (τkm )2  1, the ej L (τk ) /(α·tb) term can be ignored.

π m 2
s 2
1
α = αk1 = − (15)
µ · L · tb2 Finally, the expression of the direction vector of the k-th
incident signal in the chirp domain after SDFnT is
where L is the number of fast sampling points in the time
h  
i
domain, tb is the sampling interval, and the spectral peak Ak = 1, Ak τk2 , · · · , Ak τkm (25)
position is
τkm
Ak τkm = Bk τkm ej2πfk α



u= u1k = fk · L · tb (16) (26)

The k-th incident signal received on the m-th element Regarding the angle of the k-th incident signal only, this
(when non-reference element delay is not zero) is vector is time-invariant under far-field conditions and exclu-
sively related to the time-delay term τkm . Instead of the con-
m jπ 2(fk −µk τkm )t+µk t 2

 
k (t) = sk t − τk = Bk τk e
sm m


(17) ventional array signal model, this SDFnT-based signal model
enables the transformation of the time-varying direction
jπ µk (τkm ) −2fk τkm
h i
2
where Bk τkm = e


is a constant term matrix of non-stationary signals that resemble OFDM-LFM
related only to delay and incident angle. Similarly, SDFnT into a fixed direction matrix, allowing DoA estimation by
transform is performed on smk (t) to obtain the rotation angle employing the conventional high-resolution methods.

VOLUME 11, 2023 1353

 J. Wang et al.: Waveform Design and DoA-DoD Estimation of OFDM-LFM Signal Based on SDFnT for MIMO Radar

Algorithm 1 DoA and DoD Estimation of OFDM-LFM Sig- among them, σs2 and σn2 are the signal power spectrum and
nal Based on SDFnT-MUSIC Method noise power spectrum, respectively, s and a are signals and
Input: the signals x(t) received by the coprime array; their steering vectors, I L and I M are the identity matrice, M
Output: the DoA and DoD estimation of the wideband is the number of receiving array elements, and L is the number
θk , b
signal (b ϕk ); of snapshots in the time domain. The received data covariance
1: Initialize: Perform SDFnT on the signal of the reference matrix is
array element, then perform a spectral peak search (αk , u) to  
get the number of incident signals; R = Rs + Rn = σs2 aaH + R1 ⊗ I L . (29)
2: SDFnT transform with a scaling factor of αk for the
received signals on all array elements in order to get the
The inverse matrix is
k and peak point data X k αk , uk
m
spectral peak location um m !
R−1 H −1
1 aa R1
of the k-th incident signal on the m-th array element; R −1
= R−1
1 − ⊗ IL
σs−2 + aH R−1
1 a
3: Rk = X k X H k the related matrix of the constructed chirp
σs−2 aaH
 
domain α = R−1 ⊗ IL (30)
1 −
4: Perform eigenvalue decomposition on Rk to obtain SNR2 + M · SNR−1
the signal subspace matrix U s , and construct the spectral
function f (θ, ϕ); where the signal to noise ratio is SNR , σs2 /σn2 . According
5: By traversing the transmitting angle θ and receiving angle to [38], The likelihood function of the signal is
ϕ and searching for the peaks of the spectrum f (θ, ϕ), find  −1 H −1
the incidence angle estimation of the K LFM signals. F1 (r) = π M |R| e−r R r . (31)

For the unbiased estimator theta, its CRB is given by (32)

−1
∂ lnF1 (r)
  2
−1
CRB = J = −E (32)
C. ALGORITHM FLOW ∂θ 2
The proposed SDFnT-based DoA and DoD estimation where the J matrix is called Fisher Information Matrix (FIM).
emthod for OFDM-LFM signal is brieflly summarized in Substituting (30) and (31) into Formula (32), we obtain
Algorithm 1.
The proposed method, like the FrFT-based method, can CRB
estimate the number of targets but cannot identify multiple

∂2 σs
 −2  −1
H H H −1
targets when they are very close. = − L 2E · r aa − r R r
∂θ SNR−1 1

Lσs−2 ∂2  H   H 
  
D. ANALYSIS OF COMPUTATIONAL =− · tr aa E rr
SNR−2 + M · SNR−1 ∂θ 2
We analyze the computational complexity of the pro- (33)
posed wideband signal angle estimation algorithm. For
the propose 2D-MUSIC algorithm, time-frequency sur- In the case of only one incident angle, τs,t , τs − τt =
face transformation requires O PL logL2 , calculating the 4s,t

c is defined as the delay difference between the two array
covariance matrix needs O L(MN )2 , eigenvalue decom- elements.
3

position
n ◦ requires O (MN ) o, 2D angle finding costs
"
ω2 0s,t
2
#
∂ 2 −jωτs,t ω4s,t −jωτs,t
O 1 [(MN + 1) (MN − K )] , so the proposed algorithm’s
90 Ds,t = 2 e
∂θ
= − 2 −j
c c
e (34)
overall computational complexity is O{PL logL2 +L(MN )2 +
◦
among them 0s,t = rs cos(θ + ϕs ) − rt cos(θ + ϕt ), then we
(MN )3 + 90 1 [(MN + 1) (MN − K )]} where P is the number
of time-frequency surface transformations, K is the number get
of targets, L is the number of snapshots, 1 is the angle search M −1 M −1
X X ω2 0s,t
2
ω4s,t
step value, N and M are the number of transmitting and tr(F) = σs2 [− −j ]
receiving arrays, respectively. c2 c
s=0 t=0
M −1 M −1
σs2 ω2 X X 2
E. CRB FOR WIDEBAND DoA ESTIMATION = 0s,t . (35)
c2
The signal covariance matrix and noise covariance matrix s=0 t=0
for the received signal with unknown incident angle θ are, For the wideband signal, we divide it into J narrowband
respectively regions, and then the wideband CRB expression of far-field
h i direction finding can be obtained as
Rs = E (a ⊗ s) (a ⊗ s)H = σs2 aaH ⊗ I L (27) M −1 M −1
Mc2 X X 2 −1 −1
h i CRB = ( 0s,t ) · P (36)
Rm = E nnH = σn2 I M ⊗ I L = R1 ⊗ I L (28) 4Lπ 2
s=0 t=0

1354 VOLUME 11, 2023

J. Wang et al.: Waveform Design and DoA-DoD Estimation of OFDM-LFM Signal Based on SDFnT for MIMO Radar

TABLE 1. computational time of FrFT and SDFnT.

PJ Mfi2 ·SNR2i
P= i=0 1+M ·SNRi , fi represents the center frequency of the
i-th divided frequency band. SNRi is the frequency domain
signal to noise ratio on the i-th narrowband.
3c2
CRB = (37)
2LM (M 2 − 1)π 2 d 2 cos2 θ · P

IV. SIMULATION AND ANALYSIS FIGURE 8. The computational time of SDFnT methods and FrFT method
In this section, the computation time of the proposed SDFnT- for traversing snapshot numbers.

based waveform design method is firstly analyzed. Then,

in order to demonstrate the performance of the SDFnT-based
angle estiamtioin method for non-stationary OFDM-LFM
signals, comparisons between the proposed method and the
FrFT-based target angle estimation method reported in the lit-
erature [39] in terms of performance of DoA estimation, and
DoA-DoD estimation are presented. We consider a MIMO
radar system with 2 transmitting arrays and 4 receiving
arrays. The bandwidth of the OFDM-LFM waveform is 100
MHz and the duration is 2.56 us. The number of snapshot is
256 and the SNR is set to 10 dB. The simulation environment
is based on MATLAB 2019a, on a computer with an Intel
i5-10210U CPU Core.

A. COMPUTATION TIME OF WAVEFORM DESIGN

FIGURE 9. The computational time of SDFnT methods and FrFT method
The time required to generate OFDM-LFM orthogonal wave- for traversing search times.
forms using the SDFnT method and the FrFT method is
shown in Table 1. N stands for the number of subcarriers in
OFDM-LFM, and it ranges from 256 to 1792 in steps of 512.
The time results are averaged after 500 Monte Carlo simu-
lations. Table 1 shows that the computational time needed to
produce orthogonal waveforms using the proposed SDFnT
approach is shorter than that needed to produce them using
the FrFT method for the same number of subcarriers, and that
this time savings grows as the subcarrier count increases.

B. PERFORMANCE OF DoA ESTIMATION

In order to compare the computational times of FrFT method
and proposed SDFnT method with single and multiple tar-
gets, figures 8 and 9 respectively show the computed tim-
ings for various numbers of snapshots K and time-frequency
surface traversal times α. It is evident that the proposed FIGURE 10. RMSE of SDFnT methods and FrFT method of Traversing SNR.

SDFnT-based method requires less complicated processing

and can cut the computational time by roughly 20%. the target angles when the SNR is less than 10 dB. This is
To analyze the accuracy of the proposed method with because, the peak value following signal transformation to the
respect to angle with various signal qualities, the SNR is chirp domain is totally hidden by the noise signal. However,
set to vary from -15 dB to 10 dB with an interval of 1 dB, because the FrFT method computes the trigonometric func-
and θ1 = 15◦ and θ2 = 20◦ . After 500 Monte Carlo sim- tion numerous times, which demands more accuracy, it has a
ulations, the RMSEs of the target parameters are shown as little lower angle estimation accuracy than the novel SDFnT
in Fig. 10. It can be seen that neither method can estimate based method.

VOLUME 11, 2023 1355

 J. Wang et al.: Waveform Design and DoA-DoD Estimation of OFDM-LFM Signal Based on SDFnT for MIMO Radar

FIGURE 13. Mulit-objective SDFnT-MUSIC spatial spectrum traversing

FIGURE 11. Multi-targets MUSIC spatial spectrum. array elements.

FIGURE 14. SDFnT-MUSIC spatial spectrum of ergodic snapshots.

FIGURE 12. Single-objective SDFnT-MUSIC spatial spectrum traversing
array elements.

Figure 11 shows the MUSIC spatial spectrum, where the

angle θ is searched from −90◦ to −90◦ and 1θ = 0.01◦
represents the angular step value. The real incident angles for
the two targets are θ1 = 15.2◦ and θ2 = 18.3◦ . It is not
difficult to find that the proposed SDFnT-based method has
high angle estimate accuracy even when the two targets are
quite near to one another.
Figures 12 and 13 depict how the number of array elements
affects the precision of single- and multi-target angle esti-
mates. The findings demonstrate that the accuracy of angle
estimate is strongly influenced by the number of array ele-
ments, and that target angle estimation performance greatly
FIGURE 15. 2D-MUSIC algorithm based on SDFnT.
improves with increasing array element number.
Figure 14 depicts how the estimating performance is
affected by the number of snapshots. Naturally, increasing C. PERFORMANCE OF DoA-DoD ESTIMATION
the number of snapshots can also increase the accuracy of the The bistatic MIMO radar can detect the target even in the
angle estimation, but by comparing the cases of the number of absence of additional information like target range accord-
snapshots K = 128 and K = 256, we can see that the increase ing to the DoD and DoA information. This section uses
in snapshots does not significantly improve performance. the SDFnT method and the 2D-MUSIC algorithm to jointly
However, we already know from Fig. 8 that the increase in estimate the transceiver angle for non-stationary signals.
snapshots increases the computation time significantly, so we Figures 15 and 16 display the results of the 2D-MUSIC
need to choose the appropriate number of snapshots to get angle estimation based on the SDFnT method and its top
better performance and less computation time. view, demonstrating how the method can jointly estimate

1356 VOLUME 11, 2023

J. Wang et al.: Waveform Design and DoA-DoD Estimation of OFDM-LFM Signal Based on SDFnT for MIMO Radar

proposed SDFnT method consistently achieves high accuracy

DoD and DoA estimations in the majority of the considered
SNR region (SNR > −10 dB) and the performance of
angle estimation using the proposed approach in this paper
is marginally superior to that using the FrFT method.

V. CONCLUSION
This paper first proposed an efficient OFDM-LFM waveform
design method, which can efficiently generates orthogonal
LFM signals with unlimited number of subcarrier, based
on a novel scale discrete Fresnel transform with the kernel
function as a set of orthogonal LFM with adjustable chirp
rate. Then we further proposed a novel angle estimation
method for OFDM-LFM-based MIMO systems by combin-
ing the SDFnT method with the high-resolution MUSIC algo-
FIGURE 16. The overlooking function of 2D-MUSIC algorithm based on rithm. This approach turns the time-varying steering vector
SDFnT.
into a non-time-varying steering vector in the chirp domain,
allowing the classic subspace algorithm to be used for high-
resolution angle estimation. Simulation results show that the
proposed methods have lower computational complexity and
better estimate performance than the FrFT-based approaches.
In addition, because of their compatibility with the FFT, the
proposed SDFnT-based methods have the extra benefit of
being simple to implement in the existing OFDM systems.
Furthermore, since the SDFnT has an inverse transform, it can
be utilized for modulation and demodulation in multicarrier
communication systems, as well as the development of inno-
vative joint radar and communication systems in future.

REFERENCES
[1] D. W. Bliss and K. W. Forsythe, ‘‘Multiple-input multiple-output (MIMO)
radar and imaging: Degrees of freedom and resolution,’’ in Proc. 37th
FIGURE 17. The RMSE of DoD when traversing SNR based on
Asilomar Conf. Signals, Syst. Comput., Nov. 2003, pp. 54–59.
SDFnT-MUSIC method.
[2] J. Li and S. Petre, ‘‘MIMO radar signal processing,’’ Tech. Rep., 2009,
pp. 1–64.
[3] T. D. Backes, ‘‘Parameter identifiability in a phased-subarray MIMO
radar,’’ in Proc. IEEE Aerosp. Conf., Mar. 2014, pp. 1–6.
[4] Y. Petillot, C. Du, and J. S. Thompson, ‘‘Predicted detection performance
of MIMO radar,’’ IEEE Signal Process. Lett., vol. 15, pp. 83–86, 2008.
[5] K. W. Forsythe, D. W. Bliss, and G. S. Fawcett, ‘‘Multiple-input multiple-
output (MIMO) radar: Performance issues,’’ in Proc. Conf. Rec. 38th
Asilomar Conf. Signals, Syst. Comput., Nov. 2004, pp. 310–315.
[6] F. Meinl, M. Stolz, M. Kunert, and H. Blume, ‘‘An experimental high
performance radar system for highly automated driving,’’ in IEEE MTT-
S Int. Microw. Symp. Dig., Mar. 2017, pp. 71–74.
[7] J. Suh, J. Lee, G.-T. Gil, and S. Hong, ‘‘Time-and-frequency hybrid
multiplexing for flexible ambiguity controls of DFT-coded MIMO OFDM
radar,’’ IEEE Access, vol. 9, pp. 137793–137808, 2021.
[8] N. Levenon, ‘‘Multi frequency complementary phase-coded radar signal,’’
ET Radar Sonar Navig., vol. 147, no. 6, pp. 272–284, Dec. 2000.
[9] B. J. Donnet and I. D. Longstaff, ‘‘Combining MIMO radar with OFDM
communications,’’ in Proc. Eur. Radar Conf., Manchester, U.K., Sep. 2006,
pp. 37–40.
FIGURE 18. The RMSE of DoA when traversing SNR based on [10] Y.-H. Cao, X.-G. Xia, and S.-H. Wang, ‘‘IRCI free colocated MIMO radar
SDFnT-MUSIC method. based on sufficient cyclic prefix OFDM waveforms,’’ IEEE Trans. Aerosp.
Electron. Syst., vol. 51, no. 3, pp. 2107–2120, Jul. 2015.
[11] S. H. Wang, C. P. Li, K. C. Lee, and H. J. Su, ‘‘A novel low-complexity pre-
the transceiver angle of non-stationary signals while also coded OFDM system with reduced PAPR,’’ IEEE Trans. Signal Process.,
estimating the target number when the number of sources vol. 63, no. 6, pp. 1366–1376, Mar. 2015.
is unknown. The RMSEs of the DoD and DoA estimates in [12] F. A. P. de Figueiredo, N. F. T. Aniceto, J. Seki, I. Moerman, and
G. Fraidenraich, ‘‘Comparing f-OFDM and OFDM performance for
terms of SNR are shown in Fig. 17 and Fig. 18, where φ1 and MIMO systems considering a 5G scenario,’’ in Proc. IEEE 2nd 5G World
φ2 represent DoD, θ1 and θ2 represent DoA. As expected, the Forum (5GWF), Sep. 2019, pp. 532–535.

VOLUME 11, 2023 1357

 J. Wang et al.: Waveform Design and DoA-DoD Estimation of OFDM-LFM Signal Based on SDFnT for MIMO Radar

[13] J.-H. Kim, M. Younis, A. Moreira, and W. Wiesbeck, ‘‘A novel OFDM [37] R. O. Schmidt, ‘‘Multiple emitter location and signal parameter estima-
chirp waveform scheme for use of multiple transmitters in SAR,’’ IEEE tion,’’ IEEE Trans. Antennas Propag., vol. AP-34, no. 3, pp. 276–280,
Geosci. Remote Sens. Lett., vol. 10, no. 3, pp. 568–572, May 2013. Mar. 1986.
[14] J. H. Kim, M. Younis, A. Moreira, and W. Wiesbeck, ‘‘Spaceborne MIMO [38] X. Zeng, Z. Li, H. Huang, and G. Sun, ‘‘Focused passive synthetic aper-
synthetic aperture radar for multimodal operation,’’ IEEE Trans. Geosci. ture,’’ in Proc. IEEE Youth Conf. Inf., Comput. Telecommun., Nov. 2010,
Remote Sens., vol. 53, no. 5, pp. 2453–2466, May 2015. pp. 158–161.
[15] S.-J. Cheng, W.-Q. Wang, and H.-Z. Shao, ‘‘Spread spectrum-coded [39] C. Fang, H. Zishu, L. Hongming, and L. Jun, ‘‘The parameter setting
OFDM chirp waveform diversity design,’’ IEEE Sensors J., vol. 15, no. 10, problem of signal OFDM-LFM for MIMO radar,’’ in Proc. Int. Conf.
pp. 5694–5700, Oct. 2015. Commun., Circuits Syst., May 2008, pp. 876–880.
[16] W.-Q. Wang, ‘‘MIMO SAR OFDM chirp waveform diversity design with
random matrix modulation,’’ IEEE Trans. Geosci. Remote Sens., vol. 53,
no. 3, pp. 1615–1625, Mar. 2015.
[17] J. Shang, Z. Wang, L. Zhou, and L. Zhao, ‘‘A joint time-frequency syn-
JINGQI WANG (Member, IEEE) received the
chronization algorithm for FrFT-OFDM systems and its hardware imple-
B.E. degree in communication engineering from
mentation,’’ in Proc. 15th IEEE Int. Conf. Commun. Technol., Nov. 2013,
pp. 190–195. the Nanjing University of Aeronautics and Astro-
[18] K. C. K. Chen, Y. L. Y. Liu, and W. Z. W. Zhang, ‘‘Study on integrated nautics, Nanjing, China, in 2003, and the M.E.
radar-communication signal of OFDM-LFM based on FRFT,’’ in Proc. IET and Ph.D. degrees in electromagnetic field and
Int. Radar Conf., 2015, pp. 1–6. microwave technology from Southeast University,
[19] A. R. Nafchi, F. Sanz Estevez, E. Hamke, and R. Jordan, ‘‘Applying the Nanjing, in 2006 and 2011, respectively.
FRFT to an OFDM system for Li-Fi: A design experiment for peace She is currently a Lecturer with the School of
engineering education,’’ in Proc. World Eng. Educ. Forum Global Eng. Electronic and Optical Engineering, Nanjing Uni-
Deans Council (WEEF-GEDC), Nov. 2018, pp. 1–4. versity of Science and Technology, Nanjing. Her
[20] M. D. Larsen, A. L. Swindlehurst, and T. Svantesson, ‘‘Performance research interests include the signal processing for multicarrier radar, wave-
bounds for MIMO-OFDM channel estimation,’’ IEEE Trans. Signal Pro- form design for joint radar and communication systems, and linearization
cess., vol. 57, no. 5, pp. 1901–1916, May 2009. technologies for power amplifier.
[21] R. Shafin, L. Liu, J. Zhang, and Y.-C. Wu, ‘‘DoA estimation and capac-
ity analysis for 3-D millimeter wave massive-MIMO/FD-MIMO OFDM
systems,’’ IEEE Trans. Wireless Commun., vol. 15, no. 10, pp. 6963–6978,
Oct. 2016.
PINGPING WANG received the B.S. degree in
[22] Y. C. Lin, T. S. Lee, Y. H. Pan, and K. H. Lin, ‘‘Low-complexity high-
resolution parameter estimation for automotive MIMO radars,’’ IEEE
electronic information engineering from the Uni-
Access, vol. 8, pp. 16127–16138, 2020. versity of South China, Hunan, China, in 2020.
[23] D. Werbunat, B. Meinecke, B. Schweizer, J. Hasch, and C. Waldschmidt, She is currently pursuing the M.S. degree in elec-
‘‘OFDM-based radar network providing phase coherent DOA estimation,’’ tromagnetic field and microwave technology with
IEEE Trans. Microw. Theory Techn., vol. 69, no. 1, pp. 325–336, Jan. 2021. the Nanjing University of Science and Technol-
[24] F. Zhang, G. Bi, and Y. Q. Chen, ‘‘Tomography time-frequency transform,’’ ogy, Nanjing, China. Her current research interests
IEEE Trans. Signal Process., vol. 50, no. 6, pp. 1289–1297, Jun. 2002. include radar signal processing and joint radar and
[25] P. Rao and F. J. Taylor, ‘‘Estimation of instantaneous frequency using the communication.
discrete Wigner distribution,’’ Electron. Lett., vol. 4, no. 26, pp. 246–248,
1990.
[26] R. N. Czerwinski and D. L. Jones, ‘‘Adaptive short-time Fourier analysis,’’
IEEE Signal Process. Lett., vol. 4, no. 2, pp. 42–45, Feb. 1997.
[27] Q. Lin, T. Ran, Z. Siyong, and W. Yue, ‘‘Adaptive time-varying filter for FAN LUO received the B.S. degree in telecom-
linear FM signal in fractional Fourier domain,’’ in Proc. 6th Int. Conf. munications engineering from the Zijin College,
Signal Process., Aug. 2002, pp. 1425–1428.
Nanjing University of Science and Technology,
[28] A. B. Gershman and M. G. Amin, ‘‘Wideband direction-of-arrival estima-
Nanjing, China, in 2019, where he is currently pur-
tion of multiple chirp signals using spatial time-frequency distributions,’’
suing the M.S. degree in electronics and communi-
IEEE Signal Process. Lett., vol. 7, no. 6, pp. 152–155, Jun. 2000.
cations engineering. His research interests include
[29] F. Chen, T. Zhou, W. Yi, L. Kong, and B. Zhai, ‘‘Passive direct position
determination of multiple emitters transmitting unknown LFM signals,’’ joint radar and communication.
in Proc. IEEE Radar Conf. (RadarConf), Apr. 2018, pp. 1129–1133.
[30] X. Wang, B. Li, H. Yang, and W. Lu, ‘‘Off-grid DOA estimation for
wideband LFM signals in FRFT domain using the sensor arrays,’’ IEEE
Access, vol. 7, pp. 18500–18509, 2019.
[31] H. Guo, G. A. Sitton, and C. S. Burrus, ‘‘The quick discrete Fourier
transform,’’ in Proc. IEEE Int. Conf. Acoust., Speech Signal Process., WEN WU (Senior Member, IEEE) received
Apr. 1994, pp. III/445–III/448. the Ph.D. degree in electromagnetic field and
[32] T. Ran, P. Xianjun, S. Yu, and Z. Xinghao, ‘‘A novel discrete frac- microwave technology from Southeast University,
tional Fourier transform,’’ in Proc. CIE Int. Conf. Radar, Oct. 2001,
Nanjing, China, in 1997.
pp. 1027–1030.
He is currently a Professor with the School of
[33] X. Ouyang and J. Zhao, ‘‘Orthogonal chirp division multiplexing for coher-
Electronic and Optical Engineering, Nanjing Uni-
ent optical fiber communications,’’ J. Lightw. Technol., vol. 34, no. 18,
pp. 4376–4386, Sep. 15, 2016.
versity of Science and Technology, where he is
[34] Y. Yu, W. Wang, X. Ouyang, Z. Wang, and J. Zhao, ‘‘Discrete Fresnel also the Associate Director of the Ministerial Key
transform spread OFDM for coherent optical fiber communication,’’ IEEE Laboratory of JGMT. He has authored or coau-
Photon. Technol. Lett., vol. 30, no. 1, pp. 91–94, Jan. 1, 2018. thored over 300 journals and conference papers
[35] L. G. D. Oliveira, M. B. Alabd, B. Nuss, and T. Zwick, ‘‘An OCDM and has submitted over 30 patent applications. His current research inter-
radar-communication system,’’ in Proc. 14th Eur. Conf. Antennas Propag. ests include micro-wave and millimeter-wave theories and technologies,
(EuCAP), Mar. 2020, pp. 1–5. microwave and millimeter-wave detection, and multimode compound detec-
[36] C. Fang, H. Zishu, L. Hongming, and L. Jun, ‘‘The parameter setting tion. He was a recipient of six times of ministerial and provincial-level
problem of signal OFDM-LFM for MIMO radar,’’ in Proc. Int. Conf. science and technology awards.
Commun., Circuits Syst., May 2008, pp. 876–880.

1358 VOLUME 11, 2023