www.ietdl.

org
Published in IET Radar, Sonar and Navigation
Received on 4th July 2011
Revised on 7th January 2013
Accepted on 27th January 2013
doi: 10.1049/iet-rsn.2012.0116

ISSN 1751-8784

Detection of low observable moving target in sea

clutter via fractal characteristics in fractional
Fourier transform domain
Xiaolong Chen1, Jian Guan1, You He1, Jian Zhang2
1
Department of Electronic and Information Engineering, Naval Aeronautical and Astronautical University,
Yantai Shandong 264001, People’s Republic of China
2
Training Brigade of the Received and Reformed Equipment, Naval Aeronautical and Astronautical University,
Yantai Shandong 264001, People’s Republic of China
E-mail: [email protected]

Abstract: Effective detection of low observable moving target at sea is important for remote sensing and radar signal processing.
The non-Gaussian property of sea clutter and lack of accurate model make the detection difficult for statistics based detectors.
Also the fractal techniques in time domain cannot achieve high detection probability in heavy sea clutter. To help solve the
problems, fractal characteristics of IPIX datasets in fractional Fourier transform (FRFT) domain are analysed making use of
the fluctuation of FRFT amplitudes and moving target detection algorithms are proposed based on the fractal characteristics in
FRFT domain. Firstly, fractal model in FRFT domain is established with fractional Brownian motion model and two
judgment and extraction methods are employed for calculating the fractal characteristics in FRFT domain. It is found that sea
clutter of different polarisations exhibit fractal behaviours in FRFT domain, that is, self-similarity property, within its
corresponding scale-invariant interval. Then, we find that four specific fractal statistics in the best FRFT domain can provide
valuable information for developing simple and effective detectors. Finally, traditional amplitude detector and Hurst exponent
detector in time domain are compared and the results prove the superior detection ability of low observable moving target
without complex computations.

1 Introduction observed to be highly non-Gaussian, and even spiky [2].

Hence, sea clutter modelling is a particularly challenging
Sea clutter is the backscattered returns from a patch of the sea problem. A lot of efforts have been made to fit various
surface illuminated by a radar pulse, which is a main factor of distributions to the observed amplitude of sea clutter data,
target detection at sea [1–4]. Robust detection of moving including Rayleigh, Weibull and log-normal distributions
target at sea is one of the fundamental problems and plays a [8, 9]. A more useful model, however, including both
significant role in both military and civil fields. In the civil components in a ‘compound’ representation, is known as
fields, it has been widely used for safety navigation, natural the K-distribution, which is described in terms of Rayleigh
obstacles avoidance and ocean environmental monitoring speckle fluctuations modulated by gamma underlying radar
[5]; its applications in military can be expanded to remote cross-section (RCS) variations [10]. However, these
sensing, antisubmarine, low-flying aircrafts striking and low distributions do not address the extended tail region when
probability target detection [6]. Unlike land clutter, the sea spikes are present and the non-Gaussian characteristic in
ocean surface is irregular and rapidly fluctuant, which heavy sea conditions makes it difficult for classical
results in poor detection probability both in time and statistical mathematics to represent sea clutter accurately.
frequency domain under the low signal-to-clutter ratio For this reason, it has not been practical to design a really
(SCR) environment. If the target signal is weak comparing optimum detector based on assumed statistics of clutter and
to the sea clutter, it will be hard to be detected [7]. target signals.
An accurate model for signals backscattered from the sea To gain deeper understanding of the nature of sea clutter,
surface is a problem of primary concern among radar the concept of fractal has been employed for describing the
engineers. The development of statistical models that roughness of sea surface and the scattering from rough
properly characterise radar clutter processes is critical for surface [11, 12]. It is shown that the sea surface has a
designing optimum algorithms of detecting targets within fractal nature and is more irregular than the target signal.
radar clutter as well as for performance evaluation of those Therefore the fractal dimension (FD) of clutter is bigger
detectors. On account of radar operating at low grazing than that of target. This is the distinction, which helps radar
angles with high resolution capabilities, sea clutter is often detectors to reject the sea clutter. A large variety of

IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651 635
doi: 10.1049/iet-rsn.2012.0116 & The Institution of Engineering and Technology 2013
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org
concepts deriving from fractal geometry, such as fractal error, domain can achieve a better performance than some classical
FD [13] and multifractal analysis [14] have been employed to methods in time domain, which have been validated by real
characterise the rough surface of sea clutter, which provide intelligent pixel (IPIX) processing radar datasets [23].
effective ways to detect low observable targets. However, The remaining of the paper is organised as follows. In Section
because of the heavy sea conditions, that is, under some 2, fractal characteristic of sea clutter data in FRFT domain is
low SCR conditions, the differences of fractal characteristic analysed and two methods are used to the judgment and
between sea clutter and target in time domain have little extraction of fractal characteristic in FRFT domain. The effect
improvement for target detection. of transform order on fractal parameter estimation is also
Fractional Fourier transform (FRFT) has become a powerful studied. In Section 3, we propose four moving target
tool for time-varying signal analysis especially linear detectors. Then, real sea clutter data are used for verification
frequency modulation (LFM) signal [15]. Its advantage is and detection performances are presented in Section 4.
that FRFT of LFM signals has a strong aggregation of Finally, some concluding remarks are presented in Section 5.
energy without cross-terms interference [16]. It is especially
applicable and suitable for detection of moving target
modelled as LFM signal in some circumstances [17, 18].
The radar return of moving target after FRFT is highly 2 Fractal characteristic of sea clutter in
concentrated as a peak in the FRFT domain while the FRFT domain
energy of sea clutter in FRFT domain is mainly distributed
around the frequency domain with a fluctuant and rough 2.1 FRFT of sea clutter and moving target
surface [19, 20]. As a result, we can employ the fractal
theory to describe the geometric complexity and irregularity The FRFT is the generalisation of Fourier transform and can
of sea clutter and moving target signal in FRFT domain and be interpreted as a rotation by an angle in the time–frequency
then take the fractal characteristic as test statistic for moving plane. It is defined by means of the transformation kernel
target detection. It involves the research of fractal detection Kp(t, u), and expressed as [24]
in transform domain. Many researchers have discussed the
calculation method of FD with transform tools while the
+1
fractal differences at some specific scale in FRFT domain Fp [x](u) = x(t)Kp (t, u) dt (1)
are rarely studied, which may be another useful tool to −1
detect low observable moving target within strong sea
clutter. In [21], a novel method to detect the FD for a where p is the transform order, Fp is the FRFT operator and
composite signal consisting of a periodic signal and a fractal Kp(t, u) is indicated by (see (2))
noise is proposed. The method transforms the composite
signal to the Fourier domain, and then uses the Hough where n is an integer and α is the rotation angle, α = pπ / 2.
transform to detect the FD. In [22], a FRFT-based Hurst The kernel of FRFT is a chirp signal, as a result, if a LFM
parameter estimation method is introduced to analyse the signal is projected onto a rotated frequency axis with a
long range dependence (LRD) in time series and fractional proper angle, the energy distribution will be highly
Gaussian noises (FGN) which typically possesses LRD with concentrated with a impulse function.
known Hurst parameters is used to test the accuracy of the Within the observation time T, the received signal model of
proposed estimator. However, these methods just use the single moving target in the presence of sea clutter can be
transformation tool to calculate FD, which rarely involve a characterised as follows [20, 24]
systematical study of the fractal properties in transform
domain without a proper mathematical model. Real data are
 
also needed to testify the conclusions. x(t) = s(t) + w(t) = A exp j2pf0 t + jpmt 2
In this paper, statistical characteristic of sea clutter in FRFT
domain is studied with the popular fractal model encountered + c(t) + n(t), |t| ≤ T (3)
in physical fields and two kinds of extraction methods of
fractal characteristic in FRFT domain are employed, namely,
the ‘random walk’ process and the ‘increment’ process. After where A is the amplitude. The acceleration and initial velocity
the energy accumulation of moving target in FRFT domain, of targets are denoted as a, v0 and the instantaneous velocity
four fractal characteristics based detectors are proposed for at time t can be formulated as V = v0 + at. The chirp rate is
moving target detection, which are FD, intercept, fractal μ = 2a / λ, λ represents the wavelength of the radar, f0 is the
fitting error (FFE) and FD variance (FDV). These four centre frequency corresponds to the velocity of the targets
statistics capture the key fractal characteristic differences f0 = 2v0 / λ. w(t) is the clutter, which is made up of two
between sea clutter and low observable moving target in components, that is, sea clutter c(t) and Gaussian noise n(t).
FRFT domain, so the fractal detection methods in FRFT Then we calculate the FRFT of s(t) which is expressed as
(see (4))

⎧   
⎪
⎨ 1 − j cot a 1 2 1 2
exp j t cot a − ut csc a + u cot a if a = np
Kp (t, u) = 2p 2 2 (2)
⎩ 
⎪ 
d u − ( − 1) tn
if a = np



T
1 − j cot a ju2 cot a/2 T j((1/2)t2 cot a−ut csc a) ju2 cot a/2 2
Fa [s(t)] = e s(t)e dt = AAa e ej(( cot a+2pm)/2)t +j(2pf0 −u csc a)t dt (4)
2p −T −T

636 IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rsn.2012.0116
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org
√
When Aa = (1 − j cot a)/2p and α = arctan(−1 / 2πμ), the meters for gravity waves but centimetres for capillary
FRFT modulus is given by waves. The two-scale nature of sea surface leads naturally
  to the use of a two component model for radar sea

T
       backscatter. The large scale waves resulting from the distant
F [s(t)] = AA exp j 2pf0 − u csc a t dt  wind are modelled as a sum of single frequency signals and
a
 a −T 
the small scale waves because of the local wind and the
 
     interferences among waves are approximate to Gaussian
= 2AAa T eju cot a/2 sinc 2pf0 − u csc a T 
2

noise [8].
    In Figs. 1a4, a5, b4 and b5, the FRFT modulus of a
= 2Aa T Asinc 2pf0 − u csc a T  (5) simulated moving target at sea with SCR = 0 dB are
displayed. Since the radar echo of moving target in one
From (5), if the moving state matches the best transform angle range bin can be modelled as LFM signal among adjacent
and f0 = u csc α / 2π, the moving target signal can be well pulses, it is clearly visible that its amplitude appears as a
accumulated with a peak in FRFT domain. peak in some specific transform domain called the best
    FRFT domain. To further discuss the effects of the peak on
a0 , u0 = arg maxFa [s(t)] (6) fluctuation, we draw the spectrum of sea clutter and moving
a,u
target in the best FRFT domain ( popt = 1.056) for the two
datasets, shown in Figs. 1a3, a5, b3 and b5. Comparing
where the axis (α0, u0) is the location of the biggest FRFT Fig. 1a3 with a5, the FRFT modulus of range bin where the
modulus obtained as a sinc function. moving target is present show the higher ‘regularity’, which
Alternative methods of investigation of the mathematical may lead to the reduction of the FD value. This is also
analysis require the use of data. The simulation of data can verified for the dataset_31, which corresponds to the case of
be controlled by the user but the results are sometimes not lower sea state (Figs. 1b3 and b5).
quite accurate (statistical generation of the sea clutter) or From the above analysis, it infers that the fractal
computationally expensive. As a result, we prefer to use characteristic of sea clutter and moving target in FRFT
real data which is related to the widely used Osborn Head domain is rather different, which can be used for detection.
database, collected at Osborne Head Gunnery Range, in When the moving target interacts with sea clutter, the
November 1993, with McMaster University IPIX radar amplitude of sea clutter will be affected by the peak with a
[23]. To perform the fractal analysis in FRFT domain, we smoother surface. However the energy of sea clutter only is
have selected two representative datasets, one with higher distributed around and the amplitude varies sharply with a
sea state (19931107_135603, a significant wave height of rougher surface. As a result, the irregular degree of moving
2.3 m) named as dataset_17 and the other with lower sea target bin is smaller than that of sea clutter bin in FRFT
state (19931109_202217, a significant wave height of domain leading to a smaller FD.
0.9 m) named as dataset_31. Each analysed dataset is
relative to 14 range cells and a few of the range bins hit a
small target made of a spherical block of styrofoam of
diameter 1 m, wrapped with wire mesh. The range bin 2.2 Statistical characteristic of sea clutter in
where the target is the strongest is labelled as primary target FRFT domain
bin, which is selected to analyse the fractal behaviour in the
presence of a target and the neighbouring range bins where We will introduce one of the prototypical models for random
the target may also be visible are labelled as bins with fractals – the fractional Brownian motion (FBM) model, then
target. It is possible that the target may possess verify whether the model can describe sea clutter in FRFT
instantaneous moving state because of the fluctuation at sea. domain by analysis of statistical characteristic. The FBM
More information is summarised in Table 1. BH(t) is a Gaussian process with mean zero, continuous and
Figs. 1a1, a2, b1 and b2 show the radar echo in time and non-differentiable, which can be obtained by fractional
FRFT domain, respectively, with dataset_17 and dataset_31 integration of Brownian motion [25]. It is a non-stationary
for HH polarisation. It can be found that higher sea state model and quite sensitive to the change of scale, which
results in bigger amplitude with rougher sea surface. The indicates the self-similarity property
energy of sea clutter in FRFT domain is distributed in the
whole FRFT domain, particularly around order p = 1, that
t
is, in frequency domain. In fact, small varying acceleration 1
BH (t) = (t − l)H −(1/2) dB(l) (7)
in sea clutter also exits, which only last a very short time. G(H + 1/2) −1
As a result, the energy of sea clutter cannot be well
focused. The roughness of sea surface determines sea
clutter properties, which is often modelled as two types of where H is the Hurst exponent inversely proportional to the
waves: gravity waves which is also called as large scale irregular degree and its relation with FD is D = DT + 1 − H
waves and capillary waves namely the small scale waves (DT means the topology dimension and for one-dimensional
[2]. Wavelength is the main difference between them; time series DT = 1). The variance and covariance of FBM

Table 1 Parameters of IPIX radar datasets

Filename Significant wave Observation Grazing Sea Primary Bins with Polarisation Sampling
height, m direction angle, deg. state target bin target frequency, Hz

dataset_17 2.3 upwind <1 5–6 9 8 : 11 HH/HV/VH/VV 1000

dataset_31 0.9 downwind <1 3–4 7 6:8 HH/HV/VH/VV 1000

IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651 637
doi: 10.1049/iet-rsn.2012.0116 & The Institution of Engineering and Technology 2013
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org

Fig. 1 Radar echo in time and FRFT domain, respectively, with IPIX radar data (HH polarisation)
Dataset_17: a1 Waveform of sea clutter in time domain
a2 FRFT module of sea clutter
a3 Spectrum of sea clutter in the FRFT domain ( popt = 1.056)
a4 FRFT module of a moving target (SCR = 0 dB)
a5 Spectrum of a moving target in the best FRFT domain ( popt = 1.056)
Dataset_31: b1 Waveform of sea clutter in time domain
b2 FRFT module of sea clutter
b3 Spectrum of sea clutter in the FRFT domain ( popt = 1.056)
b4 FRFT module of a moving target (SCR = 0 dB)
b5 Spectrum of a moving target in the best FRFT domain ( popt = 1.056)

638 IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rsn.2012.0116
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org
Table 2 Statistical increment property of sea clutter in FRFT domain
μ s21 s22 s23 s24

dataset_17 HH polarisation − 1.0952 × 10 − 4 0.0379 0.0558 0.0638 0.0623

VV polarisation − 1.7908 × 10 − 5 0.0023 0.0034 0.0033 0.0037
dataset_31 HH polarisation − 3.6615 × 10 − 4 0.0544 0.0069 0.0064 0.0065
VV polarisation − 2.4116 × 10 − 5 0.0120 0.0099 0.0089 0.0191

are denoted as VV) polarisation. It is easily verified from Figs. 2a1, a3, b1
and b3 that the mean value is about zero and the
 2 distribution of increment is similar with Gaussian
E BH (t) = t 2H (8)
distribution. For a Gaussian distribution, the quantile–
  1  quantile plot (QQplot) of data results in a straight line. The
E BH (s)BH (t) = s2H + t 2H − |s − t|2H (9) normality of given data are tested by QQplot in Figs. 2a2,
2
a4, b2 and b4 showing that the Gaussian distribution is a
Its increment FGN xH(t) is Gaussian distributed, stationary good assumption. The stationary property is further studied
process with zero mean and showed in Table 2, which indicates the statistical
properties do not change with time. Hence, the increment
  series di of sea clutter in FRFT domain can be
xH (t) = BH (t + t) − BH (t)  N 0, s2H t2H t.0 (10)
approximately distributed as a stationary Gaussian
where t is delay and the autocovariance function γ(t) for the stochastic process and analysed by FBM model.
FGN process
2.3 Judgment and extraction of fractal
1  characteristic in FRFT domain
g(t) = (t + 1)2H − 2t2H + |t − 1|2H (11)
2
2.3.1 Sea clutter data treated as ‘random walk’
Hence process: We first subtract the mean from the new time
series denoted as a = {a(i), i = 1, 2, ..., N}, where
g(t) / |t|2H−2 (12)
a(i) = d(i) − m, i = 1, 2, . . . , N (17)
If the sea clutter in FRFT domain can satisfy the FBM model,
its increment should also meet the condition (10) and (12) Then we form the partial summation of a to construct a new
with statistical self-similarity property. For the real sea time series h = {h(i), i = 1, 2, ..., N}, which is a ‘random walk’
clutter data of IPIX, the verification method consists of the process of a [26, 27]
following steps:

n
Step 1: Given the sea clutter time series of N samples h(n) = a(i) (18)
i=1
y(i) = x(i) −
x, i = 1, 2, . . . , N (13)
Actually a is an ‘increment’ process of h. One then examines
where
x is the mean value of time series of sea clutter x(i). whether the following scaling-law holds or not
Step 2: Transform y(i) to FRFT domain with order p
expressed as Yp(i), then the increment of sea clutter in F(m) = k|h(n + m) − h(n)|2 l1/2 / mH (19)
FRFT domain is d = {d(i), i = 1, 2, ..., N}
where F() is called fluctuation function and the average 〈·〉 is
d(i) = Yp (i + 1) − Yp (i) (14) taken over all possible pairs of (h(n + m), h(n)), H is the Hurst
exponent and m is the scale. Take the logarithm operation of
(19) [26]
Step 3: Suppose d has the ergodicity property, then the
statistical average μ and variance σ 2 can be estimated by  
the increment log2 F(m) = log2 k|h(n + m) − h(n)|2 l1/2
= H log2 (m) + const (20)
1 N −1
m= d(i) (15)
N − 1 i=1
where ‘const’ means the constant value because of the scaling

N coefficient in (19).
1
s2 = [d(i) − m]2 (16) When the time series of sea clutter in FRFT domain Yp is
N −1 i=1 considered as ‘random walk’ process, which should be
satisfied with (20), the process under investigation is said to
be a fractal process. Given a time series of sea clutter, we
can obtain the fluctuation function log2 F(m) in FRFT
Fig. 2 shows the statistical increment distribution of sea domain at different scales m. If the fractal characteristic of
clutter in FRFT domain for both dataset_17 and dataset_31 sea clutter in FRFT domain coincides with the theoretical
with horizontal-horizontal and vertical-vertical (HH and model, log2 F(m) will increase with log2(m) in a linear

IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651 639
doi: 10.1049/iet-rsn.2012.0116 & The Institution of Engineering and Technology 2013
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org

Fig. 2 Statistical increment distribution of sea clutter in FRFT domain

Dataset_17: a1 Histogram (HH polarisation)
a2 Quantile-quantile plot (HH polarisation)
a3 Histogram (VV polarisation)
a4 Quantile-quantile plot (VV polarisation)
Dataset_31: b1 Histogram (HH polarisation)
b2 QQplot (HH polarisation)
b3 Histogram (VV polarisation)
b4 QQplot (VV polarisation)

640 IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rsn.2012.0116
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org
manner. The typical fractal curves of log2 F(m) against peak will change the degree of fluctuation and roughness in
log2(m) of both dataset_17 and dataset_31 with four FRFT domain, so that the fractal characteristic will change
different polarisations are shown in Fig. 3. It is observed as well (with smaller FD) and this can be used to
that the fractal curves are relatively linear within the time distinguish sea clutter and moving target.
range from about 30 ms (m = 25) to 1 s (m = 210). The scale
range (25–210) is called the scale-invariant interval. Thus, 2.3.2 Sea clutter data treated as ‘increment’
the sea clutter data in FRFT domain can be regarded as process: Also, the sea clutter can be modelled as an
fractal with self-similarity property within the scale range. ‘increment’ process as discussed in (17). However, the
Then, the fractal characteristic such as Hurst exponent of results of [13] show that using this model the Hurst
sea clutter in FRFT domain can be estimated by fitting a exponents of sea clutter and primary target are almost the
straight line to the log2(m)–log2F(m) curve in its same, which are very close to 1. Let us now consider Yp as
corresponding scale-invariant interval [28]. If a low an ‘increment’ process and then check whether it satisfies
observable moving target is embedded in sea clutter, the the condition (20). From Fig. 4, it is noticed that the scaling

Fig. 3 Fractal curves log2(m)–log2 F(m) of IPIX radar data in FRFT domain
The sea clutter data are considered as a ‘random walk’ process
Dataset_17: a1 #2(sea clutter bin)
a2 #3(sea clutter bin)
a3 #9(primary target bin)
Dataset_31: b1 #2(sea clutter bin)
b2 #3(sea clutter bin)
b3 #9(primary target bin)

IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651 641
doi: 10.1049/iet-rsn.2012.0116 & The Institution of Engineering and Technology 2013
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org
relations are improved and fairly linear in the range of 2–211. obtained by ‘increment’ process is not a very efficient
Data with different polarisations have the same linear trend statistic to describe the texture property of radar echo in
and the Hurst exponent can be calculated by least square FRFT domain.
linear fitting about 0.65–0.7 for dataset_17 and 0.65–0.68
for dataset_31. Although the fractal curves of the primary 2.3.3 Relation of FD in FRFT domain with time and
target bin change as linear pattern with H = 0.63, there exits range: Whether the FD in FRFT domain is independence
little difference between sea clutter and moving target over of time and range or not is physically meaningful when the
all fractal scales in FRFT domain. Thus, the characteristic physical conditions of sea wave structure and system

Fig. 4 Fractal curves log2(m)–log2 F(m) of IPIX radar data in FRFT domain
The sea clutter data is considered as an ‘increment’ process
Dataset_17: a1 #2(sea clutter bin)
a2 #3(sea clutter bin)
a3 #9(primary target bin)
Dataset_31: b1 #2(sea clutter bin)
b2 #3(sea clutter bin)
b3 #9(primary target bin)

642 IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rsn.2012.0116
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org
configuration do not change during the observation time and 1 
Mt
within the illuminated area. Before presenting the results, we mHH
r = DHH (n, m) (24)
make several assumptions as follows: (i) the total observation Mt m=1
time used for FRFT calculation is approximately seconds
long, which is short enough to neglect changes in weather A similar definition holds for VV polarisation.
conditions; (ii) the radar system does not change its In Fig. 5, the time and range signatures of FD of sea clutter
parameters; and (iii) the illuminated area is small enough to in FRFT domain are reported with dataset_17 and dataset_31
ascertain the spatially homogeneous sea surface. for both HH and VV polarisations. From Figs. 5a1 and b1, it
Then in order to show the relation of FD in FRFT domain can be found that within the time interval, that is, 1 s, the
with time and range, we define two parameters, that is, the MTFD in FRFT domain for a given range bin can be
mean time standard deviation (MTSD) σt and the mean considered constant under a tolerance of a 10% error (D =
range standard deviation (MRSD) σr 1.8–1.9 for dataset_17 and D = 1.5–1.65 for dataset_31).
Figs. 5a2 and b2 show the MRFD signature relative to a
 1/2 given sweep, which may draw the same conclusion with
1   HH 2
Mt
MTFD. At the same time, two points should be noted: (i)
sHH
t = D (n, m) − mHH
t (21) for the same polarisation, the MTFD and MRFD in FRFT
Mt m=1
domain of sea clutter under lower sea state are smaller than
those of sea clutter under higher sea state; (ii) for the same
 1/2 dataset, the MTFD and MRFD in FRFT domain of sea
1  r  2
N
clutter with HH polarisation are smaller than those of sea
sHH
r = DHH (n, m) − mHH
r (22) clutter with VV polarisation; and (iii) since range bin 6 and
Nr n=1
7 of dataset_17 and range bin 5 and 6 of dataset_31 are
close to the primary target bin, their MTFD and MRFD in
where D HH(n, m) represents the dimension of the nth range FRFT domain are relatively smaller than those of the other
bin and mth subinterval. mHH and mHH are the mean time sea clutter bins. The values of time and range signatures of
t r
FD (MTFD) μt and mean range FD (MRFD) μr, respectively FD in FRFT domain obtained for the two datasets are
reported in Table 3. The results show that, the MTSD and
MRSD as defined in (21) and (22), are smaller than 0.05,
1  rN which represents the standard deviation of the FD
mHH
t = DHH (n, m) (23) estimation error. Therefore, in general, we can conclude that
Nr n=1 the fractal characteristics of sea clutter in FRFT domain are

Fig. 5 Time and range signatures of FD of sea clutter in FRFT domain

Dataset_17: a1 FD time profile for a given range bin
b1 FD range profile for a given sweep
Dataset_31: a2 FD time profile for a given range bin
b2 FD range profile for a given sweep

IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651 643
doi: 10.1049/iet-rsn.2012.0116 & The Institution of Engineering and Technology 2013
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org
Table 3 Time and range signatures of FD in FRFT domain for sea clutter datasets
MTFD MRFD MTSD MRSD

mHH
t mVV
t mHH
r mVV
r sHH
t sVV
t sHH
r sVV
r

dataset_17 1.8355 1.8673 1.8619 1.8794 0.0356 0.0440 0.0328 0.0372

dataset_31 1.5332 1.6070 1.5574 1.6067 0.0484 0.0547 0.0397 0.0433

independent of time and range statistically, that is to say, it has little difference of FD. This is because that the sea clutter
the universal significance as the same results may be obtained can be modelled by single frequency signals and Gaussian
using other sea clutter datasets. noise in Section 2.1 and cannot be focused well in the
Based on the above analysis, when sea clutter in FRFT FRFT domain, which indicates that the roughness and
domain is referred as fractal, the following conditions irregularity are order dependent. Whereas the fractal curves
should be satisfied: (i) it should have the form of (20), of primary target bin have close relation with different
which indicates the self-similarity property within the orders in Fig. 6b. This means in the best FRFT domain the
scale-invariant interval; and (ii) the FD should be bigger fluctuation of sea clutter is affected by the accumulated
than its topological dimension with stability. Both of the energy of moving target, which results in the biggest slope
two requirements are well satisfied using dataset_17 and of fitting line, that is, Hurst exponent. Then FD in different
dataset_31 with the results in Figs. 3–5. Actually, we have FRFT domain is calculated both in sea clutter bin (Ds =
analysed many other datasets in different wind and sea 1.7 – 1.9) and primary target bin (Dt = 1.35 – 1.8), which are
conditions and, concerning the fractal characteristic in shown in Figs. 6c and d. Based on the maximum of this
FRFT domain, we obtained pretty similar results. On the difference, the best transform order can be obtained, that is,
condition that ‘random walk’ process can best describe the p0 = 1.056. From Figs. 6e and f, one notices that in the best
fractal characteristic in FRFT domain, it will be more FRFT domain the fractal characteristic for the data without
convenient and accurate to develop a robust detector for target is rather different from the data with primary target
detecting moving target within sea clutter. In the following for both HH and VV polarisations.
part, we only employ dataset_17 because of the higher sea
state, which is more persuasive for detection performances 3 Moving target detection methods based on
analysis. the fractal characteristics in FRFT domain
2.4 Effect of transform order on FD In this section, the moving target detector is designed based
on four fractal characteristics in FRFT domain and its
Given the moving target model in (1), the Fα[s(t)] can be performance is compared with classical detector. Fig. 7
calculated by DFRFT algorithm developed by Ozaktas gives the flowchart of the algorithm.
which is realised by chirp convolution [29]
√ 3.1 FD and intercept method
2pAa
Fa [s(t)] =
2F The fluctuation functions of sea clutter log2 Fc(m) in the
 2 
jpn cot a j2pu csc a  n 
 scale-invariant interval can be expressed as
N  2 
× exp jpu cot a exp − s
n=−N (2F)2 2F 2F log2 Fc (m) = (2 − Dc ) log2 (m) + Ic (28)
(25)
where the FD Dc and the intercept Ic can be obtained by the
where F denotes as the highest frequency of s(t). Then the linear fitting. It should be noted that the existence of
amplitude of peak Pa in FRFT domain is moving target at sea will change the fractal characteristic of
data with sea clutter to some extent, hence we can use the
√  √
 2pAA  2p|A|Aa0 two statistics of different range bins, respectively, to
 a0 
Pa =  (2N + 1) = Tfs (26) distinguish sea clutter and moving target. If one of the
 2F  2F following statistics is lower than threshold, target can be
declared
where fs is the sampling frequency and according to (5), the
peak width Pw is H0
 .
T1 = Di  h1 (29)
2p 2p 2p p ,
Pw = = sin a0 = sin p0 (27) H1
T csc a0 T T 2
H0
Hence, only in the best FRFT domain, the fractional Fourier  .
T 2 = Ii  h 2 (30)
spectrum of the LFM signal focuses well and the fractal ,
difference will be the biggest. H1
Let us start with the fractal curves of sea clutter bin and
primary target bin in the range of p = 1 – 1.2 with equal where T is the test statistic, i denotes the ith range bin and η is
interval. It can be observed from Fig. 6a, the fractal curves a detection threshold based on constant false alarm rate
of sea clutter in different FRFT domains are fairly stable (CFAR) method, in order to provide a certain probability of
and linear within the scale-invariant interval and there is detection for a given false alarm rate [30]. Taking the

644 IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rsn.2012.0116
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org

Fig. 6 Effects of transform order on fractal parameter estimation in FRFT domain (Dataset_17)
a Fractal curves of sea clutter with different orders (HH polarisation)
b Fractal curves of primary target with different orders (HH polarisation)
c FD estimation of sea clutter with different orders (HH polarisation)
d FD estimation of primary target with different orders (HH polarisation)
e Fractal curves in the best FRFT domain (HH polarisation)
f Fractal curves in the best FRFT domain (VV polarisation)

Fig. 7 Moving target detector based on fractal characteristic differences in FRFT domain

IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651 645
doi: 10.1049/iet-rsn.2012.0116 & The Institution of Engineering and Technology 2013
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org
non-parametric CFAR method, for example, it does not in primary target bin. Then we can use the FDV for target
require the probability distribution of the background detection
clutter, therefore η1 and η2 can be formed adaptively H1
depend on the pre-determined false alarm rate only.      .
T4 = Vi  = var Di  h (35)
, 4
3.2 FFE method H0

where V is the FDV in FRFT domain and η4 is the threshold

FFE depicts the extent of signal in line with the fractal model.
determined by CFAR method.
It can be found from Fig. 2 that fractal model can be fitted to
the structure of sea clutter in FRFT domain better than that of
moving target. The energy of moving target can be 4 Real sea clutter verification and results
concentrated as a peak in FRFT domain with a smoother analysis
surface reducing the self-similarity. Hence, the fitting error
of fractal curves of sea clutter is smaller than that of In this section, simulations are carried out to validate the four
moving target, which can be taken as test statistic. Given a proposed detectors in the earlier section with dataset_17 for
point set {xi, yi} and the straight line y = ax + b, the linear both HH and VV polarisations.
fitting error is defined as the mean distances from real
coordinate value to the straight line 4.1 Results of moving target detection

2
N  Figs. 8a–d show the target detection results among 14 range
i=1 ax i + b − yi bins via the four fractal characteristics in FRFT domain,
E= (31)
N respectively, that is, FD, intercept, FFE and FDV. We
notice that they all provide excellent abilities to distinguish
Then, a fractal detection method based on the difference of sea clutter bin and primary target bin. The FD and intercept
FFE in FRFT domain between sea clutter and moving of target bin are much lower than those of sea clutter bin
target is expected to further increase the accuracy of while the FFE and FDV are just the opposite since they
detection. Target can be declared if it exceeds the threshold. reflect the fitting degree to the fractal model. Taking the FD
The statistic is as follows in FRFT domain, for example, we may draw several useful
conclusions. (i) High resolution sea clutter at low grazing
H1 angles is highly spiky and the number of scatters is not
 .
T3 = Ei  h3 (32) enough to satisfy the central limit theorem. These complex
, interactive scattering signals lead to high FD in FRFT
H0
domain. Conversely, the echo of the moving target can be
coherently accumulated in the FRFT domain with low FD;
where E is the FFE in FRFT domain and η3 is the threshold (ii) The FD of sea clutter is stable for both HH and VV
determined by CFAR method. polarisation bigger than its topological dimension, which is
satisfied with the second condition of fractal criterion in
3.3 FDV method Section 2.3; and (iii) The biggest FD difference appears in
the primary target bin and HH dataset is more appropriate
Ideal fractal object satisfies the self-similarity at all scales, for detection. Then, we compare the detection results of
which means that the FD is independent of scales. It can be proposed methods with traditional detection methods, such
found that the fractal feature of sea clutter in FRFT domain as amplitude detector and Hurst exponent detector in time
coincides well with that of theoretical fractal model in the domain, which are shown in Figs. 8e and f. It can be seen
scale-invariant interval. However, the fractal curves of that the differences of the two test statistics of sea clutter
moving target do not increase with scale in a linear way, bins and primary target bin are not obvious especially with
which indicates that its fractal characteristic varies with the VV datasets, indicating that it is more difficult for target
scale. The reason may be that when the width of peak and detection in heavy sea clutter.
scale is close enough, there will exit great changes of Let us now examine whether a robust moving target
amplitude, which reflects in a sudden change of FD. Hence, detector within sea clutter can be developed based on the
we can calculate the FD at different scales in FRFT domain four test statistics. We have systematically conducted 1000
and employ the changes between scales for target detection. times detections with sea clutter bin and primary target bin.
The x, y coordinate of fractal curves in FRFT domain can The frequencies under the two hypotheses for HH datasets
be expressed as follows are shown in Fig. 9. It is observed that frequencies of the
four test statistics can separate the sea clutter from moving
 
xi = log2 mi  target clearly.
(33)
yi = log2 F mi
4.2 Detection performance analysis
Then the FD of adjacent scale Δi is denoted as To demonstrate the effectiveness and the detection
performances of the four methods in a quantitative manner,
   
log2 F mi − log2 F mi−1 y − yi−1 we have analysed more data. For each range bin, the
DDi =     = i (34) original samples are divided into 1000 overlapping
log2 mi − log2 mi−1 xi − xi−1
segments, each with 2000 points, which is sufficient for the
purpose of calculating the four test statistics (T1, T2, T3 and
As it is shown in Fig. 3, there exits small variation of FDs in T4) within the scale-invariant interval. Then, the value of
sea clutter bin with different scales, while remarkable changes the detection probability (Pd) is evaluated by determining

646 IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rsn.2012.0116
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org

Fig. 8 Results of different fractal characteristics for the 14 range bins (Dataset_17)
a FD
b Intercept
c FFE
d FDV
e Coherent accumulation in time domain
f Hurst exponent in time domain

the percentage by which test statistics for the 1000 datasets domain as an example, if the threshold is set at 1.75, the Pd
exceeds the corresponding threshold. The Pd for different is 90% whereas the Pfa is < 5%.
threshold values is given in Figs. 10a–d. We have also The comparison of detection performances by the four
constructed the false alarm rate (Pfa) curves, as shown in proposed detectors for HH and VV polarisations are carried
Figs. 10e–h. Take detection method via FD in FRFT out and shown in Table 4, which can reflect the true reality

IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651 647
doi: 10.1049/iet-rsn.2012.0116 & The Institution of Engineering and Technology 2013
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org

Fig. 9 Frequencies of sea clutter bin and primary target bin with HH polarisation (Dataset_17)
a FD
b Intercept
c FFE
d FDV

of the detection performance by analysing the results of where the subscript ‘s’ and ‘ci’ denote the primary target bin
different fractal characteristics for the 14 range bins in and ith sea clutter bin. Owing to the different dimensions, the
Fig. 8. According to the six detectors, that is, amplitude normalised differences can be well described the differences
detector in time domain, Hurst detector in time domain, FD, between sea clutter and primary target. The normalised
Intercept, FFE and FDV detector in FRFT domain, we differences of the six detectors with HH polarisation are
define six normalised differences as follows bigger than those with VV polarisation. The four proposed
      detectors in FRFT domain achieve better normalised
    differences than the methods using the amplitude and Hurst
As − min Aci  hs − max hci 

dAmplitude =  
  , dHurst =     exponent in time domain with VV polarisation, which
   indicates the proposed detectors perform better in rather low
As − max Aci  hs − min hci 
SCR environment. The detection probability is then
     
    analysed with the false alarm probability Pfa = 10 − 2 shown
Ds − min Dci  Is − min Ici 
dD =    , d I =    in Table 4.
   After a great deal of simulations with the same dataset,
Ds − max Dci  Is − max Ici 
several conclusions may be made: (i) it is clearly noticed
     
    that the proposed detectors outperform the traditional
Es − max Eci  Vs − max Vci  methods; (ii) the FDV achieves the highest detection rate
dE =    , dV = 
 
 
 while the Pd for amplitude detector is the lowest among the
Es − min Eci  Vs − min Vci  six, indicating the lowest correct detection probability for
(36) the same false alarm probability. This is because the energy

648 IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rsn.2012.0116
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org

Fig. 10 Analysis of detection performances with HH polarisation (Dataset_17)

a–d Detection probability against fractal characteristics in FRFT domain for primary target
e–h False alarm rate against fractal characteristics in FRFT domain for sea clutter

IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651 649
doi: 10.1049/iet-rsn.2012.0116 & The Institution of Engineering and Technology 2013
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org
Table 4 Comparison of detection performances by the four proposed detectors (dataset_17, HH polarisation)
Sea clutter bin Primary target bin Normalised difference Detection probability (Pfa = 10 − 2)

HH VV HH VV HH VV HH VV

amplitude 0.8821 0.8937 0.7215 0.8655 0.4802 0.2358 69.2% 49.1%

hurst 0.1073 0.0407 0.3886 0.1356 0.5076 0.2767 72.7% 57.6%
FD 1.7369 1.7730 1.3517 1.5150 0.6253 0.5695 90.7% 83.2%
intercept − 3.2092 − 3.1010 − 5.8976 − 4.8360 0.4828 0.3559 74.1% 71.7%
FFE 0.5980 0.5880 1.4751 1.1670 0.6644 0.5504 86.4% 80.6%
FDV 0.0895 0.0741 0.4422 0.2704 0.8342 0.7747 93.6% 87.5%

Table 5 Computational complexity analysis (sea clutter bin of dataset_17, HH polarisation)

FD Intercept FFE FDV

value (N = 512) 1.9660 −0.4087 0.6004 0.1454

value (N = 1024) 1.8417 −1.0446 0.5629 0.1272
value (N = 2048) 1.8335 −0.9964 0.5557 0.1221
calculation time T when N = 1024a 0.0994 s 0.0994 s 0.1026 s 0.1057 s

a
Main configuration of the computer
CPU: Intel Core(TM)2 Duo E7400 2.8 GHz; RAM: 4 G DDR b 1333; Operating System: Windows XP SP3; Software: Matlab 2008a

of moving target is better accumulated after FRFT, which very short integration time. On the above conditions, the
improves the SCR. Then in the best FRFT domain it can required samples for the proposed four methods in this
achieve larger difference of fractal characteristic between paper are not enough for high detection performance and
sea clutter and moving target; (iii) the FD and FFE have the accurate estimation of moving state. Fortunately, the
similar detection performance because they both employ the Radon-Fourier transform (RFT) is proposed to deal with the
method of fractal curve fitting; and (iv) among the four ARU effect and is proven to be a generalised Doppler filter
methods, that is, FD, intercept, FFE and FDV, the FDV is bank [31]. As a result, more works can be done combining
more suitable for target detection in case of high sea state. the merits of RFT and the proposed methods in order to
solve the problems and restrictions in real applications.

4.3 Analysis of the computational complexity

5 Conclusions
Since the digital computation of FRFT is efficient and
requires only a sequence of FFTs with the overall In this paper, moving target detection within sea clutter is
computation complexity O(N log N) [29], it can be easily studied and four novel detectors are proposed based on the
implemented in real-time using currently available high fractal characteristics in FRFT domain with IPIX radar data.
speed FFT. Also it will save a lot of time using the grading The results of simulation with real sea clutter indicate that
iterative method to search for the best transform order [20, the detectors are effective and several useful remarks can be
24]. Since the fractal characteristics are estimated according concluded: (i) sea clutter of different polarisations in FRFT
to the fractal curves within the scale-invariant interval 25– domain has self-similarity properties within its
210, it needs about 1000 samples to achieve the true corresponding scale-invariant interval; (ii) the FD in FRFT
real-time processing, which is about 1 s with 1000 Hz domain calculated by FBM model is bigger than its
sampling frequency of IPIX radar. Table 5 gives the topological dimension; (iii) the biggest differences of fractal
computational complexity of calculating fractal characteristic between sea clutter bins and primary target
characteristic in FRFT domain with only sea clutter range bin appear in the best FRFT domain, which can be used for
bin for HH datasets. We notice that it can ensure the moving target detection; and (iv) the proposed detectors
requirements of efficiency and accuracy with only 1024 have better detection performances than the traditional
points. On the platform of popular PC, the FD and intercept amplitude detector and Hurst detector especially in low
need the same time about 0.0994 s because they just do the SCR environment.
linear fitting calculation while the FFE and FDV are a bit We have to admit that we do not give the detailed
time-consuming. It is worth noting that the four fractal mathematical derivation to understand exactly how the
characteristics in FRFT domain have certain errors because moving target return interact with sea clutter in FRFT
of the following effects: (i) the presence of noise; (ii) the domain, and further investigation is needed. However, we
best transform order estimation error; and (iii) the fitting know that when the two processes interact with each other
error of a straight line. the fractal characteristic in FRFT domain changes. Therefore
For detection of the far-range or weak target, the long-time we may be able to determine whether the signal consists of
coherent integration is needed in order to obtain more pulses only sea clutter or whether it consists of both sea clutter and
from target or improve the SCR gain. The range walk, that is, a moving target return especially in low observation
the linearly varied distance between target and radar of even a conditions. This opens the door to the possibility of using
slowly moving target will exceed several range bins, which is the fractal characteristic in FRFT domain of radar signals to
called the effect of across range unit (ARU). For a high-speed classify moving target or to distinguish it from sea clutter.
target, this kind of ARU effect may also be drawn even in a Since the methods do not require any training, it maybe

650 IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651
& The Institution of Engineering and Technology 2013 doi: 10.1049/iet-rsn.2012.0116
 17518792, 2013, 6, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-rsn.2012.0116, Wiley Online Library on [11/05/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
www.ietdl.org
readily developed into automated target detectors within sea 14 Guan, J., Liu, N.B., Zhang, J., Jie, S.: ‘Multifractal correlation
clutter for real-time operation. characteristic for radar detecting low-observable target in sea clutter’,
Signal Process., 2010, 90, (2), pp. 523–535
15 White, P.R., Locke, J.: ‘Performance of methods based on the fractional
6 Acknowledgments Fourier transform for the detection of linear frequency modulated
signals’, IET Signal Process., 2012, 6, (5), pp. 478–483
This work was in part by the National Natural Science 16 Namias, V.: ‘The fractional order Fourier transform and its application to
Foundation of China under grants nos. 61002045, quantum mechanics’, J. Inst. Appl. Math., 1980, 25, (2), pp. 241–265
17 Chen, X.L., Guan, J., He, Y.: ‘Sea clutter suppression and moving target
61179017 and 61201445, and by the Mountain Tai detection method based on clutter map cancellation in FRFT domain’.
Scholars of China. The authors wish to thank the Int. Conf. Radar, 2011, pp. 438–441
anonymous reviewers for their very helpful suggestions. 18 Tao, R., Zhang, N., Wang, Y.: ‘Analysing and compensating the effects
of range and Doppler frequency migrations in linear frequency
modulation pulse compression radar’, IET Radar Sonar Navig., 2011,
7 References 5, (1), pp. 12–22
19 Elgamel, S.A., Soraghan, J.: ‘Enhanced monopulse tracking radar using
1 Farina, A., Gini, F., Greco, M.V., Verrazzan, L.: ‘High resolution of
optimum fractional Fourier transform’, IET Radar Sonar Navig., 2011,
recorded live sea clutter data: statistical analysis data’, IEE Proc.
Radar Sonar Navig., 1997, 144, (3), pp. 121–130 5, (1), pp. 74–82
2 Javier, C.M., Javier, G.M., Álvaro, B.C., Alberto, A.L.: ‘Statistical 20 Chen, X.L., Guan, J.: ‘A fast FRFT based detection algorithm of
analysis of a high-resolution sea-clutter database’, IEEE Trans. multiple moving targets in sea clutter’. IEEE Int. Radar Conf., 2010,
Geosci. Remote Sens., 2010, 48, (4), pp. 2024–2037 pp. 402–406
3 Li, X., Xu, X.: ‘Numerical study on the phase and amplitude statistics of 21 Yuen, S.Y., Fong, C.K., Chan, K.L., Leung, Y.W.: ‘Fractal dimension
backscattered signals from time-evolving sea surfaces’, IET Radar estimation and noise filtering using Hough transform’, Signal
Sonar Navig., 2011, 5, (5), pp. 551–560 Process., 2004, 84, pp. 907–917
4 Chen, X.L., Wang, G.Q., Huang, Y., Guan, J.: ‘Study on the fractional 22 Chen, Y.Q., Sun, R.T., Zhou, A.H.: ‘An improved Hurst parameter
power spectrum of the scattered echoes from 1-D time-varying sea estimator based on fractional Fourier transform’. Int. Design
surface’. Asia-Pacific Conf. Antennas and Propagation, 2012, pp. Engineering Technical Conf. & Computers and Information in
321–322 Engineering Conf., 2007, pp. 1–11
5 Nunziata, F., Migliaccio, M., Gambardella, A.: ‘Pedestal height for sea 23 http://www.soma.ece.mcmaster.ca/ipix/dartmouth/datasets.html
oil slick observation’, IET Radar Sonar Navig., 2011, 5, (2), pp. 24 Guan, J., Chen, X.L., Huang, Y., He, Y.: ‘Adaptive fractional Fourier
103–110 transform-based detection algorithm for moving target in heavy sea
6 Ward, K.D., Watts, S.: ‘Use of sea clutter models in radar design and clutter’, IET Radar Sonar Navig, 2012, 6, (5), pp. 389–401
development’, IET Radar Sonar Navig., 2010, 4, (2), pp. 146–157 25 Sheng, H., Chen, Y.Q., Qiu, T.: ‘On the robustness of Hurst estimators’,
7 Shi, Y.L., Shui, P.L.: ‘Target detection in high-resolution sea clutter via IET Signal Process., 2011, 5, (2), pp. 209–225
block-adaptive clutter suppression’, IET Radar Sonar Navig., 2011, 5, 26 Hu, J., Gao, J.B., Posner, F.L., Zheng, Y., Tung, W.W.: ‘Target
(1), pp. 48–57 detection within sea clutter: a comparative study by fractal scaling
8 Gini, F., Greco, M.: ‘Texture modeling, estimation, and validation using analyses’, Fractals, 2006, 14, pp. 187–204
measured sea clutter data’, IEE Proc. Radar Sonar Navig., 2002, 149, 27 Gao, J.B., Cao, Y.H., Tung, W.W., Hu, J.: ‘Multiscale analysis of
(3), pp. 115–124 complex time series-integration of chaos and random fractal theory,
9 Guan, J., Zhang, X.L.: ‘Subspace detection for range and Doppler and beyond’ (Wiley Interscience, New York, 2007), p. 352
distributed targets with Rao and Wald tests’, Signal Process., 2011,
28 Martorella, M., Berizzi, F., Mese, E.D.: ‘On the fractal dimension of sea
91, pp. 51–60
surface backscattered signal at low grazing angle’, IEEE Trans.
10 Blacknell, D.: ‘Comparison of parameter estimators for K-distribution’,
Antennas Propag., 2004, 52, (5), pp. 1193–1204
IEE Proc. Radar Sonar Navig., 1994, 141, (1), pp. 45–52
11 Salmasi, M., Modarres-Hashemi, M.: ‘Design and analysis of fractal 29 Ozaktas, H.M., Arikan, O., Alper Kutay, M., Bozdaği, G.: ‘Digital
detector for high resolution radars’, Chaos Solitons Fractals, 2009, computation of the fractional Fourier transform’, IEEE Trans. Signal
40, pp. 2133–2145 Process., 1996, 44, (9), pp. 2141–2150
12 Lo, T., Leung, H., Litva, J., Haykin, S.: ‘Fractal characterization of 30 He, Y., Guan, J., Meng, X.W.: ‘Radar target detection and CFAR
sea-scattered signals and detection of sea-surface targets’, IEE Proc. processing’ (Tsinghua University Press, Beijing, China, 2011, 2nd
F, 1993, 140, (4), pp. 243–250 edn.), pp. 204–221
13 Hu, J., Tung, W.W., Gao, J.B.: ‘Detection of low observable targets 31 Xu, J., Yu, J., Peng, Y.N., Xia, X.G.: ‘Radon-Fourier transform for radar
within sea clutter by structure function based multifractal analysis’, target detection, I: generalized Doppler filter bank’, IEEE Trans. Aerosp.
IEEE Trans. Antennas Propag., 2006, 54, (1), pp. 136–143 Electron. Syst., 2011, 47, (2), pp. 1186–1202

IET Radar Sonar Navig., 2013, Vol. 7, Iss. 6, pp. 635–651 651
doi: 10.1049/iet-rsn.2012.0116 & The Institution of Engineering and Technology 2013