Sparse Discrete Fractional Fourier Transform and Its Applications
Sparse Discrete Fractional Fourier Transform and Its Applications
Sparse Discrete Fractional Fourier Transform and Its Applications
Abstract—The discrete fractional Fourier transform is a pow- results of the linear combination type do not match those of
erful signal processing tool with broad applications for nonsta- the continuous fractional Fourier transform (FrFT). In contrast,
tionary signals. In this paper, we propose a sparse discrete frac- the sampling based approach has a closed form expression
tional Fourier transform (SDFrFT) algorithm to reduce the com-
putational complexity when dealing with large data sets that are
with a relatively low complexity of , and the
sparsely represented in the fractional Fourier domain. The pro- transformed results approach that of the continuous FrFT [9].
posed technique achieves multicomponent resolution in addition Therefore, the sampling based DFrFT is widely employed in
to its low computational complexity and robustness against noise. engineering applications.
In addition, we apply the SDFrFT to the synchronization of high Among the various types of DFrFT algorithms, the lowest
dynamic direct-sequence spread-spectrum signals. Furthermore,
complexity is achieved by the Pei’s algorithm [9]. The Pei’s al-
a sparse fractional cross ambiguity function (SFrCAF) is devel-
oped, and the application of SFrCAF to a passive coherent loca- gorithm can be further optimized via a novel sub-linear algo-
tion system is presented. The experiment results confirm that the rithm for DFT named sparse Fourier transform (SFT) developed
proposed approach can substantially reduce the computation com- by Haitham et al. [10], [11]. When the input data have a large
plexity without degrading the precision. size with a sparse spectrum, this algorithm reduces the com-
Index Terms—Cross ambiguity function, global positioning plexity of DFT to , where stands for
system, passive bistatic radar, sparse discrete fractional Fourier the number of large coefficients in the frequency domain. Con-
transform. sider a wideband chirp signal with a sparse feature in the frac-
tional Fourier domain, to accelerate the time-frequency anal-
I. INTRODUCTION ysis of such signals, we propose an efficient scheme through
redesigning Pei’s algorithm by exploiting the advantage of the
1053-587X © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
Authorized licensed use limited to: Viettel Group. Downloaded on May 06,2022 at 07:36:05 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: SPARSE DISCRETE FRACTIONAL FOURIER TRANSFORM AND ITS APPLICATIONS 6583
The contribution of this paper is fivefold: (1) We propose the [21]. Define a flat window function , which is
concept and algorithm of SDFrFT; (2) We analyze its impor- a symmetric vector, . Let denote the
tant properties such as the capability of resolving multiple signal window length in the time domain. Suppose that
components; (3) We apply the SDFrFT to the fast synchroniza- is the frequency domain expression of ,
tion of high dynamic GPS signals; (4) We develop sparse frac- whose range obeys
tional cross ambiguity function (SFrCAF) to reduce the compu-
,
tational complexity of radar signal processing; (5) We apply the (5)
,
proposed SFrCAF to a PCL system to yield desirable results.
The rest of the paper is arranged as follows. In Section II, the where and are the truncation factors of the pass-
proposed SDFrFT algorithm is presented, and its relationship band and stopband, respectively, and denotes the
with the Pei’s algorithm and the SFT is discussed. Simulation extent of ripple oscillation. Define a signal
results and performance analysis of the proposed algorithm are , , then the support of sat-
given in Section III. In Section IV, we apply the SFrCAF to the isfies .
fast acquisition of high dynamic DSSS signals. In Section V, Step 4) Let be an exact divisor of integer . If ,
the principle of the proposed SFrCAF and its application to the construct a signal
PCL signal processing is demonstrated. The paper is concluded
in Section VI.
II. METHODOLOGY
(6)
A. SDFrFT Algorithm
1) Algorithm Flow: From a practice perspective, the compu- On the other hand, if , substitute the FFT
tational efficiency of an algorithm is a critical factor. The main operation with IFFT. Assume that is the fre-
steps of the proposed SDFrFT algorithm are as follows: quency domain expression of signal . It can be
Step 1) Construct the input signal of the SFT stage from the proved that [10], [29]
original input signal by a chirp multiplication.
Note that must be sparse in fractional Fourier (7)
domain and nonperiodic, and satisfy the Dirichlet
condition. (7) indicates that aliasing in the time domain cor-
responds to subsampling in the frequency domain.
(1) Store the value of and parameter employed
in (2).
where is the sampling interval of the input signal, Step 5) Define a hash function
is a real number representing the rotation angle of
FrFT. (8)
Step 2) To tear apart the nearby coefficients in the spectrum,
and an offset function
a permutation is adopted to reorder the signal’s fre-
quency domain . This process is conducted by (9)
modifying the time-domain signal as we do not
have access to the input signal’s Fourier spectrum, Step 6) Location loops: Define another set
which would require performing a DFT [21]. We
(10)
permutate the constructed signal as follows:
Authorized licensed use limited to: Viettel Group. Downloaded on May 06,2022 at 07:36:05 UTC from IEEE Xplore. Restrictions apply.
6584 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 62, NO. 24, DECEMBER 15, 2014
(15)
(16)
(13)
Step 9) By multiplying another chirp function to the estima- According to (17), the energy of
tion result obtained from the above steps, the will concentrate at
output of the SDFrFT algorithm is finally given by
(18)
(14) and it is proved that the best estimation precision can be ob-
where is the sampling interval of the output tained with . Then, can be correctly estimated from .
signal, and is the length of the DFrFT output. However, influenced by the input channel noise, the estima-
The detailed overall computation architecture for tion precision may not be sufficient for some application sce-
the situation is presented in Fig. 1. narios. For these cases, a fine search for the value of within
2) Selection of : With regard to the selection of the value a limited range around the estimated result is required. In the
of , there are two cases in practice: process, the selection of the step size mainly depends on
Case 1) The value of is already known. This kind of situ- the chirp rate resolution and application requirement. Here we
ation exists in many applications, for example, the derive the upper bound of under the constraint of chirp rate
matched filtering in the linear frequency modula- resolution . From (18) we know that
tion radar or in the synthetic aperture radar (SAR)
imaging. (19)
Case 2) The value of is unknown. For this case, we estimate
the value of by the discrete polynomial-phase
transform (DPT) method [22], [23]. The estimation Let be the time length of the signal, and choose . Then
precision can be further improved by searching with we can get
a finer step size within a limited range around the es-
timated value of . (20)
We explain how to choose the value of as follows.
An important method to estimate the rotation angle is the Comparing (1) with (17), we can find that
maximum likelihood estimation (MLE) technique [24]. In [25],
the discrete chirp Fourier transform, which is the discrete form
of the MLE, is proposed to estimate the chirp rate. However, due
to the exhausting two-dimensional maximization process of the
(21)
MLE, suboptimal methods are preferred. The phase unwrapping
method [26] is developed based on the finite difference operator, where denotes first-order derivative operator.
Authorized licensed use limited to: Viettel Group. Downloaded on May 06,2022 at 07:36:05 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: SPARSE DISCRETE FRACTIONAL FOURIER TRANSFORM AND ITS APPLICATIONS 6585
B. Relation to the Pei’s Sampling-Type Algorithm and the then locate and estimate the large coefficients in a manner sim-
Concept of SFT ilar to the sketching/streaming algorithms, where either iteration
The continuous FrFT [27] is defined as (22), at the bottom or interpolation, the expensive process in the previous methods,
of the page, where denotes the fractional Fourier domain is needed. This makes it possible for further improvement of the
frequency, is an integer, , and the phase of algorithm efficiency on the basis of the Pei’s sampling-type al-
is constrained in the range of . gorithm. Fortunately, it happens that the algorithm architecture
The Pei’s sampling-type algorithm [9] is derived based on of the Pei’s algorithm is suited for this kind of modification.
(22). First, the input and the output signals are directly sampled The revised Pei’s algorithm, termed SDFrFT, is designed for
by the intervals and , respectively. Second, to satisfy the the signals that meet the following descriptions: The signal is
reversible property, the sampling interval is restricted by non-stationary with a large scale, and is -sparse in the frac-
tional Fourier domain, where the signal size and the number
of large coefficients satisfy . This kind of signal is
(23) common in many applications, such as SAR signal processing
and nuclear magnetic resonance imaging.
Denote the length of input signal by . Then, the constraint
must be satisfied. We only discuss the situ-
ation in this paper. In this case, the form of DFrFT can be ob- III. PERFORMANCE OF PROPOSED SDFRFT
tained as shown in (24) at the bottom of the page.
Generally, if , the Pei’s sampling-type algorithm can A. Resolution Performance
be seen as two times of multiplication with chirp signals and one In the following an example is given to illustrate the resolu-
time of FFT. Therefore, the overall multiplication complexity of tion performance of the proposed SDFrFT in the multi-rotation
the Pei’s algorithm is . angle case. The initial frequencies of the four frequency com-
As the most efficient numerical algorithm of DFrFT, the Pei’s ponents are 100, 200, 300 and 300.1 Hz, respectively, and the
sampling-type algorithm is suited for a broad spectrum of ap- chirp rates of these components are 10, 11.85, 13.85 and 13.85
plications. However, the computational complexity will be high Hz/s, respectively. The sampling rate is , and the
when the data length is large, in which the FFT stage ac- data length is . The input signal is corrupted by a white
counts for a major proportion. When the signal is sparse, i.e., Gaussian noise, and the SNRs of the four frequency components
most of its coefficients are zero or negligible, it is recently re- are 12, 18, 24 and 24 dB, respectively. In the simulation
vealed that the computational complexity of DFT can be signifi- process, the number of computed large coefficients in the fre-
cantly reduced by a novel fast algorithm named SFT [10], which quency domain is set to . The loop number parameters
is far superior to the FFT. The key idea of SFT is to first parti- are set as and , respectively. The filter pa-
tion the frequency domain of the sparse signal into individual rameters are set as , , , and
buckets using a specially designed filter that is concentrated . The length of subsampled FFT is . In
both in time and frequency domains, which is obtained by con- every location loop, as many as maximum magnitudes
volving a Dolph-Chebyshev function with a box-car function, are searched out from .
,
(22)
,
,
, (24)
,
.
Authorized licensed use limited to: Viettel Group. Downloaded on May 06,2022 at 07:36:05 UTC from IEEE Xplore. Restrictions apply.
6586 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 62, NO. 24, DECEMBER 15, 2014
Fig. 2. The resolution performance of SDFrFT: (a) The frequency domain of the input signal. (b) (d) (f) The matched-order DFrFT of the four components
respectively. (c) (e) (g) The matched-order SDFrFT of the four components, respectively.
The simulation results are shown in Fig. 2, where Fig. 2(a) sparse component in the fractional Fourier domain can be
shows the frequency domain magnitude of the input signal, guaranteed by the proposed SDFrFT. On the other hand,
and the rest are arranged in 3 rows and 2 columns. The results in the estimated fractional Fourier domain, the components
in each row are obtained by setting the rotation angle in which do not behave sparse and focused will be estimated
the fractional Fourier domain such that one of the frequency as dense fractional spectral lines with lower amplitude as
components is focused. The simulation results demonstrate compared with the correctly estimated large values. The
that, in this multi-rotation angle case, the estimation precision enlargements in Figs. 2(f) and (g) reveal the local details of
of the fractional frequency and the amplitude value of the the immediately adjacent spectral lines with the same chirp
Authorized licensed use limited to: Viettel Group. Downloaded on May 06,2022 at 07:36:05 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: SPARSE DISCRETE FRACTIONAL FOURIER TRANSFORM AND ITS APPLICATIONS 6587
Fig. 3. Comparison of the computational complexity between conventional Fig. 4. Robustness vs. SNR.
DFrFT and SDFrFT approaches.
(25)
Fig. 4 plots the average error of the SDFrFT obtained from
complex multiplication operations, where the function the numeric simulation results, which confirms the robustness
expresses the cardinality of a set. of the algorithm under noisy circumstance.
The comparative result of the computation complexity be-
tween SDFrFT and DFrFT is shown in Fig. 3. Note that the re- IV. APPLICATION TO THE SYNCHRONIZATION OF HIGH
sult is simply based on the number of complex multiplications DYNAMIC DSSS SIGNAL
in the unoptimized algorithm flow as depicted in (25). In the DSSS communication and navigation systems [30] have the
simulation process, we assume the number of computed large advantages of low spectral density, high information security
coefficients in the frequency domain to be . The loop and resistance to jamming, and are easy to realize multiple ac-
number parameters are set as and , respec- cess communications and high-precision measurements. Hence,
tively. It can be seen that, when the data length is increased to they are widely used in both civilian and military applications.
a moderate level, the advantage of the SDFrFT over the DFrFT The well-known GPS [31] is an example of DSSS system. To
in computational complexity becomes more evident. ensure correct despreading and demodulation, signal synchro-
The proposed SDFrFT algorithm is designed based on the nization is needed at the receiver end. The synchronization in
SFT theory [10] and code versions 1 and 2 [28]. As is pointed a DSSS system normally consists of two steps: acquisition and
out in [21], the computational complexity of these two versions tracking, where acquisition is the prerequisite for tracking. GPS
closely correlates with the signal size . The version 3 and receivers are now frequently used in the field of aerospace engi-
4 codes are not published yet. However, the theories of these neering. The commonly occurring high dynamic relative motion
two versions have been described in [29], and some analysis between the navigation satellite and the receiver platform will
of version 3 can be found in [21]. It is proved that, in code induce acute variations in the phase of the carrier. The high ve-
version 3, the correlation between signal size and computational locity and acceleration of the motion are characterized as a large
complexity becomes less significant. Thus it is reasonable to Doppler shift and its derivative.
expect that the SDFrFT based on SFT code version 3 will not Conventional FFT based fast acquisition approaches
necessarily require such a large signal size to exceed the classic [32]–[34] solely compensate for the Doppler shift compo-
algorithm. nent caused by the high velocity, whereas the impact of the
On the other hand, our algorithm is already computation- change rate of the Doppler frequency is ignored. However,
ally faster for signals with length around , which is a quite if the change rate of the Doppler frequency is high, then the
common size in many application scenarios. Some of the exam- spectral expansion results in a reduction of the signal peak, and
ples will be presented in Section IV and Section V. thereby, makes the acquisition difficult, especially with an ex-
tremely low SNR. In this section, we propose a fast acquisition
C. Algorithm Robustness method based on the SDFrFT to synchronize high dynamic
As illustrated in [10], the SFT’s reduced runtime does not DSSS signals. By compensating the quadratic phase term with
compromise its robustness to noise. The robustness to noise of SDFrFT, a notably enhanced acquisition performance can be
the proposed algorithm is examined by simulations. Let achieved.
Authorized licensed use limited to: Viettel Group. Downloaded on May 06,2022 at 07:36:05 UTC from IEEE Xplore. Restrictions apply.
6588 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 62, NO. 24, DECEMBER 15, 2014
A. Principle
Fig. 7. The relationship between acceleration and transform order.
Let and be the initial velocity and acceleration of the
receiver platform relative to the transmitter respectively. Let adopted to further reduce the computational complexity. When
denote the carrier wavelength, so the Doppler frequency at searching for the Doppler frequency , the input signal
time can be written as is first multiplied by to obtain , where
. As subsampling a signal in the frequency domain
(27)
is equivalent to aliasing it in the time domain, and vice versa,
Thus, the received signal can be expressed as signal is aliased to obtain its subsampled version as
(29)
Authorized licensed use limited to: Viettel Group. Downloaded on May 06,2022 at 07:36:05 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: SPARSE DISCRETE FRACTIONAL FOURIER TRANSFORM AND ITS APPLICATIONS 6589
A. Principle
Fig. 9. Simulation results of GPS acquisition using different approaches: In this section, the definition and the derivation of the novel
(a) FFT. (b) DFrFT. (c) SFT. (d) SDFrFT. SFrCAF are discussed in a PCL radar scenario.
A PCL system utilizes the direct wave signal and the target
echo signal to calculate CAF. The CAF is calculated as
and quadrature (Q) channels are conveyed to the acquisition and
tracking module to reconstruct the navigation message. (31)
The following simulation is based on the coarse/acquisition
(C/A) code. The C/A code is a pseudo-random (PN) binary se-
quence, which is transmitted at a rate of 1.023 Mchips/s with where is the echo signal received by the surveillance an-
the information data rate of 50 bps. The PN sequences only tenna, is the direct path signal received by the reference
strongly correlate when they are exactly aligned. We choose the antenna, and denote the time delay and the Doppler shift,
L1 wave band as the carrier frequency, whose center frequency respectively, and is the integration time. It is worth noting
is 1575.42 MHz. that the CAF in (31) can be interpreted as the Fourier transform
The other simulation parameters are listed in Table I. Fig. 9 of the product of the delayed version of and the conjugate
shows the simulation results using FFT, DFrFT, SFT and SD- of .
FrFT. It can be seen that, with the acceleration de-chirped, a The discrete definition of CAF can be written as follows:
more concentrating correlation peak can be obtained by using
FrFT approach than by FFT approach. The SDFrFT method (32)
only outputs the most significant peak, and the algorithm per-
forms well even in a relatively low SNR environment. By re- where and are the sampled echo and reference sig-
ferring to Fig. 3, we draw the conclusion that the proposed nals, respectively. In addition, is the number of delay bins,
method can greatly increase both the probability and the speed is the number of Doppler shift bins, and refers to the in-
of acquisition. tegration length of data. Then, the integration time of CAF is
Authorized licensed use limited to: Viettel Group. Downloaded on May 06,2022 at 07:36:05 UTC from IEEE Xplore. Restrictions apply.
6590 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 62, NO. 24, DECEMBER 15, 2014
Substituting (34) into (35) and (36), then performing Taylor se-
ries expansion at with the quadratic and higher terms
neglected, we can get
(37)
(36) (41)
(34)
Authorized licensed use limited to: Viettel Group. Downloaded on May 06,2022 at 07:36:05 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: SPARSE DISCRETE FRACTIONAL FOURIER TRANSFORM AND ITS APPLICATIONS 6591
Fig. 11. Simulation results of coherent integration using different approaches: (a) CAF result with FFT. (b) Side view of Doppler-amplitude section of the CAF
map in (a). (c) SFrCAF result with SFT, namely . (d) Side view of Doppler-amplitude section of the SFrCAF map in (c). (e) SFrCAF result, where Doppler
migration has been compensated at . (f) Side view of Doppler-amplitude section of the SFrCAF map in (e).
For the additional dimension to the conventional CAF, the In the case where a priori information suggests that the target
FrCAF is more time-consuming, and thus greatly limits its remains in its state of radial uniform motion during the integra-
application range. By adopting the proposed SDFrFT based tion time, the SDFrFT process can be further simplified to the
method, a significant reduction in the overall runtime can be SFT.
achieved. Thus, the proposed -order SFrCAF can be expressed
as
B. Algorithm Verification
(42)
The aforementioned algorithm is validated by the following
where denotes the -order SDFrFT operation to simulation and real data experiment, where the digital video
signal . broadcast signal is adopted as the non-cooperative transmitted
Authorized licensed use limited to: Viettel Group. Downloaded on May 06,2022 at 07:36:05 UTC from IEEE Xplore. Restrictions apply.
6592 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 62, NO. 24, DECEMBER 15, 2014
Fig. 12. Real data experiment results of coherent integration using different approaches: (a) CAF result of real data. (b) Side view of Doppler-amplitude section
of the CAF map in (a). (c) SFrCAF result with SFT, namely . (d) Side view of Doppler-amplitude section of the SFrCAF map in (c). (e) SFrCAF result of
real data, where . (f) Side view of Doppler-amplitude section of the SFrCAF map in (e).
Authorized licensed use limited to: Viettel Group. Downloaded on May 06,2022 at 07:36:05 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: SPARSE DISCRETE FRACTIONAL FOURIER TRANSFORM AND ITS APPLICATIONS 6593
VI. CONCLUSION
The objective of this paper was to develop a numerical algo-
rithm for the fast computation of DFrFT when the signal spec-
trum can be sparsely represented in the fractional Fourier do-
main. By recurring to the merit of SFT, we have redesigned the
Pei’s algorithm and proposed a novel approach, which signif-
icantly outperforms the existing algorithms in the runtime as-
pect. The application of the proposed SDFrFT algorithm for the
Fig. 13. Superposition of detection results over data files sequence.
fast synchronization of the high dynamic DSSS signal was then
presented. We have also proposed a SFrCAF for radar signal
processing and applied it to the coherent integration in a PCL
of the bistatic velocity are estimated as 437.2 Hz, 501.8 , system. The simulation results clearly demonstrated the appli-
and , respectively. cability of the proposed algorithm to the fast analysis of non-sta-
The effectiveness of the proposed SFrCAF is verified by the
tionary signals with a large size and sparse spectrum in the frac-
performance as shown in Fig. 12, where the motion parameters
tional Fourier domain. Our future work will concentrate on fur-
of the target are accurately estimated by SDFrFT, and a fine de-
ther optimizing the algorithm, and generalizing the method to
tection performance is achieved with a notably reduced compu-
the case of the discrete linear canonical transform.
tational complexity. By comparing the algorithm performances
shown in Figs. 12(d) and (f), the target’s amplitude in Fig. 12(f)
ACKNOWLEDGMENT
is approximately 4.9 dB higher than that in Fig. 12(d). There-
fore, we can conclude that the peak energy of the accelerating The authors would like to thank the anonymous reviewers
target is more concentrated with SDFrFT than with SFT, so that for their valuable comments, which have helped improve the
the target can be better distinguished from interference. It is ra- quality and clarity of this paper. Tao Shan would also like to
tional to predict that in CAF application, the advantage of SD- thank the China Scholarship Council for support.
FrFT over SFT will be more pronounced when the acceleration
is higher.
In the following, we further demonstrate target detection over REFERENCES
multiple coherent processing intervals by performing SFrCAF [1] S. C. Pei and W. L. Hsue, “Random discrete fractional Fourier trans-
form,” IEEE Signal Process. Lett., vol. 16, no. 12, pp. 1015–1018, Dec.
on a sequence of 9 consecutive data files. The length of each
2009.
data file is 0.5 s, i.e., samples, in which the first [2] J. W. Cooley and J. W. Tukey, “An algorithm for the machine compu-
3932160 samples are utilized. Then we draw the superposi- tation of complex Fourier series,” Math. Comput., vol. 19, no. 90, pp.
tion of the processing results over these data files plotted in 297–301, Apr. 1965.
Fig. 13, yielding the trajectory of the target that moves away [3] S. C. Pei, C. C. Tseng, and M. H. Yeh, “A new discrete fractional
Fourier transform based on constrained eigendecomposition of DFT
from the radar. In particular, the interval between the neigh- matrix by Largrange multiplier method,” IEEE Trans. Circuits Syst.
boring trace points matches the product of the target Doppler II, Analog Digit. Signal Process., vol. 46, no. 9, pp. 1240–1245, Sep.
and the length of data file, and the changing rate of the Doppler 1999.
frequency also matches the estimated acceleration. On the other [4] C. Candan, M. A. Kutay, and H. M. Ozaktas, “The discrete fractional
Fourier transform,” IEEE Trans. Signal Process., vol. 48, no. 5, pp.
hand, other non-target components randomly scatter on the CAF
1329–1337, May 2000.
surface, and no connection between them can be observed. The [5] M. T. Hanna, “Fractional discrete Fourier transform of type IV based
offline data detection results have also been compared against on the eigenanalysis of a nearly tridiagonal matrix,” Digit. Signal
the recorded log of the ADS-B receiver in our experimental data Process., vol. 22, no. 6, pp. 1095–1106, Dec. 2012.
acquisition system to verify their consistency with the ground [6] B. Santhanam and J. H. McClellan, “The discrete rotational Fourier
transform,” IEEE Trans. Signal Process., vol. 44, no. 4, pp. 994–998,
truth. As such, through the superposition of the consecutive pro- Apr. 1996.
cessing results, we are able to clearly identify and track a weak [7] T. Erseghe, P. Kraniauskas, and G. Cariolaro, “Unified fractional
target in the presence of other strong echo components. Fourier transform and sampling theorem,” IEEE Trans. Signal
To summarize, the proposed SFrCAF is applicable to the Process., vol. 47, no. 12, pp. 3419–3423, Dec. 1999.
[8] H. M. Ozaktas, O. Ankan, M. A. Kutay, and G. Bozdagt, “Digital
detection scenarios where the data length is sufficiently large
computation of the fractional Fourier transform,” IEEE Trans. Signal
and the radial acceleration remains roughly stable during the Process., vol. 44, no. 9, pp. 2141–2150, Sep. 1996.
integration time. When SFrCAF is adopted, a relatively higher [9] S. C. Pei and J. J. Ding, “Closed-form discrete fractional and affine
integration gain can be obtained with a faster acceleration rate Fourier transforms,” IEEE Trans. Signal Process., vol. 48, no. 5, pp.
of the target, and an accelerated algorithm computation can be 1338–1353, May 2000.
[10] H. Hassanieh, P. Indyk, D. Katabi, and E. Price, “Simple and prac-
achieved with a larger data length. From the implementation tical algorithm for sparse Fourier transform,” in Proc. 23rd Annu.
perspective, the inherent data-parallelism of the proposed ACM-SIAM Symp. Discrete Algorithms, Kyoto, Japan, Jan. 2012, pp.
SFrCAF, which is embodied in the computing patterns of 1183–1194.
Authorized licensed use limited to: Viettel Group. Downloaded on May 06,2022 at 07:36:05 UTC from IEEE Xplore. Restrictions apply.
6594 IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 62, NO. 24, DECEMBER 15, 2014
[11] A. C. Gilbert, P. Indyk, M. Iwen, and L. Schmidt, “Recent develop- [35] K. P. Bentz, “Computation of the cross ambiguity function using
ments in the sparse Fourier transform: A compressed Fourier transform perfect reconstruction DFT filter banks,” Ph.D. dissertation, George
for big data,” IEEE Signal Process. Mag., vol. 31, no. 5, pp. 91–100, Mason Univ., Fairfax, VA, USA, 2007.
Sep. 2014. [36] J. E. Palmer, H. A. Harms, S. J. Searle, and L. M. Davis, “DVB-T
[12] J. Markel, “FFT pruning,” IEEE Trans. Audio Electroacoust., vol. 19, passive radar signal processing,” IEEE Trans. Signal Process., vol. 61,
no. 4, pp. 305–311, Dec. 1971. no. 8, pp. 2116–2126, Apr. 2013.
[13] R. Tao, B. Z. Li, and Y. Wang, “Spectral analysis and reconstruction for [37] H. D. Griffiths and C. J. Baker, “Passive coherent location radar sys-
periodic nonuniformly sampled signals in fractional Fourier domain,” tems. Part 1: Performance prediction,” Proc. Inst. Electr. Eng.—Radar
IEEE Trans. Signal Process., vol. 55, no. 7, pp. 3541–3547, Jul. 2007. Sonar Navig., vol. 152, no. 3, pp. 153–159, Jun. 2005.
[14] R. Tao, F. Zhang, and Y. Wang, “Fractional power spectrum,” IEEE [38] H. D. Griffiths, C. J. Baker, H. Ghaleb, R. Ramakrishnan, and E.
Trans. Signal Process., vol. 56, no. 9, pp. 4199–4206, Sep. 2008. Willman, “Measurement and analysis of ambiguity functions of off-air
[15] R. Tao, B. Deng, W. Q. Zhang, and Y. Wang, “Sampling and sam- signals for passive coherent location,” Electron. Lett., vol. 39, no. 13,
pling rate conversion of band limited signals in the fractional Fourier pp. 1005–1007, Jun. 2003.
transform domain,” IEEE Trans. Signal Process., vol. 56, no. 1, pp. [39] R. Tolimieri and S. Winograd, “Computing the ambiguity surface,”
158–171, Jan. 2008. IEEE Trans. Acoust., Speech, Signal Process., vol. 33, no. 5, pp.
[16] R. Tao, F. Zhang, and Y. Wang, “Sampling random signals in a frac- 1239–1245, May 1985.
tional Fourier domain,” Signal Process., vol. 91, no. 6, pp. 1394–1400, [40] C. L. Yatrakis, “Computing the cross-ambiguity function—A review,”
Jun. 2011. B.S. thesis, State Univ. of New York, Binghamton, NY, USA, 2005.
[17] R. Tao, X. M. Li, Y. L. Li, and Y. Wang, “Time delay estimation of [41] Y. Feng, T. Shan, Z. H. Zhuo, and R. Tao, “The migration compen-
chirp signals in the fractional Fourier domain,” IEEE Trans. Signal sation methods for DTV based passive radar,” in Proc. IEEE Radar
Process., vol. 57, no. 7, pp. 2852–2855, Jul. 2009. Conf., Ottawa, ON, Canada, May 2013, pp. 1–4.
[18] R. Tao, Y. L. Li, and Y. Wang, “Short-time fractional Fourier transform [42] D. W. O’Hagan, H. Kuschel, M. Ummenhofer, J. Heckenbach, and J.
and its applications,” IEEE Trans. Signal Process., vol. 58, no. 5, pp. Schell, “A multi-frequency hybrid passive radar concept for medium
2568–2580, May 2010. range air surveillance,” IEEE Aerosp. Electron. Syst. Mag., vol. 27, no.
[19] E. Sejdic, I. Djurovic, and L. Stankovic, “Fractional Fourier trans- 10, pp. 6–15, Oct. 2012.
form as a signal processing tool: An overview of recent developments,”
Signal Process., vol. 91, no. 6, pp. 1351–1369, June 2011. Shengheng Liu (S’14) was born in 1987. He was
[20] A. Omid, H. Ezz, H. Haitham, A. Abhinav, K. Dina, C. P. Anantha, admitted to the Ph.D. program directly from B.S. de-
and S. Vladimir, “A 0.75-million-point fourier-transform chip for gree, exempting from the qualification examination
frequency-sparse signals,” in IEEE Int. Solid-State Circuits Conf. in the year 2010 for his exceptional performance
Dig. Tech. Papers (ISSCC), San Francisco, CA, USA, Feb. 2014, pp. during undergraduate years in Beijing Institute of
458–459. Technology, and he is currently working toward
[21] J. Schumacher, “High performance sparse fast Fourier rransform,” the Ph.D. degree. His research interests include
M.S. thesis, ETH Zurich, Zurich, Switzerland, 2013. fractional Fourier transform and passive bistatic
[22] S. Peleg and B. Friedlander, “The discrete polynomial-phase trans- radar signal processing.
form,” IEEE Trans. Signal Process., vol. 43, no. 8, pp. 1901–1914,
Aug. 1995.
[23] S. Peleg and B. Friedlander, “Multicomponent signal analysis using
the polynomial-phase transform,” IEEE Trans. Aerosp. Electron. Syst.,
vol. 32, no. 1, pp. 378–387, Jan. 1996. Tao Shan was born in 1969. He received his B.S.
[24] T. Abotzoglou, “Fast maximum likelihood joint estimation of fre- degree from Xidian University, Xi’an, in 1991 and
quency and frequency rate,” IEEE Trans. Acoust., Speech Signal Ph.D. degree from Beijing Institute of Technology
Process., vol. 22, no. 6, pp. 708–715, Jun. 1986. in 2004. He is an associate professor with the School
[25] X. G. Xia, “Discrete chirp-Fourier transform and its application to of Information and Electronics, Beijing Institute of
chirp rate estimation,” IEEE Trans. Signal Process., vol. 48, no. 11, Technology. Currently, he is a visiting scholar at the
pp. 3122–3133, Nov. 2000. Center for Advanced Communications, Villanova
[26] P. M. Djuric and S. M. Kay, “Parameter estimation of chirp signals,” University. He was a recipient of the first prize of
IEEE Trans. Signal Process., vol. 38, no. 12, pp. 2118–2126, Dec. science and technology progress awarded by the
1990. Ministry of Education in 2006 and 2007 respectively.
[27] L. B. Almeida, “The fractional Fourier transform and time-frequency His research interests include radar signal processing
representations,” IEEE Trans. Signal Process., vol. 42, no. 11, pp. and complex information system theory.
3084–3091, Nov. 1994.
[28] L. B. Almeida, Dec. 2013, The Code sFFT-1.0–2.0 Package [Online].
Available: http://groups.csail.mit.edu/netmit/sFFT/sFFT-1.0-2.0.tar.gz
[29] H. Hassanieh, P. Indyk, D. Katabi, and E. Price, “Nearly optimal sparse Ran Tao (M’00–SM’04) was born in 1964. He re-
Fourier transform,” in Proc. 44th Symp. Theory Comput., New York, ceived the B.S. degree from Electronic Engineering
NY, USA, Jan. 2012, pp. 563–578. Institute of PLA, Hefei, in 1985 and the M.S. and
[30] A. J. Viterbi, CDMA Principles of Spread Spectrum Communication. Ph.D. degrees from Harbin Institute of Technology,
Reading, MA, USA: Addison-Wesley, 1995. Harbin, in 1990 and 1993, respectively. In 2001,
[31] B. Hofmann-Wellenhof, H. Lichtenegger, and J. Collins, Global Posi- he was a senior visiting scholar in the University of
Michigan, Ann Arbor. He is currently a Professor
tioning System: Theory and Practice. Vienna, Italy: Springer, 1993.
with the School of Information and Electronics,
[32] S. M. Spangenberg, I. Scott, S. McLaughlin, G. J. Povey, D. G. Cruick-
Beijing Institute of Technology, Beijing, China. He
shank, and P. M. Grant, “An FFT-based approach for fast acquisition
was a recipient of National Science Foundation of
in spread spectrum communication systems,” Wireless Pers. Commun., China for Distinguished Young Scholars in 2006,
vol. 13, no. 1–2, pp. 27–55, May 2000. and a Distinguished Professor of Changjiang Scholars Program in 2009. He
[33] D. Akopian, “Fast FFT based GPS satellite acquisition methods,” Proc. has been a Chief Professor of the Creative Research Groups of the National
Inst. Electr. Eng.–Radar Sonar Navig., vol. 152, no. 4, pp. 277–286, Natural Science Foundation of China since 2014, and he was a Chief Professor
Aug. 2005. of the Program for Changjiang Scholars and Innovative Research Team in
[34] H. Hassanieh, F. Adib, D. Katabi, and P. Indyk, “Faster GPS via the University during 2010 to 2012. He is currently the Vice Chair of IEEE China
sparse Fourier transform,” in Proc. ACM MobiCom, Istanbul, Turkey, Council. He is also the Vice Chair of the International Union of Radio Science
Aug. 2012, pp. 353–364. (URSI) China Council and a Member of Wireless Communication and Signal
Authorized licensed use limited to: Viettel Group. Downloaded on May 06,2022 at 07:36:05 UTC from IEEE Xplore. Restrictions apply.
LIU et al.: SPARSE DISCRETE FRACTIONAL FOURIER TRANSFORM AND ITS APPLICATIONS 6595
Processing Commission of URSI. He was a recipient of the first prize of science Guo Zhang was born in 1989. She received her B.S.
and technology progress in 2006, 2007, respectively, and the first prize of and M.S. degree from Beijing Institute of Technology
natural science in 2013, both awarded by the Ministry of Education. His current in 2011 and 2014, respectively. She is currently with
research interests include fractional Fourier transform and its applications, Beijing Telecom Planning & Designing Institute Co.,
theory and technology for radar and communication systems. He has 3 books Ltd. Her research interest is radar signal processing.
and more than 100 peer-reviewed journal articles.
Authorized licensed use limited to: Viettel Group. Downloaded on May 06,2022 at 07:36:05 UTC from IEEE Xplore. Restrictions apply.