Optik 127 (2016) 2613–2616

Contents lists available at ScienceDirect

Optik
journal homepage: www.elsevier.de/ijleo

A generalized convolution theorem for the special affine Fourier

transform and its application to filtering
Xiyang Zhi a , Deyun Wei b,∗ , Wei Zhang a
a
Research Center for Space Optical Engineering, Harbin Institute of Technology, Harbin 150001, China
b
School of Mathematics and Statistics, Xidian University, Xi’an 710071, China

a r t i c l e i n f o a b s t r a c t

Article history: The special affine Fourier transform (SAFT), which is a time-shifted and frequency-modulated version of
Received 1 June 2015 the linear canonical transform (LCT), has been shown to be a powerful tool for signal processing and optics.
Accepted 23 November 2015 Many properties for this transform are already known, but an extension of convolution theorem of Fourier
transform (FT) is still not having a widely accepted closed form expression. The purpose of this paper is
Keywords: to introduce a new convolution structure for the SAFT that preserves the convolution theorem for the FT,
Special affine Fourier transform
which states that the FT of the convolution of two functions is the product of their Fourier transforms.
Convolution and product theorem
Moreover, some of well-known results about the convolution theorem in FT domain, fractional Fourier
Filtering
transform (FRFT) domain, LCT domain are shown to be special cases of our achieved results. Last, as an
application, utilizing the new convolution theorem, we investigate the multiplicative filter in the SAFT
domain. The new convolution structure is easy to implement in the designing of filters.
© 2015 Elsevier GmbH. All rights reserved.

1. Introduction
The special affine Fourier transform (SAFT) [1–4], also called the
offset linear canonical transform (OLCT) [4] or the inhomogeneous
canonical transform [3], is a six-parameter (a, b, c, d, u0 , ω0 ) class
of linear integral transform. The SAFT encompasses a number of
important transform in digital signal processing and optical system
modeling. The well-known signal processing operations, such as
the Fourier transform (FT), the offset FT [3,4], the fractional Fourier
transform (FRFT) [5,6], the offset FRFT [3,4], the Fresnel transform
[7], the linear canonical transform (LCT) [8–10] and the scaling
operations are all special case of the SAFT. With the progression of
LCT theory, SAFT also has evolved as an interesting tool. The SAFT
is more general and flexible than the original LCT for its two extra
parameter u0 and ω0 . Recently, along with applications of the LCT in
the signal processing community [5,9,11–15], the role of the SAFT
for signal processing has also been noticed. It has found many appli-
cations in optics, signal processing, and many other applications
[3,4,16–18].
With intensive research of the SAFT, many properties have been
found including time shift, phase shift, scaling, differentiation, inte-
gration and so on [3,16]. Simultaneously, as the generalization of FT,
∗ Corresponding author. Tel.: +86 18291835435.
E-mail addresses: zhixiyang@hit.edu.cn (X. Zhi), dywei@xidian.edu.cn (D. Wei),
wzhang6262@hit.edu.cn (W. Zhang).
the relevant theory of SAFT has been developed including the con-
volution theorem, uncertainty principle, sampling theory and so
on [16–18], which are generalizations of the corresponding prop-
erties of the FT, FRFT and LCT [3,9,19–28]. Conventional convolution
operations for FT are fundamental in the theory of linear time-
invariant (LTI) system [9]. The output of any continuous-time LTI
system is found via the convolution of the input signal with the sys-
tem impulse response. As the SAFT has found wide applications in
optic and signal processing fields, it is theoretically interesting and
practically useful to consider the convolution theory in the SAFT
domain. However, the convolution theorems don’t have the ele-
gance and simplicity comparable to that of the FT, which states
that the FT of the convolution of two functions is the product of
their Fourier transforms.
Convolution theorem for a linear integral transform can be for-
mulated in several ways. Recently, Xiang and Qin [16] introduced
a new convolution operation that is more suitable for the SAFT and
by which the SAFT of the convolution of the two functions is the
product of their SAFTs and a phase factor. However, on the one
hand the convolution theorem for the SAFT derived in [16] with
the modified convolution operation also contains an extra chirp
factor and hence does not exactly parallel the theorem given by
FT. On the other hand, there possesses different chirp multiplica-
tions, which are difficult to implement in the engineering based
on it is nearly impossible to generate a chirp signal accurately. In
this paper, we propose a new convolution structure for the SAFT,
which is different from the convolution structure derived in [16].

http://dx.doi.org/10.1016/j.ijleo.2015.11.211
0030-4026/© 2015 Elsevier GmbH. All rights reserved.
2614 X. Zhi et al. / Optik 127 (2016) 2613–2616

Table 1 Property 1. The space shift property

Some of the specific cases of the saft.
2 /2
Parameter A Corresponding transform
S A [f (t − )] (u) = FA (u − a) e−j(ac )+jc(u−u0 )+jaω0 (4)

A = (a, b, c, d, 0, 0) Linear canonical transform Property 2. The phase shift property

(LCT)   2 /2
A = (cos , sin , − sin , cos , u0 , ω0 ) Offset fractional Fourier S A f (t) ejt (u) = FA (u − b) e−j(bd )+jd(u−u0 )+jbω0 (5)
transform (OFRFT)
A = (cos , sin , − sin , cos , 0, 0) Fractional Fourier transform Property 3. The space shift and phase shift properties
(FRFT) A
 j t

A = (0, 1, − 1, 0, u0 , ω0 ) Offset Fourier transform (OFT)
S f (t − ) e (u)
A = (0, 1, − 1, 0, 0, 0) Fourier transform (FT) 2 +bd 2 /2
A = (1, b, 0, 1, 0, 0) Fresnel transform = FA (u − a − b) e−j(ac )+j(c+d)(u−u0 )−jbc +j(a+b)ω0
A = (1, 0, 0, 1, u0 , 0) Time shift
A = (1, 0, 0, 1, 0, ω0 ) Frequency shift (6)
A = (d−1 , 0, 0, d, 0, 0) Time scaling

2.2. The convolution theory

Based on the expression for the generalized translation in the SAFT
Convolution and correlation operations are fundamental in the
domain, the generalized convolution theorem can be derived in
theory of LTI system. Moreover, convolution and correlation are
the SAFT domain, which shows that the generalized convolution of
widely used in signal processing, as well as in optics, in pattern
two signals in time domain is equivalent to simple multiplication
recognition or in the description of image formation with inco-
of their SAFTs in the SAFT domains. This result is an extension of
herent illumination [16,24–30]. The convolution operation in FT
the convolution theorem from the FT to the SAFT domain, and can
domain is defined as
be more useful in practical analog filtering in the SAFT domain. We
also show that the convolution theorem in FT or FRFT domain can 
+∞

be looked as special cases of our achieved results. f (t) ⊗ g (t) = f () g (t − ) d (7)
The rest of the paper is organized as follows. In Section 2,
−∞
we provide a brief review of the SAFT and convolution theory. A
new convolution theorem for SAFT is derived based on general- FT
f (t) ⊗ g (t) ←−F (u) G (u) (8)
ized translation in Section 3. In Section 4, the multiplicative filter is
investigated in the SAFT domain. The paper is concluded in Section where ‘⊗’ denotes the conventional convolution operation.
5. Recently, the convolution theorem has been derived in the SAFT
domain by Xiang [16] as follows:
2. Preliminaries  
+∞
1 j(d/2b)u2
z (t) = e 0 f () g (t − ) e−j(a/b)(t−) dt (9)
2.1. Special affine Fourier transform (SAFT) j2b
−∞
2 +2u(du −bω )
The SAFT was introduced by Abe and Sheridan [1,2] who studied S [z (t)] (u) = FA (u) GA (u) e(j/2b)[−du
A 0 0 ] (10)
a transformation in phase space that was associated with a general,
inhomogeneous, lossless linear mapping. The SAFT of a signal f (t) The SAFT of the convolution of the two functions is the product of
with real parameter A = (a, b, c, d, u0 , ω0 ) is defined as [1–4] their SAFTs and a phase factor. However, the convolution theorem
⎧ +∞
doesn’t have the elegance and simplicity comparable to that of the
⎨ f (t) hA (u, t) dt, b=
/ 0
FT.
A
FA (u) = S [f (t)] (u) =
⎩√−∞j(cd/2)(u−u0 )2 +jω0 u 3. New convolution structure for the SAFT based on
de f [d (u − u0 )] , b=0
(1) generalized translation

where In this section, we seek to modify the ordinary convolution

   structure using the generalized translation based on the kernel of
1 j(1/2b) d(u2 +u2 )−2u(du0 −bω0 )+2t(u0 −u)+at 2
the SAFT. We will derive the new convolution theorem for the SAFT
hA (u, t) = e 0 (2)
j2b based on the generalized translation. The new convolution theorem
has the elegance and simplicity comparable to the classical result
a, b, c, d, u0 , ω0 are real numbers satisfying ad − bc = 1. We only
for the FT. Some of the well-known results about the convolution
consider the case of b = / 0, since the SAFT is just a chirp multipli-
theorem in FT domain, FRFT domain, LCT domain are shown to be
cation operation if b = 0. And without loss of generality, we assume
special cases of our achieved results.
b > 0 in the following sections. The inverse of the SAFT is given by
parameters A−1 = (d, − b, − c, a, bω0 − du0 , cu0 − aω0 ) as follows:
3.1. Generalized translation and general framework of

+∞ convolution theory
−1
f (t) = S A [FA (u)] (t) = C FA (u) hA−1 (u, t) du (3)
Generalized translation is necessary in dealing with the signal
−∞ transforms with non-exponential kernels to have the translation

or shift property of comparable simplicity to that of the Laplace or
j(1/2) cdu2 −2adu0 ω0 +abω2
where C = e 0 0 . This can be verified by using
Fourier transforms [31].
the Definition (1). The SAFT can model a number of optical opera- If we consider a general signal transform and its Fourier-type
tions such as rotation and magnification (see Table 1). inverse given by
The SAFT has the following important space shift and phase shift 
properties [4,16], which are used to derive the new convolution f (t) =  (ω) F (ω) h (ω, t) dω (11)
theorems for SAFT in this paper.
 X. Zhi et al. / Optik 127 (2016) 2613–2616 2615


Similarly
F (ω) =  (t) f (t) h∗ (ω, t) dt (12)

S A [f (t) × g (t)] (u) = FA GA (u) (20)

then the -generalized translation of signal f (t) denoted by f t

Proof. According to the expression for the generalized translation
is given by [31]
 in the SAFT domains obtained in (17), we have

f t =  (ω) F (ω) h (ω, t) h∗ (ω, ) dω (13) 

+∞
j(a(t 2 − 2 )/2b) j(1/b)u0 (t−) 1

g(t) = Ne e √ GA (u)e−j((t−)u/b) du(21)
2
where  in the argument of the function f t is the generalized −∞
delay operator for the generalized translation, h (ω, t) is the ker-
 
nel of the transform, h∗ (ω, ) stands for the complex conjugater of where N = 1/ (j2b) 1/ (−jb), the generalized convolution
h (ω, ), and  (ω) is the weight function. The shift property for the defined as
FT expressed below
∞
FT

−jω f (t) g (t) =

f (t − ) ←−F (ω) e (14) f () g t d (22)

is consistent with the above definition of the generalized transla- −∞

tion as can be easily verified. Substituting (21) into (22), we can express (22) as
In the general framework of convolution theory (see [32,Ch. 4]),
it is known that to every integral transformation I, one can, at least 
+∞
2 − 2 )/2b
theoretically, associate with it a convolution operation, , such that f (t) g(t) = M f ()ej(a(t ) ej(1/b)u0 (t−)

I (f  g) = I (f ) I (g) (15) −∞

For example, the convolution operation associated with the 

+∞

Hankel transform is too complicated to be stated here, but the × GA (u)e−j((t−)u/b) dud (23)
interested reader can find the details in [32,Sect. 21.6].
−∞
 
3.2. New convolution theorem in the SAFT domains where M = 1/ (j2b) 1/ (−j2b). This expression in (23) can
be simplified to
In this subsection, we consider the extension of the generalized
convolution [31] using the SAFT kernel given in (1). Based on the SAFT
f (t) g (t) −→FA (u) GA (u) (24)
generalized translation, for the special case of the SAFT, with the
weight function  (ω) taken as unity and the kernel h taken equal Similarly, we can get
to the kernel of the SAFT given by (1), the generalized translation
SAFT
in (13) can be expressed as f (t) × g (t) −→FA (u) GA (u) (25)


+∞
The proof of this Eq. (25) is similar to the proof of Eq. (24) and is
f t =  (u) FA (u) hA (u, t) h∗A (u, ) du
omitted.
 +∞
−∞ (16) According to (24) and (25), we can obtain
2 − 2 )
=M FA (u) ej(a/2b)(t ej(1/b)(u0 −u)(t−) du
SAFT
−∞ f (t) g (t) ←−FA (u) GA (u) (26)
 
where M = 1/ (j2b) 1/ (−j2b), alternatively, We have introduced expressions for the SAFT’s of a convolution

+∞ and a product of two functions. Based on generalized transla-

) ej(1/b)u0 (t−) √1
2 − 2 )/2b tion, their convolution formula generalizes very nicely the classical
f t = Nej(a(t FA (u) e−j((t−)u/b) du
2 result for FT. This shows that the generalized convolution of two sig-
−∞ nals in time domain is equivalent to simple multiplication of their
(17) SAFTs in the SAFT domains. Eq. (26) is particularly useful in filter
  design as discussed in [16]. Moreover, some of well-known results
where N = 1/ (j2b) 1/ (−jb). This is the required expression

about the convolution theorem in FT domain, FRFT domain, and LCT
for f t , the -generalized translation of signal f (t) by amount

domain are shown to be special cases of our achieved results.
. This expression for the f t in (17) will be used to derive an
expression for the generalized convolution theorem in the SAFT Corollary 1. If the parameter of SAFT A = (0, 1, −1, 0, 0, 0), then
domains. Theorem 1 reduces to the conventional convolution in FT domain
as follows:
Definition 1. For any signal f (t), g (t), let us define a new convo-
FT
lution operation  for SAFT as follows: f (t) ⊗ g (t) ←−F (ω) G (ω) (27)
∞


where ‘⊗’ denotes the conventional convolution operation.
z (t) = f g (t) = f () g t d (18)
Corollary 2. If the parameter of SAFT A = (a, b, c, d, 0, 0), then
−∞
Theorem 1 collapses to the generalized convolution for LCT.

where g t is the -generalized translation of signal g (t). LCT

f (t) g (t) ←−FA (u) GA (u) (28)
Theorem 1. Let z (t) = f g (t) and FA , GA , ZA denote the SAFT of
f, g and z, respectively. Then The proof of Corollary 1 and Corollary 2 is easy and is omitted.
In addition, convolution theorems for the specific SAFT case FRFT
S A [z (t)] (u) = ZA (u) = FA (u) GA (u) (19) can also be obtained from the above-derived theorems [25–28].
2616
H A (u )
rin ( t ) S A ( .) × −
S A ( .)
Fig. 1. Multiplicative filter in the SAFT domain.
4. Multiplicative filtering in the SAFT domain
In this section, we mainly discuss the multiplicative filtering
in the SAFT domain and show that the generalized convolution
structure is easy to implement in designing filters.
The model of the multiplicative filtering in the SAFT domain is
shown in Fig. 1 [13]. Many models of multiplicative filters can be
achieved by designing H(A) (u), such as low pass, high pass, band
pass, band stop, and so on. For example, if the received signal com-
prises of the desired signal f (t) and noise n (t) together, then noise
can be discarded to a large extent and Signal to Noise Ratio (SNR)
can be increased through a multiplicative filter in the SAFT domain
when the SAFT components of noise NA (u) and desired signal SA (u)
have no overlapping or minimal overlapping.
In [16], authors derived a convolution structure for SAFT. They
discuss the design methods and the performance of the multiplica-
tive filter design by the SAFT. Moreover, they showed that above
mentioned multiplicative filter in the SAFT domain can also be
achieved through convolution in the time domain. However, the
convolution theorem for the LCT given in [16] lacks simplicity com-
parable to our results. Our theorem states the powerful result that
the convolution of two signals in time domain results in simple
multiplication of their SAFTs in the SAFT domain. Those results are
extensions of the convolution theorem of the FT to the SAFT domain,
and can be more useful in practical analog filtering in SAFT domains.
According to Theorem 1, with the assumption that function GA (u) in
(26), acts as the transfer function of the multiplicative filter, HA (u),
is designed as
GA (u) = HA (u)
Therefore we obtain
rout (t) = rin (t) g (t)
where g (t) is the inverse SAFT of G(A) (u).
Eqs. (26) and (29) is particularly useful in filter design. For exam-
ple, if we are interested only in the frequency spectrum of the SAFT
in the region [u1 , u2 ] of a signal f, we choose the filter impulse
response, g, so that GA is constant over [u1 , u2 ], and zero or of rapid
decay outside that region. Passing the output of the filter, yields that
part of the spectrum of f over [u1 , u2 ]. This is easier to implement
than the one suggested in [16].
5. Conclusion
In this paper, we have introduced expressions for the SAFT of
a convolution and a product of two functions. First, we propose a
new convolution theorem for the SAFT based on generalized trans-
lation. Moreover, our convolution structure generalizes very nicely
the classical result for FT. This theorem states the powerful result
that the convolution of two signals in time domain results in simple
multiplication of their SAFTs in the SAFT domain. Then, we show
that the convolution theorem in FT, LCT domain can be looks as
special cases of our achieved results. Last, utilizing the new con-
volution theorem derived in this paper, multiplicative filter in the
SAFT domain has been further investigated. It is shown that our
convolution can be more useful in practical analog filtering in SAFT
domain.
X. Zhi et al. / Optik 127 (2016) 2613–2616
Acknowledgements
This work was supported in part by the Natural Science Basic
rout ( t )
Research Plan in Shaanxi Province of China (Grant nos. 2014JQ1037
and 2015JM6339), and the Fundamental Research Fund for the Cen-
tral Universities under (Grant nos. NSIY181413 and JB150711).
References
[1] S. Abe, J.T. Sheridan, Optical operations on wave functions as the Abelian
subgroups of the special affine Fourier transformation, Opt. Lett. 19 (1994)
1801–1803.
[2] S. Abe, J. Sheridan, Generalization of the fractional Fourier transformation to an
arbitrary linear lossless transformation an operator approach, J. Phys. A: Math.
Gen. 27 (1994) 4179–4185.
[3] S.-C. Pei, J.-J. Ding, Eigenfunctions of the offset Fourier, fractional Fourier, and
linear canonical transforms, JOSA A: Opt. Image Sci. Vision 20 (2003) 522–532.
[4] S.-C. Pei, J.-J. Ding, Eigenfunctions of Fourier and fractional Fourier transforms
with complex offsets and parameters, IEEE Trans. Circuits Syst. I 54 (7) (2007)
1599–1611.
[5] S.C. Pei, J.J. Ding, Relations between fractional operations and time–frequency
distributions, and their applications, IEEE Trans. Signal Process. 49 (2001)
1638–1655.
[6] L.B. Almeida, The fractional Fourier transform and time frequency representa-
tions, IEEE Trans. Signal Process. 42 (11) (1994) 3084–3091.
[7] D.F.V. James, G.S. Agarwal, The generalized Fresnel transform and its applica-
tions to optics, Opt. Commun. 126 (5) (1996) 207–212.
[8] M. Moshinsky, C. Quesne, Linear canonical transformations and their unitary
representations, J. Math. Phys. 12 (1971) 1772–1780.
[9] H.M. Ozaktas, Z. Zalevsky, M.A. Kutay, The Fractional Fourier Transform with
Applications in Optics and Signal Processing, Wiley, New York, NY, 2001.
[10] T. Alieva, M.J. Bastiaans, Properties of the linear canonical integral transforma-
tion, J. Opt. Soc. Am. A 24 (11) (2007) 3658–3665.
[11] M.B. Barshan, H.M. Ozaktas, Optimal filtering with linear canonical transfor-
mation, Opt. Commun. 135 (1997) 32–36.
[12] D. Wei, Y. Li, Generalized wavelet transform based on the convolution operator
in the linear canonical transform domain, Optik 125 (2014) 4491–4496.
[13] D. Wei, Q. Ran, Y. Li, J. Ma, L. Tan, A convolution and product theorem for the
linear canonical transform, IEEE Signal Process. Lett. 16 (2009) 853–856.
[14] D. Wei, Q. Ran, Y. Li, New convolution theorem for the linear canonical trans-
form and its translation invariance property, Optik 123 (2012) 1478–1481.
[15] K.K. Sharma, S.D. Joshi, Signal separation using linear canonical and fractional
Fourier transforms, Opt. Commun. 265 (2) (2006) 454–460.
[16] Q. Xiang, K. Qin, Convolution, correlation, and sampling theorems for the offset
linear canonical transform, Signal Image Video Process. 8 (2014) 433–442.
[17] A. Stern, Sampling of compact signals in the offset linear canonical domain,
(29) Signal Image Video Process 1 (4) (2007) 359–367.
[18] Q. Xiang, K.Y. Qin, Q.Z. Huang, Multichannel sampling of signals band-limited
in offset linear canonical transform domains, Circuits Syst. Signal Process. 32
(5) (2013) 2385–2406.
(30) [19] A. Stern, Sampling of linear canonical transformed signals, Signal Process. 86
(2006) 1421–1425.
[20] B.-Z. Li, R. Tao, Y. Wang, New sampling formulae related to linear canonical
transform, Signal Process. 87 (2007) 983–990.
[21] J.J. Healy, J.T. Sheridan, Sampling and discretization of the linear canonical
transform, Signal Process. 89 (2009) 641–648.
[22] D. Wei, Y. Li, Reconstruction of multidimensional bandlimited signals from
multichannel samples in the linear canonical transform domain, IET Signal
Process. 8 (6) (2014) 647–657.
[23] D. Wei, Y. Li, Multichannel sampling theorem for bandpass signals in the linear
canonical transform domain, Optik 125 (2014) 3434–3438.
[24] O. Akay, G.F. Boudreaux-Bartels, Fractional convolution and correlation via
operator methods and application to detection of linear FM signals, IEEE Trans.
Signal Process. 49 (2001) 979–993.
[25] L.B. Almeida, Product and convolution theorems for the fractional Fourier trans-
form, IEEE Signal Process. Lett. 4 (1997) 15–17.
[26] A.I. Zayed, A convolution and product theorem for the fractional Fourier trans-
form, IEEE Signal Proc. Lett. 5 (1998) 101–103.
[27] R. Torres, P. Pellat-Finet, Y. Torres, Fractional convolution, fractional corre-
lation and their translation invariance properties, Signal Process. 90 (2010)
1976–1984.
[28] J. Shi, X. Liu, N. Zhang, Generalized convolution and product theorems asso-
ciated whit linear canonical transform, Signal Image Video Process. 8 (2014)
967–974.
[29] D. Mendlovic, Y. Bitran, R.G. Dorsch, A.W. Lohmann, Optical fractional correla-
tion: experimental results, J. Opt. Soc. Am. A 12 (1995) 1665–1670.
[30] H.M. Ozaktas, B. Barshan, D. Mendlovic, L. Onural, Convolution, filtering, and
multiplexing in fractional Fourier domains and their relationship to chirp and
wavelet transforms, J. Opt. Soc. Am. A 11 (1994) 547–559.
[31] J. Robert, I.I. Marks, Advanced Topics in Shannon Sampling and Interpolation
Theory, Springer-Verlag, New York, 1993.
[32] A.I. Zayed, Function and Generalized Function Transformations, CRC Press, Boca
Raton, FL, 1996.