Professional Documents
Culture Documents
Fractional Fourier Transform: A Survey: August 2012
Fractional Fourier Transform: A Survey: August 2012
Fractional Fourier Transform: A Survey: August 2012
net/publication/232846454
CITATIONS READS
4 2,628
1 author:
Krishna Battula
Jawaharlal Nehru Technological University Kakinada
65 PUBLICATIONS 648 CITATIONS
SEE PROFILE
All content following this page was uploaded by Krishna Battula on 11 March 2018.
B.T.Krishna
Department of Electronics and Communication Engineering
Gitam Institute of Technology
GITAM University, Visakhapatnam, India
tkbattula@gmail.com
j (u2 + t2 ) cotα−jutcosecα
q
1−jcotα
e α 6= kπ
2
2π
Kα (u, t) =
∂ (u − t) α = 2kπ
∂ (u + t) α = (2k + 1) π
(2)
where α = aπ/2 is rotation angle in the phase plane
and a is a constant ranging from 0 to 4. The Trans-
form Kernel is also defined as[21, 23, 28], 1/α, sinα = α
∞
r
X 1 −j ( u−t
2α )
2
Kα (u, t) = e −jnα
ψn (t)ψn (u) (3) Kα (u, t) = e (6)
j2πα
n=0
2. A Fourier transform with its argument scaled 2. When α = 2π FRFT of the signal is same as the
by cosecα. identity operator.
3. Multiplication by another chirp signal 3. When α = π/2 FRFT gives the Fourier trans-
form of the signal. When α = π/2 ,cotα =
4. Multiplication by another constant 0, cosecα = 1 and
r
1 −jut
Kα (u, t) = e (8)
The inverse fractional Fourier transform is defined as[2, 2π
5, 12, 22, 23, 24, 27], r
Z ∞
Z ∞ 1 −jut
f (t) = Fα (u) K−α (u, t) du (4) F (u) =
π f (t) e dt = F (u) (9)
2
−∞ 2π
−∞
The relationship between fractional domain with the 4. The FRFT of order −α is the inverse of the
traditional time-frequency plane can be expressed in FRFT of order α .
matrix form as,
5. Successive applications of FRFT are equivalent
t cosα sinα u to a single transform whose order is equal to the
= (5)
ω −sinα cosα v sum of the individual orders.
FRFT will have an advantage of providing rotation Fα Fβ = Fα+β (10)
in the time−frequency plane. FRFT of Some of the
commonly used signals is shown in Fig.1. From Fig it 6. According to Parseval’s theorem
is observed that FRFT of a Gaussian is also Gaussian. Z ∞ Z ∞
Fig.1 illustrates how the FRFT of a rectangular signal |f (t)|2 dt = |Fα (u)|2 du (11)
evolves into a sinc function as α is varied from 0 to −∞ −∞
jat2
Let g(t) = c f (t)e , The above expression reduces
to,
Figure 1: FRFT of some common signals
Z∞
2 1
below. e−jau Fα (u) = √ g(t)e−j2abut dt = G(2abu)
2π
−∞
(14)
1. When the value of α = 0, the FRFT of the sig- 2
nal is the signal itself. When α = 0, cotα = Fα (u) = ejau G(2abu) (15)
where a 6= 0. The kernels of FRFT for different values
Table 1: FRFT of some common signals of α can be regarded as wavelet family.
Signal FRFT
√ ju2 tanα
1 1+jtanα e− 2 3. DISCRETE FRFT
−t2 −u2 The evaluation of continuous time fractional fourier
e 2 e 2
transform is done by it’s discrete version named Dis-
√ j(v2 +u2 )tanα crete Fractional Fourier Transform(DFRFT)[5, 24, 28,
ejvt 1+jtanα ejuvsecα− 2
9, 1]. Although, the properties of CFRFT have been
u2 +τ 2
q
δ (t−τ ) 1−jcotα j
e 2
cotα−juτ cosecα well studied, a universally agreed upon discrete coun-
2π
terpart is yet to be found. The DFRFT should satisfy
f (−t) Fα (−u) the following properties
v2 cos α sin α
f (t)ejvt ejuv cos α−j 2 Fα (u − v sin α)
2
−juτ cos α+j τ cos2α sin α
1. Unitary property for the Transform to be re-
f (t − τ ) e Fα (u − τ cos α) versible .
tf (t) ucosαFα (u) + jsinαFα′ (u)
2. Additivity of orders needed for rotation.
df
dt
cosαFα′ (u) + jusinαFα (u)
3. Generalization of discrete Fourier transform (DFT)
Rt ju2 tan α Ru jz 2 tan α
f (t′ )dt′ sec αe− 2 Fα (z)e 2 dz is needed to complete theory.
a a
ju2 4. Similarity with CFRFT is needed from the point
f (t)∗g(t) e− 2 (
cot απ
2 ) F α (u)Gα (u)
of view of Theory.
where g(t) and G(u) are Fourier Transform pair. Additionally, it is required to have a closed form and
be efficient to implement. Pei and Deng has classified
The Wigner Distribution(WD) of a signal f (t) is de- some of the possible definitions of DFRFT in 2000[17,
fined as, 20]. The straight forward way to obtain discrete ver-
Z ∞ sion is to sample the continuous time fractional fourier
τ ∗ τ −juτ transform (Direct form type DFRFT ) and is given by,
Wf (t, u) = f t+ f t− e dτ (16)
−∞ 2 2
N
X
The Wigner distribution depicts the distribution of Fα (m) = Kα (m, n)f (n) (20)
signals energy over the entire time-frequency plane. n=−N
For a given time domain signal f (t) the Wigner dis-
Kernel Kα (m, n) is defined as,
tribution of its Fourier Transform can be obtained
by rotating the Wigner Distribution of f (t) by π2 in jcotα
(m2 ∆u2 +n2 ∆t2 ) e −j2πmnS
Kα (m, n) = Cα e 2 2M +1 (21)
Time−Frequency plane in clockwise direction. Sim-
ilarly the Wigner distribution of αth order FRFT of where,
f (t) is obtained by rotating the Wigner distribution r
of f (t) by απ in Time−Frequency plane[3, 6, 19, 12]. | sin α| − jsgn(sin α) cos α
2 Cα = (22)
In mathematical form it is represented as, 2M + 1
WFα (t, u) = W f (t cos (φ) − u sin (φ) , t sin (φ) + u cos (φ)) f (n) denotes samples of input with n = (−N, N ),
(17) Fα (m) is the samples of output with m = (−M, M ), S
where Fα is the αth order FRFT of function f (t). is integer prime to 2M +1, and ∆u∆t = 2πS sin α
2M +1
. The
disadvantage of direct form DFRFT is that it does
The Ambiguity Function(AF) for f (t) is defined as, not obey unitary, additivity, reversibility and closed
form definitions. However an improved version of di-
Z ∞
∗ −j2πf t rect type DFRFT is possible. The second kind of
A (τ, f ) = f (t)f (t−τ ) e dt (18)
−∞ DFRFT is Linear combination type DFRFT, which
is derived by using the linear combination of DFT,
The ambiguity function and wigner distribution are linearity,inverse operation and IDFT. It is given by,
related by two-dimensional fourier transform. Wigner
N
distribution gives and idea about the distribution of X
the energy of a signal, whereas, the ambiguity function Fα [f (n)] = Kα (n, k)f (n) (23)
n=0
gives a correlative interpretation. So, It can be easily
said that FRFT and ambiguity function are related by where
a rotation angle[13, 29]. 1 (α)
Kα (n, k) = a0 (α)δ(n − k) + a√
−j2πnk
N
e N
j2πnk (24)
3 (α)
The continuous wavelet transform (CWT) of the sig- +a2 (α)δ [(n + k)N ] + a√ N
e N
nal f (t) is defined as,
Z ∞ Eigen vector decomposition type DFRFT is obtained
1 1 u−t by by finding the eigen vectors and eigen values of
Lψ f (a, t) = √ ψ f (u)du (19)
cψ a
p
|a| −∞ DFT matrix and taking the fractional power. This was
invented by Pei and Deng, and the results obtained completely. Finally, FRFT of order −α is applied to
will match to FRFT. get back the signal. In order to illustrate the noise re-
2
N
X moval operation, a gaussian signal,f (t) = e−π(t−4) , is
Fα [f (n)] = uk [n] uTk [n] e−jαk (25) 2
added by the noise, n(t) = e−jπt rect( 16 t
) , and frac-
k=0 tional order α = 0.5 is chosen. The original Signal,
Where uk [n] is kth discrete Hermite-Gaussian function Noise, FRFT of Noise and Signal, Recovered signal
and k 6= N , for N odd k 6= N −1, for N even. In matrix are shown in Fig.2.
form,
Fα [f (n)] = U EU T (26)
Where U is discrete Hermite-Gaussian matrix consists
of discrete Hermite-Gaussian functions, E is a diago-
nal matrix that consist of the eigen values e−jkα , k =
0, 1, 2, ...N of DFrFT matrix Fα as diagonal elements.
Group theory type DFRFT is obtained by multiplying
DFT with periodic chirps. This is used only when the
fractional order equals some specified angles. Compar-
ison of Linear weighted type, Sampling type and eigen
decomposition type DFRFT is presented in Table.3.
From the Table.3 it is evident that Linear weighted