Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
1. Introduction
Let ϕ ∈ L1 (R) ∩ L2 (R). Then for a signal f in L2 (R), the Gabor transform Gϕ f
of f with respect to the window ϕ is defined by
∞
−1/2
(Gϕ f )(b, ξ) = (2π) e−ixξ f (x)ϕ(x − b)dx, b, ξ ∈ R.
−∞
Equivalently,
where ( , )L2 (R) is the inner product in L2 (R), Mξ and T−b are the modulation
operator and the translation operator given by
and
(T−b h)(x) = h(x − b)
for all measurable functions h on R and all x in R. In signal analysis,
(Gϕ f )(b, ξ) gives the time-frequency content of the signal f at time b and fre-
quency ξ by placing the window ϕ at time b. See, for instance, [2].
The time-frequency analysis of signals based on the Gabor transform Gϕ is
undermined by the fact that the window ϕ is not automatically wide for low
frequency and not automatically narrow for high frequency. Rectifying this
deficiency is one motivation for the wavelet transforms that we now recall.
Let ϕ ∈ L2 (R) be such that
∞ |ϕ̂(ξ)|2
dξ < ∞, (1)
−∞ |ξ|
and the convergence is understood to take place in L2 (R). The Fourier trans-
form ϕ̂ is sometimes denoted by F ϕ. The condition (1) is known as the ad-
missibility condition for a function ϕ in L2 (R) to be eligible as a window or a
“mother wavelet”. See Chapter 2 of [3] and Chapter 18 of [10] for details.
Let ϕ ∈ L2 (R) be a mother wavelet. Then for all b in R and a in R \ {0},
we define the wavelet ϕb,a by
1 x−b
ϕb,a (x) = ϕ , x ∈ R.
|a| a
1 x
(D21/a h)(x) = h
|a| a
for all measurable functions h on R and all x in R. Now, we define the wavelet
transform Ωϕ f of a function f in L2 (R) to be the function on R × (R \ {0}) by
Then, as a hybrid of the Gabor transform and the wavelet transform, the Stock-
well transform Sϕ f of a signal f in L2 (R) with respect to the window ϕ is
defined by
∞
−1/2
(Sϕ f )(b, ξ) = (2π) |ξ| e−ixξ f (x)ϕ(ξ(x − b))dx (2)
−∞
where
ϕb,ξ = (2π)−1/2 Mξ T−b D1ξ ϕ
4 QIANG GUO AND M. W. WONG
6 Qiang Guo e Man Wah Wong
where
ψ(x) = eix ϕ(x), x ∈ R.
Let f be a Schwartz function on R such that f (b) > 0 for all b in R. Then as
ξ → ∞,
|(Sϕ f )(b, ξ)| → a(1 − e−2 )| f (b)|, b ∈ R.
/ L2 (R × R).
This shows that Sϕ f ∈
ϕL2 (R) = 1
6 QIANG GUO AND M. W. WONG
8 Qiang Guo e Man Wah Wong
and
∞ |ϕ̂(ξ − 1)|2
dξ < ∞. (5)
−∞ |ξ|
Then for all f and g in L2 (R),
1
∞ ∞ db dξ
( f , g)L2 (R) = (Sϕ f )(b, ξ)(Sϕ g)(b, ξ) ,
cϕ −∞ −∞ |ξ|
where
∞ |ϕ̂(ξ − 1)|2
cϕ = dξ.
−∞ |ξ|
If we pick ϕ to be the function on R defined by
2
ϕ(x) = (2π)−1/2 e−x /2 , x ∈ R, (6)
then we get the Stockwell transform first introduced in [8]. Unfortunately, this
window ϕ does not satisfy the admissibility condition (5). As in Example 1, Sϕ
does not map L2 (R) into L2 (R × R). The range R(Sϕ ) given by R(Sϕ ) = {Sϕ f :
f ∈ L2 (R)} is computed in [11].
The aim of this paper is to introduce the modified Stockwell transforms Sϕs
parametrized by s, where 1 ≤ s ≤ ∞ and
∞
ϕ(x) dx = 1.
−∞
These modified Stockwell transforms, which are given in Section 2, include the
classical Stockwell transforms when s = 1 and are reminiscent of the wavelet
transforms when s = 2. The significance of the modified Stockwell transforms
is described in Section 2 by displaying the modified Stockwell spectra of a sig-
nal for different values of s. We also give in Section 2 the resolution of the
identity formula, the continuous and the discrete inversion formula for modi-
fied Stockwell transforms. In Section 3, we give a reconstruction formula for a
signal f based on the absolutely referenced phase information of the modified
Stockwell transform and show that a slight variation of the same theme gives a
formula for the Riesz potential of a signal f based on the corresponding modi-
fied Stockwell transform. The range R(Sϕs ) of the Stockwell transform Sϕs given
by
R(Sϕs ) = {Sϕs f : f ∈ L2 (R)}
is computed in Section 4 and it turns out to be a reproducing kernel Hilbert
space. For every signal f in L2 (R), the graph G(Sϕs f ) given by
G(Sϕs f ) = {(Sϕs f )(b, ξ) : b ∈ R, ξ ∈ R \ {0}}
MODIFIED STOCKWELL TRANSFORMS 7
Modified Stockwell Transforms 9
is the time-frequency spectrum of the signal f . So, the main result on the de-
scription of the range R(Sϕs ) of the modified Stockwell transform Sϕs in Section
4 actually sheds light on the spectrum of the modified Stockwell transform Sϕs
in time-frequency analysis. In Section 5, localization operators for the mod-
ified Stockwell transforms are introduced and shown to be Weyl transforms
after a brief recall of Wigner transforms and Weyl transforms.
100
0.5
frequency
amplitude
200
0
300
−0.5
400
500
0 0.5 1 1.5 2 0 0.5 1 1.5 2
time time
Modified Stockwell Spectrum with s=2 Modified Stockwell Spectrum with s=8
0 0
100 100
frequency
frequency
200 200
300 300
400 400
500 500
0 0.5 1 1.5 2 0 0.5 1 1.5 2
time time
ϕL2 (R) = 1.
MODIFIED STOCKWELL TRANSFORMS 9
Modified Stockwell Transforms 11
and
|ϕ̂(ξ − 1)|2
∞
dξ < ∞.
−∞ |ξ|
Then for all f and g in L2 (R), we get
1
∞ ∞ db dξ
( f , g)L2 (R) = (Sϕs f )(b, ξ)(Sϕs g)(b, ξ) ,
cϕ −∞ −∞ |ξ|1−(2/s′ )
where
∞ |ϕ̂(ξ − 1)|2
cϕ = dξ.
−∞ |ξ|
and
b = nb0 ξm
0.
For all m and n in Z, we let Φm,n and Φsm,n be defined by
Φm,n = ϕnb0 ξm0 ,ξm0
and −m
nb0 ξm
0 ,ξ0
Φm,n
s = ϕs .
Then we have the following discrete inversion formula for a signal from its
Stockwell transform.
The following result is in fact Lemma 2 in Chapter 5 of the book [6] cast in
the definition of the Fourier transform and the notation used in this paper.
where
Γ 2s1 ′
′
γ(1/s′ ) = π1/2 21/s
.
Γ 12 − 2s1 ′
We can give a formula, which is an analog of the formula (4) for modified
Stockwell transforms.
4. Spectra
Using the resolution of the identity formula given in Theorem 6, we can show
that the range of a modified Stockwell transform is a reproducing kernel Hilbert
space.
Proof. It is clear from Theorem 6 that Sϕs : L2 (R) → L2 R × R, |ξ|db dξ
′
1−(2/s )
Fj = Sϕs f j .
Then by (7), we see that { f j }∞j=1 is a Cauchy sequence in L2 (R). So, there
exists a function f in L2 (R) such that
fj → f
in L2 (R) as j → ∞. Therefore
Sϕs f j → Sϕs f
in L2 R × R, |ξ|db dξ
1−(2/s′ ) as j → ∞. Hence
F = Sϕs f
Theorem 11. Let ϕ be as in Theorem 10. Then for 1 ≤ s ≤ ∞, the range R(Sϕs )
of the modified Stockwell transform Sϕs is a reproducing kernel Hilbert space.
H ∋ f → f (x) ∈ C
K(x, y) = gx (y), x, y ∈ X,
b,ξ
K(b, ξ; b′ , ξ′ ) = (Sϕs ϕs )(b′ , ξ′ )
5. Localization Operators
1
∞ ∞ db dξ
s
(Lσ,ϕ u, v)L2 (R) = σ(b, ξ)(Sϕs u)(b, ξ)(Sϕs v)(b, ξ)
cϕ −∞ −∞ |ξ|1−(2/s′ )
14 QIANG GUO AND M. W. WONG
16 Qiang Guo e Man Wah Wong
1
∞ ∞ db dξ
s
(Lσ,ϕ u, v)L2 (R) = σ(b, ξ)(u, ϕb,ξ b,ξ
s )L2 (R) (ϕs , v)L2 (R)
cϕ −∞ −∞ |ξ|1−(2/s′ )
Then we have the Moyal identity to the effect that for functions u1 , v1 , u2 and
v2 in L2 (R),
(W (u1 , v1 ),W (u2 , v2 ))L2 (R2 ) = (u1 , u2 )L2 (R) (v1 , v2 )L2 (R) .
where
1
∞ ∞
−1/2
τ(q, p) = (2π) σ(b, ξ)W (ϕ, ϕ)(ξ(q − p), (p − ξ)/ξ)db dξ
cϕ −∞ −∞
Proof of Theorem 12 Let u and v be functions in L2 (R). Then using the Moyal
identity, we get
s
(Lσ,ϕ u, v)L2 (R)
1 ∞ ∞ db dξ
= σ(b, ξ)(u, ϕb,ξ b,ξ
s )L2 (R) (ϕs , v)L2 (R)
cϕ −∞ −∞ |ξ|1−(2/s′ )
∞ ∞
1 db dξ
= σ(b, ξ)(W (u, v),W (ϕb,ξ b,ξ
s , ϕs ))L2 (R2 ) . (9)
cϕ −∞ −∞ |ξ|1−(2/s′ )
But, by Lemma 2,
(W (u, v),W (ϕb,ξ b,ξ
s , ϕs ))L2 (R2 )
|ξ|(2/s)−1 ∞ ∞
= W (u, v)(q, p)W (ϕ, ϕ)(ξ(q − b), (p − ξ)/ξ) dq d p.
2π −∞ −∞
So, plugging the preceding formula into (9), using Fubini’s theorem and The-
orem 4.3 in [9], we get
∞ ∞
s −1/2
(Lσ,ϕ u, v)L2 (R) = (2π) τ(q, p)W (u, v)(q, p) dq d p,
−∞ −∞
where τ is as asserted.
6. Conclusions
References
[1] A RONSZAJN N., Theory of reproducing kernels, Trans. Amer. Math. Soc.
68 (1950), 337–404.
[2] C OHEN L., Time-Frequency Analysis, Prentice-Hall, 1995.
[3] DAUBECHIES I., Ten Lectures on Wavelets, SIAM, 1992.
[4] D U J., W ONG M.W. AND Z HU H., Continuous and discrete inversion
formulas for the Stockwell transform, Integral Transforms Spec. Funct.
18 (2007), 537–543.
[5] G ROSSMANN A. AND M ORLET J., Decomposition of Hardy functions
into square integrable wavelets of constant shape, SIAM J. Math. Anal.
15 (1984), 723–736.
[6] S TEIN E.M., Singular Integrals and Differentiability Properties of Func-
tions, Princeton University Press, 1970.
[7] S TOCKWELL R.G., Why use the S transform?, in Pseudo-Differential
Operators: Partial Differential Equations and Time-Frequency Analysis,
Editors: L. Rodino, B.-W. Schulze and M.W. Wong, Fields Institute Com-
munications Series 52, American Mathematical Society, 2007, 279–309.
[8] S TOCKWELL R.G., M ANSINHA L. AND L OWE R.P., Localization of
the complex spectrum: the S transform, IEEE Trans. Signal Processing
44 (1996), 998–1001.
[9] W ONG M.W., Weyl Transforms, Springer-Verlag, 1998.
[10] W ONG M.W., Wavelet Transforms and Localization Operators, Operator
Theory: Advances and Applications 136, Birkhäuser, 2002.
[11] W ONG M.W. AND Z HU H., A characterization of the Stockwell
spectrum, in Modern Trends in Pseudo-Differential Operators, Editors:
J. Toft, M.W. Wong and H. Zhu, Operator Theory: Advances and
Applications 172, Birkhäuser, 2007, 251–257
9HQJRQRLQILQHFRQVLGHUDWLRSHUDWRULGLORFDOL]]D]LRQHGHILQLWLLQWHUPLQLGL
WUDVIRUPDWDGL6WRFNZHOOPRGLILFDWDHGLHVVLYLHQHULFRVWUXLWRLOVLPERORGL
:H\O
/D &RPPLVVLRQH ULWLHQH FKH L ULVXOWDWL SUHVHQWDWL VLDQR RULJLQDOL H GL
LQWHUHVVHVLDGDOSXQWRGLYLVWDGHOODWHRULDFKHGHOOHDSSOLFD]LRQL3URSRQH
SHUWDQWRDOO¶$FFDGHPLDGHOOH6FLHQ]HODSXEEOLFD]LRQHGHOOD0HPRULD
7RULQRJHQQDLR
/D&RPPLVVLRQH
*LRYDQQL0RQHJDWR
/XLJL5RGLQR
)XOYLD6NRI