GP 503: Geophysical Signal Processing

Lecture – 8

Instructor: E. Chandrasekhar
Department of Earth Sciences
Indian Institute of Technology Bombay
E-mail: [email protected]
Windowing:
1. Box-car window or Rectangular window:
1
Mathematically, it is defined as w(t ) = 1 / T 0≤|t |≤T
=0 |t |>T
-T 0 T
t
Its frequency characteristics are defined as
T T sinc(ω
2Sinc (ωT)
/2)
W (ω ) = ∫ w(t ) e −iωt dt = ∫ (1 / T ) e −iωt dt 1
−T −T

 sin ωT 
T
= (1 / T ) ∫ e − iωt
dt = 2   = 2 sinc(ωT)
−T  ωT  0
ω

As a result, since W(ω) has side lobes, it is less well preferred for smoothing operations.
How will the side lobes affect the spectrum?
Windowing:
2. Bartlett Window or Triangular window: w(t)
1
1 t
w(t ) = (1 − ) 0 ≤ t ≤ T
T T
=0 t >T t
Its FT is given by
0
1 t  − iω t T
1 t  − iω t
W (ω ) = ∫ 1 −  e dt + ∫ 1 −  e dt
−T
T T 0
T T
This can be simplified to get
2  t  e iω t + e − iω t
T
2  t
T

W (ω ) = ∫ 1 −  ( )dt = ∫ 1 −  cos ω t dt
T 0 T  2 T 0 T 
Windowing:
2. Bartlett Window or Triangular window:

The above equation can be solved to get

W (ω ) = Sinc (ωT / 2)
2

Sinc 2 (ωT / 2)

w(t)

⇔
1

Bartlett window can minimize the spectral leakage

Windowing:
3. Hanning window
1 𝜋𝜋𝜋𝜋
𝑊𝑊 𝑡𝑡 = 1 + cos 0 ≤ 𝑡𝑡 < 𝑇𝑇
2𝑇𝑇 𝑇𝑇

= 0 |𝑡𝑡| > 𝑇𝑇

4. Hamming window

1 𝜋𝜋𝜋𝜋
𝑊𝑊 𝑡𝑡 = 0.54 + 0.46 cos 0 ≤ 𝑡𝑡 < 𝑇𝑇
𝑇𝑇 𝑇𝑇

= 0 |𝑡𝑡| > 𝑇𝑇
Windowing:
5. Cosine tapering (Edge effect removal):

In signal processing, tapering off of the given data is always important, preferably
with a cosine function. This helps in removing the edge effects in the data.
Z-transform: Relation with Fourier transform
We know that the Z-transform of a discrete signal, 𝑥𝑥𝑡𝑡 , is defined as
∞
X ( z) = ∑ t
x
t = −∞
z t
(1)

Considering z = e iω and replacing the summation with integration in (1), we get

∞
X ( e iω ) = ∫
i ωt
x (t ) e dt
−∞
∞
⇒ X (ω ) = ∫ x (t ) e i ωt
dt (2)
−∞

Invoking the Dirac delta function, 𝑥𝑥 𝑡𝑡 in discrete form can be written as

𝑥𝑥 𝑡𝑡 = �𝑥𝑥𝑙𝑙 𝛿𝛿(𝑡𝑡 − 𝑙𝑙 ) (3)

𝑙𝑙
Z-transform: Relation with Fourier transform
The inverse Fourier transform of (2) is
∞
1
𝑥𝑥(𝑡𝑡) = � 𝑋𝑋(𝜔𝜔) 𝑒𝑒 −𝑖𝑖𝑖𝑖𝑖𝑖 𝑑𝑑𝑑𝑑 (4)
2𝜋𝜋
−∞

We know that the integration of 𝑒𝑒 𝑖𝑖𝑖𝑖𝑖𝑖 (or 𝑧𝑧 𝑛𝑛 ) over the integral −𝜋𝜋 ≤ 𝜔𝜔 < 𝜋𝜋 is nonzero for
𝑛𝑛 = 0 and is equal to 0 for all 𝑛𝑛 ≠ 0. That is,

𝜋𝜋 𝜋𝜋
1 𝑖𝑖𝑖𝑖𝑖𝑖
1 1 for 𝑛𝑛 = 0
� 𝑒𝑒 𝑑𝑑𝑑𝑑 = � (cos 𝑛𝑛𝑛𝑛 + 𝑖𝑖 sin 𝑛𝑛𝑛𝑛)𝑑𝑑𝑑𝑑 = � (5)
2𝜋𝜋 2𝜋𝜋 0 for 𝑛𝑛 ≠ 0
−𝜋𝜋 −𝜋𝜋
Z-transform: Relation with Fourier transform
Therefore, accordingly changing the limits of integration and in terms of discretized
function, equation (4) can be written as
𝜋𝜋
1
𝑥𝑥𝑡𝑡 = ��… + 𝑥𝑥−2 𝑒𝑒 −2𝑖𝑖𝑖𝑖 + 𝑥𝑥−1 𝑒𝑒 −𝑖𝑖𝑖𝑖 + 𝑥𝑥0 + 𝑥𝑥1 𝑒𝑒 𝑖𝑖𝑖𝑖 + 𝑥𝑥2 𝑒𝑒 2𝑖𝑖𝑖𝑖 + ⋯ )𝑒𝑒 −𝑖𝑖𝑖𝑖𝑖𝑖 𝑑𝑑𝑑𝑑 (6)
2𝜋𝜋
−𝜋𝜋
Therefore, by virtue of equation (5), all terms in the equation (6) would vanish except
the term with the coefficient of 𝑧𝑧 with power zero, which would contribute to the
integral. Accordingly, from equation (6) we have,
𝜋𝜋
1
𝑥𝑥𝑡𝑡 = 𝑥𝑥𝑡𝑡 � 𝑑𝑑𝑑𝑑 = 𝑥𝑥𝑡𝑡 (7)
2𝜋𝜋
−𝜋𝜋
This proves that how the inverse Fourier transform helps to identify the coefficients of
powers of 𝑧𝑧.
Any Questions?