Sampling and Reconstruction

V. Rajbabu
[email protected]
EE 603: Digital Signal Processing and Applications

Department of Electrical Engineering

Indian Institute of Technology Bombay

18 Aug. 2021

1 / 13
 Sampling of CT Signals

Figure: Ideal C-to-D [DTSP3].

Obtaining DT sequence from CT signal through periodic

sampling
x[n] = xc (nT ), −∞ < n < ∞
where T - sampling period
fs = T1 - sampling frequency in samples per second,
Ωs = 2πfs = 2πT - sampling frequency in radians per second

2 / 13
 Sampling: Mathematical Representation

∞
X
s(t) = δ(t − nT )
n=−∞

Figure: Using impulse train [DTSP3].

3 / 13
 Sampling: Mathematical Representation

∞
X
s(t) = δ(t − nT )
n=−∞

xs (t) = xc (t)s(t)
X∞
= xc (t)δ(t − nT )
n=−∞
X∞
= xc (nT )δ(t − nT )
n=−∞

Figure: Using impulse train [DTSP3].

3 / 13
 Sampling: Mathematical Representation

2π X
S(jΩ) = δ(Ω − k Ωs )
T
k

Figure: Frequency domain

representation [DTSP3].
4 / 13
 Sampling: Mathematical Representation

2π X
S(jΩ) = δ(Ω − k Ωs )
T
k

1
Xs (jΩ) = Xc (jΩ) ∗ S(jΩ)
2π
1X
= Xc (j(Ω − k Ωs ))
T
k

Figure: Frequency domain

representation [DTSP3].
4 / 13
 Relating Xs (jΩ) and X (ejω )
Continuous time
∞
P
Sampled data: xs (t) = xc (nT )δ(t − nT )
n=−∞
∞
xc (nT )e−jΩnT
P
Fourier Transform: Xs (jΩ) =
n=−∞

5 / 13
 Relating Xs (jΩ) and X (ejω )
Continuous time
∞
P
Sampled data: xs (t) = xc (nT )δ(t − nT )
n=−∞
∞
xc (nT )e−jΩnT
P
Fourier Transform: Xs (jΩ) =
n=−∞
Discrete time
Sampled data: x[n] = xc (nT )
Fourier Transform: X (ejω ) = x[n]e−jωn
P
n

5 / 13
 Relating Xs (jΩ) and X (ejω )
Continuous time
∞
P
Sampled data: xs (t) = xc (nT )δ(t − nT )
n=−∞
∞
xc (nT )e−jΩnT
P
Fourier Transform: Xs (jΩ) =
n=−∞
Discrete time
Sampled data: x[n] = xc (nT )
Fourier Transform: X (ejω ) = x[n]e−jωn
P
n
Relating them

Xs (jΩ) = X (ejω ) = X (ejΩT )

ω=ΩT

5 / 13
 Relating Xs (jΩ) and X (ejω )
Continuous time
∞
P
Sampled data: xs (t) = xc (nT )δ(t − nT )
n=−∞
∞
xc (nT )e−jΩnT
P
Fourier Transform: Xs (jΩ) =
n=−∞
Discrete time
Sampled data: x[n] = xc (nT )
Fourier Transform: X (ejω ) = x[n]e−jωn
P
n
Relating them

Xs (jΩ) = X (ejω ) = X (ejΩT )

ω=ΩT
For the sampled signal,
1X 1X ω 2π
Xs (jΩ) = Xc (j(Ω−k Ωs )) =⇒ X (ejω ) = Xc (j( −k ))
T T T T
k k

is a frequency scaled version of Xs (jΩ) with ω = ΩT

5 / 13
Relating Xs (jΩ) and X (ejω )

6 / 13
Exact Recovery
If Xc (jΩ) is bandlimited
(BL) to ΩN , i.e.,

Xc (jΩ) = 0 for |Ω| ≥ ΩN

7 / 13
Exact Recovery
If Xc (jΩ) is bandlimited
(BL) to ΩN , i.e.,

Xc (jΩ) = 0 for |Ω| ≥ ΩN

and if

Ωs −ΩN ≥ ΩN or Ωs ≥ 2ΩN

7 / 13
Exact Recovery
If Xc (jΩ) is bandlimited
(BL) to ΩN , i.e.,

Xc (jΩ) = 0 for |Ω| ≥ ΩN

and if

Ωs −ΩN ≥ ΩN or Ωs ≥ 2ΩN

then xc (t) can be

recovered from xs (t)
using an ideal LPF as
reconstruction filter, i.e.,

Xr (jΩ) = Xc (jΩ)

7 / 13
 Nyquist-Shannon Sampling Theorem
Let xc (t) be a bandlimited signal with

Xc (jΩ) = 0 for |Ω| ≥ ΩN

then xc (t) is uniquely determined by its samples

x[n] = xc (nT ), n = 0, ±1, ±2, · · · ,

if
2π
Ωs = ≥ 2ΩN .
T
ΩN - Nyquist frequency, 2ΩN is Nyquist rate

8 / 13
Ideal Reconstruction

9 / 13
 Ideal Reconstruction
Reconstruction filter with cut-off Ωc = Ω2s = Tπ
(
1 0 ≤ Ω ≤ Tπ
Hr (jΩ) =
0 otherwise.

Corresponding impulse response

sin πt
T
hr (t) = πt
T

10 / 13
Ideal Reconstruction

11 / 13
 Ideal Reconstruction

Figure: Reconstruction from samples [DTSP3].

∞
P
xs (t) = xc (nT )δ(t − nT ) is input to a CT LPF
n=−∞
sin πt/T Ωs π
reconstruction filter hr (t) = πt/T with cut-off Ωc = 2 = T to
get
∞
X
xr (t) = x[n]hr (t − nT )
n=−∞
∞
X sin π(t − nT )/T
= x[n]
n=−∞
π(t − nT )/T
12 / 13
Example: Sinusoidal Sampling

13 / 13