Sampling and Reconstruction

DSP Material

Contact us for specific DSP Material requests, support in

assignments, and exam preparation at [email protected]

www.dspmaterial.com

Contents
1 Introduction 2

2 Sampling in Time and Frequency 3

2.1 Sampling in the Time Domain . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 The Effect of Sampling in the Frequency Domain . . . . . . . . . . . . . 4
2.3 Nyquist Criterion and Aliasing . . . . . . . . . . . . . . . . . . . . . . . . 5
2.4 Derivation of the Fourier transform of the sampled signal . . . . . . . . 5

3 Reconstruction in Time and Frequency 7

3.1 Reconstruction in the Time Domain . . . . . . . . . . . . . . . . . . . . . 7
3.2 Reconstruction in the Frequency Domain with Low-Pass Filtering . . . . 8
3.3 Duality Between Time and Frequency Domains . . . . . . . . . . . . . . 10

1
1 Introduction
This tutorial explores the fundamental concepts of sampling and reconstruction, focusing
on their significance in digital signal processing. Sampling converts a continuous signal
into a discrete signal, enabling digital processing, storage, and transmission. Reconstruc-
tion is the process of recovering the original continuous signal from its discrete samples.
This process is important as it enables and supports the following:

• Digital Transformation: Sampling bridges the analog and digital worlds by con-
verting continuous signals into discrete data that can be processed, stored, and
transmitted in digital systems.

• Efficiency in Processing: Digital signals are easier to manipulate, compress, and

enhance compared to their analog counterparts. Sampling enables efficient data
handling in applications such as audio processing, video streaming, and telecom-
munication.

• Preserving Information: Proper sampling ensures that no information is lost

during conversion, allowing the signal to be faithfully reconstructed without distor-
tion or aliasing.

• Wide Applicability:

– Audio Processing: Music streaming services rely on sampling to digitize

sound for storage and playback.
– Medical Imaging: CT scans, MRIs, and ultrasound systems use sampling
to convert analog measurements into digital data for analysis.
– Radar and Communication Systems: Sampling is critical in modern radar
systems and mobile communication technologies.

• Standardization: Sampling enables the standardization of data formats (e.g.,

MP3, JPEG), ensuring interoperability between devices and systems.

• Signal Reconstruction: Sampling provides the foundation for reconstructing sig-

nals accurately, which is essential in error-free communication systems.

• Reduced Noise Sensitivity: Digital signals sampled and processed correctly are
less prone to noise interference compared to analog signals.

2
2 Sampling in Time and Frequency
Sampling a signal in the time domain creates periodic replicas of its spectrum in the
frequency domain. This section explains this process.

2.1 Sampling in the Time Domain

When a continuous-time signal x(t) is sampled at intervals Ts (sampling period), the
sampled signal can be written as:
∞
X
xs (t) = x(t) · δ(t − nTs )
n=−∞

where δ(t) is the Dirac delta function, and Ts = f1s is the sampling period.
In discrete terms, the samples are represented as:

x[n] = x(nTs ),

where n is an integer representing the sample index n = 0, ±1, ±2, . . ..

3
2.2 The Effect of Sampling in the Frequency Domain
The Fourier transform of the sampled signal xs (t) is given by:
∞  
1 X k
Xs (f ) = X f−
Ts k=−∞ Ts

This equation shows that the spectrum of the sampled signal consists of periodic replicas
of the original spectrum X(f ), separated by the sampling frequency fs = T1s .
We note that the Fourier transform of a continuous-time signal x(t) is defined as:
Z ∞
X(f ) = x(t)e−j2πf t dt
−∞

Explanation of Terms:
- X(f ): The Fourier transform of x(t), representing the signal in the frequency domain.
- x(t): The original time-domain signal.
- f : Frequency in Hz.
- e−j2πf t : The complex exponential kernel, which decomposes x(t) into its frequency
components.

4
2.3 Nyquist Criterion and Aliasing
To avoid overlapping of the spectral replicas, the sampling frequency fs must satisfy:

fs > 2fm

where fm is the maximum frequency component of the original signal. If fs ≤ 2fm ,

aliasing occurs, causing the spectral replicas to overlap. This results in distortion and
the original signal cannot be reconstructed.

2.4 Derivation of the Fourier transform of the sampled signal

We consider the sampling process and the mathematical properties of the Fourier trans-
form.
Step 1: Sampling a Signal in the Time Domain Let x(t) be a continuous-time signal.
Its sampled version xs (t) is defined as:

xs (t) = x(t) · p(t)

where p(t) is the impulse train:

5
 ∞
X
p(t) = δ(t − nTs )
n=−∞

Here:
- Ts = f1s is the sampling period.
- fs is the sampling frequency.
The product xs (t) represents the sampled signal.
Step 2: Fourier Transform of the Impulse Train The Fourier transform of p(t) is
another impulse train P (f ), given by:
∞
1 X
P (f ) = δ(f − kfs )
Ts k=−∞

where k is an integer, and the impulses are spaced by fs = T1s in the frequency domain.
Step 3: Fourier Transform of the Sampled Signal Using the modulation property of
the Fourier transform, the Fourier transform of xs (t) becomes:

Xs (f ) = F{x(t) · p(t)}
By the convolution theorem, the Fourier transform of a product in the time domain
is the convolution of their respective Fourier transforms:

Xs (f ) = F{x(t)} ∗ F{p(t)}
Substituting F{x(t)} = X(f ) (the Fourier transform of the original signal) and
F{p(t)} = P (f ), we have:

Xs (f ) = X(f ) ∗ P (f )
Step 4: Convolving X(f ) with P (f ) Since P (f ) is an impulse train, the convolution
simplifies to a replication of X(f ) at multiples of fs :
∞
1 X
Xs (f ) = X(f − kfs )
Ts k=−∞
Here:
- X(f − kfs ): X(f ) shifted by multiples of fs .
- T1s : A scaling factor due to the sampling period.
This result shows that the spectrum of the sampled signal Xs (f ) consists of the original
spectrum X(f ) periodically replicated at intervals of fs .
Step 5: Nyquist Criterion For perfect reconstruction, the replicas X(f − kfs ) should
not overlap. This requires the sampling frequency fs to be greater than twice the maxi-
mum frequency fmax of X(f ):

fs > 2fmax
This condition is the Nyquist criterion, which ensures that the original signal can be
recovered from its samples without aliasing.
Summary: Fourier Transform of the Sampled Signal The Fourier transform of the
sampled signal xs (t) is:

6
 ∞
1 X
Xs (f ) = X(f − kfs )
Ts k=−∞

This result describes the periodic replication of the spectrum X(f ) in the frequency
domain due to sampling.

3 Reconstruction in Time and Frequency

Next, we demonstrate the reconstruction process of a sampled signal xs (t) in both the time
domain (using sinc interpolation) and the frequency domain (using low-pass filtering). We
also comment on the duality between convolution in the time domain and multiplication
in the frequency domain.

3.1 Reconstruction in the Time Domain

The sampled signal xs (t) is represented as:

xs (t) = x(t) · p(t)

where p(t) is the impulse train:
∞
X
p(t) = δ(t − nTs )
n=−∞

In the time domain, xs (t) is a train of scaled impulses:

∞
X
xs (t) = x(nTs )δ(t − nTs )
n=−∞

To reconstruct x(t), we use sinc interpolation, expressed as:

∞  
X t − nTs
x(t) = x(nTs ) · sinc
n=−∞
Ts

Why the sinc Function?

The sinc function is the impulse response of an ideal low-pass filter in the time domain:
 
t sin (πt/Ts )
sinc =
Ts πt/Ts
- Each sinc kernel is centered at t = nTs .
- The sinc kernels are scaled by the sampled values x(nTs ).
The convolution in the time domain is given by:

x(t) = xs (t) ∗ h(t)

 
where h(t) = sinc Tts . This convolution spreads the impulses of xs (t) into overlap-
ping sinc functions, reconstructing x(t).

7
3.2 Reconstruction in the Frequency Domain with Low-Pass
Filtering
The sampled signal xs (t) in the frequency domain is represented as:
∞
1 X
Xs (f ) = X(f − kfs )
Ts k=−∞
Here:
- X(f ) is the Fourier transform of x(t).
- Xs (f ) is the spectrum of the sampled signal.
- fs = T1s is the sampling frequency.
To recover X(f ), we apply an ideal low-pass filter H(f ) with a cutoff frequency
fc = f2s . The filter is defined as:
(
1, |f | ≤ f2s
H(f ) =
0, |f | > f2s
The output spectrum is:

Xrec (f ) = Xs (f ) · H(f )

8
 The multiplication in the frequency domain isolates the baseband spectrum X(f )
while eliminating higher-frequency replicas X(f − kfs ) for |k| ≥ 1.

Inverse Fourier Transform

Finally, taking the inverse Fourier transform of Xrec (f ), we obtain the reconstructed
time-domain signal:
Z ∞
x(t) = Xrec (f )ej2πf t df
−∞

9
3.3 Duality Between Time and Frequency Domains
Time Domain:
- Reconstruction is a convolution of the sampled signal xs (t) with a sinc function h(t).
- The sinc function spreads each impulse in xs (t) into a continuous waveform, filling
the gaps between samples.
Frequency Domain:
- Reconstruction is a multiplication of the sampled signal spectrum Xs (f ) with an
ideal low-pass filter H(f ).
- The filter isolates the baseband spectrum X(f ), removing higher-frequency replicas.
Key Observation:
1. Convolution in the Time Domain corresponds to Multiplication in the Frequency
Domain.
2. Multiplication in the Frequency Domain corresponds to Convolution in the Time
Domain.
This duality is a fundamental principle of the Fourier transform. In reconstruction:
- In the time domain, the sinc kernel smooths the impulses into a continuous signal.
- In the frequency domain, the low-pass filter eliminates aliasing artifacts by retaining
only the baseband.

10
Key Insights
• Time-Frequency Duality: Sampling in the time domain creates periodic replicas
in the frequency domain. Reconstruction in the time domain corresponds to filtering
in the frequency domain.

• Role of Sinc Function: The sinc function is the impulse response of the ideal
low-pass filter and serves as the interpolation kernel in the time domain.

• Nyquist Rate: The sampling frequency must be at least twice the highest fre-
quency component of the signal to ensure perfect reconstruction.

11