Module 4

06 November 2022 10:51 PM

Sampling of Continuous Time (Analog) Signals
• The sampling is the process of conversion of a continuous time signal into a discrete time signal.

• The sampling is performed by taking samples of continuous time signal at definite intervals of time.

• Usually, the time interval between two successive samples will be same and such type of sampling is called periodic or
uniform sampling.
• The time interval between successive samples is called sampling time (or sampling period or sampling interval), and it is
denoted by Ts. The unit of sampling period is second (s). (The lower units are milli-second (ms) and micro-second ms)).

• The inverse of sampling period is called sampling frequency (or sampling rate), and it is denoted by Fs. The unit of
sampling frequency is Hertz (Hz). (The higher units are kHz and MHz).

• A sequence 𝑓[𝑛] (or 𝑓[𝑘]) can be obtained from a continuous-time signal 𝑓(𝑡) through sampling. Here, the value of 𝑓[𝑛]
for some integer 𝑛 is equal to 𝑓(𝑛𝑇 ), the value of 𝑓(𝑡) at time 𝑡=𝑛𝑇 .

• So, we sample at regular intervals of width 𝑻𝒔 seconds or, equivalently, with a sampling rate of 𝑓 =1/𝑇 Hz.
 loss of signal information
• Loss of information about the signal 𝑓(𝑡) can be made insignificant if we sample at a very high rate (i.e., select very
small 𝑇 ).

The question now is whether or not, for a finite sampling rate, 𝑓 , can we perfectly reconstruct 𝑓(𝑡) from its samples, 𝑓[𝑛].
If the answer is yes, then we need to answer the following questions:
(1) What is a sufficient sampling rate that allows for perfect signal reconstruction?
(2) Are there restrictions on the nature of the continuous-time signal 𝑓(𝑡) being sampled?
(3) Can we derive a formula to reconstruct 𝑓(𝑡) from its samples 𝑓[𝑛]?
(4) Can we realize a physical (or simulated) system that accepts 𝑓[𝑛] and generates 𝑓(𝑡)?

Shannon’s Sampling Theorem

Let 𝑓(𝑡) be a real-valued, continuous-time, band-limited (finite bandwidth) signal with bandwidth 𝐵. And let 𝑓[𝑛] be the
sequence of numbers obtained by sampling 𝑓(𝑡) at a sampling rate of 𝑓 samples per second (Hz);

i.e., a sample is taken every 𝑇 =1/𝑓 seconds. Then, 𝑓(𝑡) can be perfectly reconstructed from its samples 𝑓[𝑛] if and only
if 𝑓 >2𝐵.
The sampling rate must exceed twice the bandwidth of the signal (2𝐵 is referred to as Nyquist sampling rate).
Furthermore, 𝑓(𝑡) can be uniquely reconstructed by the following interpolation formula:

The sinc function , also called the

"sampling function," is a function that
Sampling Process arises frequently in signal processing
and the theory of Fourier transforms.
The full name of the function is "sine
cardinal".
sinc(x) = sin(x)/x

New Section 1 Page 1

 The real signal 𝑓(𝑡) is generally not bandlimited, therefore the reconstructed signal will have some degree of
distortion.

Another source of distortion is the approximate nature of the reconstruction filter (a low-pass filter) that is used
to reconstruct (interpolate) the signal.
Sampling, in theory, consists of simply modulating (multiplying) the signal 𝑓(𝑡) by another signal, 𝛿 (𝑡), which
consists of a train of unit-impulses separated (in time) by 𝑇 seconds.

Hence the sampled signal as 𝑔(𝑡) can be expressed as,

Employing the sampling property of the delta function [namely, 𝑓(𝑡)𝛿(𝑡−𝑎)=𝑓(𝑎)𝛿(𝑡−𝑎)], the above expression
can be expressed as

The following figure depicts the sampling process for an arbitrary, band limited signal.
Impulse train sampling

In order to get a better picture of the sampling process by considering the signals in the frequency domain.
Let us apply the Fourier transform to the sampled signal

Employ the frequency-convolution property 𝑓1(𝑡)𝑓2(𝑡)↔(1/2𝜋)𝐹1(𝜔)∗𝐹2(𝜔).

Also note that 𝛿 (𝑡) is a periodic signal (of fundamental frequency 𝜔 =2𝜋/𝑇 =2𝜋𝑓 ) having the following exponential
Fourier series representation

By employing the Fourier Pair 𝒆𝒋𝒏𝝎𝟎𝒕 ↔2πδ(ω−n𝝎𝟎 ) and the superposition property, then the Fourier
transform of 𝛿 (𝑡) would be

New Section 1 Page 2

 where the distributive property and the convolution with a shifted impulse, 𝐹(𝜔)∗𝛿(𝜔−n𝜔 )=F(ω−n𝜔 ) were
applied.

For convenience, express the above result in terms of the Hz frequency, 𝑓, as (recall that 𝑓 =𝜔 /2𝜋)

From the above equation, it can be stated that, in the frequency domain, the sampled signal is essentially a
superposition of shifted (by integer multiples of the sampling frequency, 𝑓 ) versions of the spectrum of the signal
𝑓(𝑡). Also, the amplitude of the spectrum 𝐺(𝑓) is scaled by the sampling rate, 𝑓 .

The following figure depicts, graphically, the magnitude spectra |𝐹(𝑓)| and |𝐺(𝑓)| as a function of frequency, 𝑓.

• Note how the magnitude spectrum |𝐹(𝑓)| is duplicated every 𝑓𝑠 (Hz) in the spectrum for |𝐺(𝑓)|.
• These duplicates are known as the image spectra.
• It shows that if the sampling frequency 𝑓 is twice the bandwidth (i.e., 𝑓 =2𝐵) of the bandlimited signal 𝑓(𝑡),
then there will be no overlap between |𝐹(𝑓)| and its images, therefore an ideal (brick-wall) low-pass filter
with a cutoff frequency 𝑓 =𝑓 /2=𝐵 (and dc gain 𝑇 ) can be used to perfectly reconstruct 𝑓(𝑡)

• Increasing the sampling rate beyond 2𝐵 (𝑓 >2𝐵) allows for further separation between |𝐹(𝑓)| and its adjacent
image spectra, and may allow for a practical filter (say a 4th or higher order Butterworth filter) to properly
New Section 1 Page 3
• Increasing the sampling rate beyond 2𝐵 (𝑓 >2𝐵) allows for further separation between |𝐹(𝑓)| and its adjacent
image spectra, and may allow for a practical filter (say a 4th or higher order Butterworth filter) to properly
reconstruct the signal 𝑓(𝑡).

• On the other hand if we under sample, 𝑓 <2𝐵, then there will be overlap with the image spectra and the
reconstructed signal will always have distortion (aliasing).

• The 𝑓 =2𝐵 threshold value on the sampling frequency is known as the Nyquist sampling rate.

Aliasing is an effect of the sampling that causes different signals to become indistinguishable.
Due to aliasing, the signal reconstructed from samples may become different than the original
continuous signal. This can drastically deteriorate the performance if proper care is not taken.

New Section 1 Page 4

07 November 2022 09:55 AM

Signal Reconstruction

In control system, hold operation becomes the most popular way of reconstruction due to its simplicity and low cost.

Problem of data reconstruction can be formulated as:

“ Given a sequence of numbers, f(0), f(T), f(2T), · · · , f(kt), · · · , a continuous time signal f(t), t ≥0, is to be reconstructed from the information
contained in the sequence.”

Zero Order Hold

• Higher the order of the derivatives to be estimated is, larger will be the number of delayed pulses required.

• Since time delay degrades the stability of a closed loop control system, using higher order derivatives of f(t) for more accurate
reconstruction often causes serious stability problem.

• Moreover a high order extrapolation requires complex circuitry and results in high cost.

• For the above reasons, use of only the first term in the power series to approximate 𝑓(𝑡) during the time interval 𝑘𝑇 ≤ 𝑡 < (𝑘 + 1)𝑇 is very
popular and the device for this type of extrapolation is known as zero-order extrapolator or zero order hold.
Figure illustrates the operation of a ZOH

The accuracy of zero order hold (ZOH) depends on the sampling frequency. When 𝑇 → 0, the output of ZOH approaches the
continuous time signal.

Zero order hold is again a linear device which satisfies the principle of superposition.

Transfer function of Zero order hold circuit

Mathematically, the output of the zero-order hold is given by,

𝑥 (𝑡) = 𝑥(𝑘); 𝑓𝑜𝑟 𝑘𝑇 ≤ 𝑘 ≤ (𝑘 + 1)𝑇

Therefore,
𝑥 (𝑡) = 𝑥(0); 𝑓𝑜𝑟 0 ≤ 𝑡 ≤ 𝑇

𝑥 (𝑡) = 𝑥(𝑇); 𝑓𝑜𝑟 𝑇 ≤ 𝑡 ≤ 2𝑇

𝑥 (𝑡) = 𝑥(2𝑇); 𝑓𝑜𝑟2𝑇 ≤ 𝑡 ≤ 3𝑇 𝑎𝑛𝑑 𝑠𝑜 𝑜𝑛

Also, the impulse response of the zero-order hold is given by,

1 𝑓𝑜𝑟0 ≤ 𝑡 ≤ 𝑇
ℎ(𝑡) =
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

𝑥 𝑡 ℎ(𝑡)

New Section 1 Page 5

 The output 𝑥 (𝑡) of a zero-order hold is given by the convolution of its impulse response ℎ(𝑡) and its
input 𝑥(𝑘𝑇), i.e.,

𝑥 (𝑡) = 𝑥(𝑘𝑇) ∗ ℎ(𝑡)

⇒ 𝑥 (𝑡) = 𝑥(2𝑇)ℎ(𝑡 − 𝑘𝑇)

ℎ(𝑡)

Since the impulse response of the zero-order hold is given by

ℎ(𝑡) = 𝑢(𝑡) − 𝑢(𝑡 − 𝑇)

⇒ ℎ(𝑡 − 𝑘𝑇) = 𝑢(𝑡 − 𝑘𝑇) − 𝑢[𝑡 − (𝑘 + 1)𝑇]

∴𝑥 (𝑡) = 𝑥 𝑘𝑇 {𝑢 𝑡 − 𝑘𝑇 − 𝑢 𝑡 − 𝑘 + 1 𝑇 }

By taking Laplace transform on both sides, we get,

𝐿[𝑥 (𝑡)] = 𝐿 𝑥(𝑘𝑇){𝑢(𝑡 − 𝑘𝑇) − 𝑢[𝑡 − (𝑘 + 1)𝑇]}

⇒ 𝑒 𝑒 ( ) 1−𝑒
𝑥 (𝑠) = 𝑥(𝑘𝑇) ⎯⎯⎯⎯⎯− ⎯⎯⎯⎯⎯⎯⎯⎯ = ⎯⎯⎯⎯⎯⎯⎯⎯ 𝑥(𝑘𝑇)𝑒
𝑠 𝑠 𝑠

Transfer function of Zero order hold circuit

1−e
x (s) = ⎯⎯⎯⎯⎯⎯⎯⎯ X ∗ (s)
s
x (s) 1−e
⎯⎯⎯⎯⎯ = ⎯⎯⎯⎯⎯⎯⎯⎯
X ∗ (s) s

First order Hold

When the 1st two terms of the power series are used to extrapolate f(t), over the time interval 𝑘𝑇 ≤ 𝑡 < (𝑘 + 1)𝑇 , the device is called
a first order hold (FOH). Thus

Impulse response of FOH is obtained by applying a unit impulse at 𝑡 = 0, the corresponding output is obtained by
setting 𝑘 = 0, 1, 2, . . . . .

New Section 1 Page 6

One can verify that according to the above expression, when 0 ≤ 𝑡 < 𝑇, only the first term produces a nonzero value which is nothing but
(1 + 𝑡/𝑇).

Similarly, when 𝑇 ≤ 𝑡 < 2𝑇, first two terms produce nonzero values and the resultant is (1 − 𝑡/𝑇). In case of 𝑡 ≥ 2𝑇, all three terms
produce nonzero values and the resultant is 0.

Comparison of the reconstructed outputs of ZOH and FOH

New Section 1 Page 7

21 November 2022 10:59 AM

New Section 1 Page 8

New Section 1 Page 9
New Section 1 Page 10
23 November 2022 10:56 AM

New Section 1 Page 11

New Section 1 Page 12
New Section 1 Page 13
New Section 1 Page 14
New Section 1 Page 15
23 November 2022 01:56 PM

New Section 1 Page 16

New Section 1 Page 17
New Section 1 Page 18
New Section 1 Page 19
New Section 1 Page 20
New Section 1 Page 21
12 December 2022 09:33 AM

New Section 1 Page 22

New Section 1 Page 23
New Section 1 Page 24
New Section 1 Page 25
05 December 2022 10:04 AM

New Section 1 Page 26