CAN207 Signals and Systems

Part 2 – Discrete-Time Signals and Systems

Lecture-D3
Sampling and Reconstruction

Zhao Wang
[email protected]
Room EE322

1
Content
• 1. Sampling
– 1.0. What and Why?
– 1.1. Sampling in Time domain (TD)
– 1.2. Sampling in Frequency domain (FD)
– 1.3. Nyquist Theorem

• 2. Reconstruction
– 2.1 Interpolation
– 2.2 Reconstruction theory
– 2.3 Reconstruction in FD – filtering
– 2.4 Reconstruction in TD – interpolation
– 2.5 Realisation
2
1.0 Digital Processing of CT Signals
• Most signals in nature are continuous in time
=> Need a way for “digital processing of continuous-time
signals” => SAMPLING!
Ω /2
( ) [ ] Digital Signal [ ] Ideal ( )
Ideal Sampling
Processing Interpolation
(a) Ideal data flow for the digital processing of continuous-time signals

Ω /2 Ω /2
( ) Anti-aliasing [ ] [ ] Reconstruction ( )
Filter
S/H ADC DSP DAC Filter

(b) Practical data flow for the digital processing of continuous-time signals

3
1.0 Analog -> Digital ->Analog

( ) Anti-aliasing [ ] [ ] Reconstruction ( )
Filter
S/H ADC DSP DAC Filter

• Conversion of the continuous-time signal into a discrete-time signal

– Anti-aliasing filter – to prevent potentially detrimental effects of sampling
– Sample & Hold – discrete in time and keep the sampling values for a while to
allow the A/D converter to do its job
– Analog to Digital Converter (A/D) – conversion in amplitude
• Processing of the discrete-time signal
– Digital Signal Processing –Filter, digital processor
• Conversion of the processed discrete-time signal back into a cont.-
time signal
– Digital to analog converter (D/A) –to obtain the continuous signal
– Reconstruction / smoothing filter -smooth out the signal from the D/A
4
1.1 Sampling in Time domain (TD)
• A discrete-time sequence is developed by uniformly
sampling the continuous-time signal xa(t)

[ ]= ( ) = ( )

• The time variable - time t is related to the discrete time

variable n only at discrete-time instants tn
= 1/ (Sampling period, second, second/sample)
2
= = = = 1/ (Sampling frequency, Hz, cycles/second)
Ω
Ω =2 / (Sampling angular frequency, radian/second) 5
1.1 Sampling in Time domain (TD)
• Consider = cos(Ω + )
• Now [ ] = cos(Ω + )
2 Ω
= cos( + ) = cos( + )= ( )
Ω
2 Ω
– Where = =Ω
Ω

– is the (normalized) digital angular frequency of the signal

• Unit: radians/sample
– Ω is the analog angular frequency of signal
• Unit: radians/second
– Ω is the sampling analog angular frequency
• Unit: radians/second
6
1.1 Sampling in Time domain (TD)
• In mathematics, the periodic sampling is modelled as the
multiplication of continuous signal and impulse train

( )

7
1.1 Sampling in Time domain (TD)
• The system to convert the continuous-time (CT) signal ( )
to a discrete-time (DT) signal [ ] is shown:

( ) ( ) Pulse train [ ]
( )
( ) Sequence

= = ( − )

( ) ( ) [ ]

2 3 4 5 6 7

8
1.1 How signal changed after sampling
• In time domain, continuous -> discrete

• Different sampling rates, different details

– More samples = higher sampling rate/frequency = more detail
= more information kept = more resource occupation
– Less samples = lower sampling rate/frequency = less details
= more information loss = less resource occupation
• How to choose the sampling period/rate?
9
1.2 Frequency domain analyses
• Review: CTFT of a pulse train

CTFT
= ( − ) Ω =Ω (Ω − Ω )
2
Ω =2 =

Ω
Ω
1

− 0 −Ω 0 Ω

TD: Time domain FD: Frequency domain

10
 1.2 Sampling in Frequency domain (FD)
• In TD: multiplication between and
=
• In FD: convolution between Ω and Ω
CTFT

1 1
Ω = Ω ∗Ω (Ω − Ω ) = [ Ω− Ω ]
2

11
 ( ) ( ) Pulse train [ ]
1.2 Sampling in FD ( )
( )
Sequence

• An alternative expression of Ω is:

= ( − ) Recall
1
Ω = [ Ω− Ω ]
CTFT

Ω =

Ω = =
Since: =

= 1
= [ Ω− Ω ]

1 2
= −

12
1.2 Sampling in Frequency domain (FD)
• Discretization in TD => Periodicity in FD
Ω =
1/

−Ω /2 Ω /2
−Ω −Ω Ω Ω
Ω
Ω

1/

−
−2 −Ω Ω 2 13
1.3 Nyquist-Shannon Theorem
• The spectrum of the sampled signal contains all the
information of the original CT signal
– So the CT signal can be recovered without any loss;

1/

−
−2 −Ω Ω 2

– But a condition needs to be satisfied:

Ω ≤ 2Ω ≤ Ω
– The Nyquist-Shannon theorem!
14
1.3 Over-sampling and under-sampling
• Over-sampling: 2Ω <Ω

• Down-sampling: 2Ω >Ω

15
Wrap-up of SAMPLING
• What is sampling process?
– The first step to convert a continuous-time signal to a discrete-time
signal;
• In time domain: multiplication the CT signal to a pulse train,
then convert the modulated pulse train to sequence;
• In frequency domain: copy and shift (create infinite replica)
the spectrum of the CT signal;
– The CT signal can be recovered from the sampled signal if Nyquist
theorem is satisfied, i.e., 2Ω ≤ Ω
• Nyquist theorem

16
2.1 Interpolation
• What is interpolation?
– In the mathematical field of numerical analysis, interpolation is a
method of constructing new data points within the range of a
discrete set of known data points.
– In this module, interpolation is a procedure whereby we convert a
discrete-time (DT) sequence x[n] to a continuous-time (CT) function
x(t).
– Requirement: for the CT function x(t), its values at multiples of Ts
should be equal to the corresponding points of the DT sequence x[n]:
( ) = [ ]

The interpolation problem now reduces to “filling the gap” between

these instants.
17
 2.1 Interpolation methods
• 1. Zero-order/Local Interpolation
= ( )

• 2. First-order/Linear Interpolation
− <
=

• 3. Higher-order/Polynomial
Interpolation

Which one should we use in signal reconstruction?

18
2.1 Reconstruction with interpolation
• Interpolation, the fitting of a continuous signal to a set of sample
values, is a commonly used procedure for reconstructing a
function, either approximately or exactly, from samples.
– One simple interpolation procedure is the zero-order hold.
– Another useful form of interpolation is linear interpolation, whereby
adjacent sample points are connected by a straight line as shown
below.

– In more complicated interpolation formulas, sample points may be

connected by higher order polynomials or other mathematical
functions, like sinc().
19
 2.2 Reconstruction Theory
Time ( )
Domain
Interpolation

Sampling

Fourier Trans Fourier Trans

Frequency
Ω
Domain Filtering

Periodization

20
2.3 Reconstruction filter in Frequency domain
• Reconstruction or smoothing filter is used to eliminate all
the replicas of the spectrum outside the baseband

• Ideal lowpass filter

, Ω <Ω
– Frequency domain Ω =
0, Ω >Ω

21
2.4 Reconstruction filter in time domain
• This lowpass filter in time domain is a “sinc” function:
– Time domain
1
ℎ = ( Ω) Ω= Ω
2 2
sin(Ω /2) sin( / )
= = = sinc( )
Ω /2 /

– Multiply with ( Ω) (in FD) is equivalent to convolve with

ℎ ( ) (in TD), the recovered signal = ∗ℎ

– Impulse train : = ( − )

22
2.4 Reconstruction filter in time domain
• Convolution between the discretized signal and
the reconstruction lowpass filter ℎ :
= ∗ℎ = ℎ − = sinc( − )

• The values are interpolated as a

linear combination of the time-
shifted sinc functions
• The amplitudes are scaled
according to the sample values
at the center locations of the
sinc (the interpolation functions)

23
2.5 Realization

Reconstruction

( )

( )

24
Wrap-up of RECONSTRUCTION
• Continuous-time signal can be reconstructed from
the discrete-time sequence;
• Reconstruction can be realized as
– In time domain: interpolation;
– In frequency domain: filtering.
• Ideal reconstruction:
Filter with ideal Interpolate with
lowpass filter sinc function
Frequency domain Time domain

25
Lect-D3 Practices
• Exercise 1. A continuous-time signal ( ) is the linear combination of
the components with 300Hz, 1.2kHz and 3.5kHz frequencies. Sampling
( ) with a sampling frequency 2 kHz gets a sequence x[n]. Sending
x[n] through an ideal lowpass filter with the cutoff frequency of 900Hz
get a continuous-time output ( ).
• What are the frequency components in ( )?

• Exercise 2. The signal x t = sin πt + 4sin(3πt)cos(2πt), where t is

in ms, is sampled at a rate of 3 kHz.
• Find the frequency components in x t .

26
Lect-D3 Practices
• Exercise 3. The spectrum X(jΩ) of a continuous-time signal x(t) is
shown below. x(t) is sampled with the sampling angular frequency Ωs,
and get a discrete-time sequence x[k], whose spectrum is X(jω).

– a) What is the relationship between X(jΩ) and X(jω)? Write / Derive the equation
to link them;
– b) If Ωs = 200 rad/s, sketch X(jω) with all the labels for horizontal and vertical
axes;
– c) If Ωs = 200 rad/s, can x(t) be reconstructed from x[k]? Explain the reason.

27