Digital Signal Processing

Solved HW for Day 12

Question 1: Another view of Sampling
One of the standard ways of describing the sampling operation relies on the concept of
”modulation by a pulse train”. Choose a sampling interval Ts and define a continuous-time
pulse train p(t) as:
X∞
p(t) = δ(t − kTs ).
k=−∞
The Fourier Transform of the pulse train is
∞
2π X 2π
P(jΩ) = δ(Ω − k )
Ts Ts
k=−∞

This is tricky to show, so just take the result as is. The ”sampled” signal is simply the
modulation of an arbitrary-continuous time signal x(t) by the pulse train:
xs (t) = p(t) x(t).
Derive the Fourier transform of xs (t) and show that if x(t) is bandlimited to π/Ts then we
can reconstruct x(t) from xs (t).
1
Solution of question 1
By using the modulation theorem, we have

Xs (jΩ) = X (jΩ)P(jΩ)
Z
2π
Z X  2π

= X (j Ω̃)P(j(Ω − Ω̃))d Ω̃ = X (j Ω̃) δ Ω − Ω̃ − k d Ω̃
R Ts R Ts
k∈Z

Xs (jΩ) = X (jΩ) ∗ P(jΩ)

Z
2π
Z X  2π

= X (j Ω̃)P(j(Ω − Ω̃))d Ω̃ = X (j Ω̃) δ Ω − Ω̃ − k d Ω̃
R Ts R Ts
k∈Z
    
2π X 2π 2π X 2π
Z
= X (j Ω̃)δ Ω − Ω̃ − k d Ω̃ = X j Ω−k .
Ts R Ts Ts Ts
k∈Z k∈Z

2
Solution of question 1

In other words, the spectrum of the delta-modulated signal is just the periodic repetition (with
period (2π/Ts )) of the original spectrum. If the latter is bandlimited to (π/Ts ) there will be
no overlap and therefore x(t) can be obtained simply by lowpass filtering xs (t) (in the
continuous-time domain).

3
Question 2: Aliasing can be good! Part-1
Consider a signal xc (t) with the following spectrum:

◮ What is the bandwidth of the signal? What is the minimum sampling period in order to
satisfy the sampling theorem?
◮ Take a sampling period Ts = Ωπ0 ; clearly, with this sampling period, there will be aliasing.
Plot the DTFT of the discrete-time signal xa [n] = xc (nTs ).
◮ Suggest a block diagram to reconstruct xc (t) from xa [n].
◮ Therefore, we can exploit aliasing to reduce the sampling frequency necessary to sample a
bandpass signal. What is the minimum sampling frequency to be able to reconstruct with
the above strategy a real signal whose frequency support on the positive axis is [Ω0 , Ω1 ]?
4
Solution of question 2

◮ The highest nonzero frequency is 2Ω0 and therefore xc (t) is 2Ω0 -bandlimited for a total
bandwidth of 4Ω0 .
◮ The maximum sampling period (i.e. the inverse of the minimum sampling frequency)
which satisfies the sampling theorem is Ts = π/(2Ω0 ).
◮ Note however that the total support over which the (positive) spectrum is nonzero is the
interval [Ω0 , 2Ω0 ] so that one could say that the total effective positive bandwidth of the
signal is just Ω0 .

5
Solution of question 2
Q: Plot the DTFT of the discrete-time signal xa [n] = xc (nTs ).

◮ The digital spectrum will be the rescaled version of the periodized continuous-time
spectrum
∞
X
X̃c (jΩ) = Xc (j(Ω − 2kΩ0 )).
k=−∞

◮ The general term Xc (jΩ − j2kΩ0 ) is nonzero for Ω0 ≤ |Ω − 2kΩ0 | ≤ 2Ω0 for k ∈ Z. or
equivalently
(2k + 1)Ω0 ≤ Ω ≤ (2k + 2)Ω0
(2k − 2)Ω0 ≤ Ω ≤ (2k − 1)Ω0
◮ Non-overlapping intervals, therefore, no disruptive superpositions of the copies of the
spectrum!
6
Solution of question 2
◮ The digital spectrum is
∞
1 X ω 2πk
X (e jω ) = Xc (j −j )
Ts Ts Ts
k=−∞

which looks like this (with 2π-periodicity, of course):

7
Solution of question 2

Q: Suggest a block diagram to reconstruct xc (t) from xa [n].

Here’s a possible scheme (verify that it works):

◮ Sinc-interpolate xa [n] with period Ts to obtain xb (t)
◮ Multiply xb (t) by cos(2Ω0 t) in the continuous time domain to obtain xp (t) (i.e. modulate
by a carrier at frequency (Ω0 /π) Hz).
◮ Bandpass filter xp (t) with an ideal bandpass filter with (positive) passband equal to
[Ω0 , 2Ω0 ] to obtain xc (t).

8
Solution of question 2
Q: What is the minimum sampling frequency to be able to reconstruct with the above strategy
a real signal whose frequency support on the positive axis is [Ω0 , Ω1 ]?

◮ The effective positive bandwidth of such a signal is Ω∆ = (Ω1 − Ω0 ).

◮ The sampling frequency must be at least equal to the effective total bandwidth, so a first
condition on the maximum allowable sampling period: Tmax < π/Ω∆ .
◮ Case 1: assume Ω1 is a multiple of the bandwidth, i.e. Ω1 = MΩ∆ for some integer M (in
the previous case, it was M = 2).
In this case, the argument we made in the previous point can be easily generalized: if we
pick Ts = π/Ω∆ and sample we have that
∞
X
X̃c (jΩ) = Xc (j(Ω − 2kΩ∆ )).
k=−∞

9
Solution of question 2

◮ The general term Xc (jΩ − j2kΩ∆ ) is nonzero only for

Ω0 ≤ |Ω − 2kΩ∆ | ≤ Ω1 for k ∈ Z.

◮ Since Ω0 = Ω1 − Ω∆ = (M − 1)Ω∆ , this translates to

(2k + M − 1)Ω∆ ≤ Ω ≤ (2k + M)Ω∆

(2k − M)Ω∆ ≤ Ω ≤ (2k − M + 1)Ω∆

again non-overlapping intervals!

10
Solution of question 2

◮ Case 2: Ω1 is not a multiple of the bandwidth.

The easiest thing to do is to change the lower frequency Ω0 to a new frequency Ω′0 so
that the new bandwidth Ω1 − Ω′0 divides Ω1 exactly. In other words we set a new lower
frequency Ω′0 so that it will be Ω1 = M(Ω1 − Ω′0 ) for some integer M; it is easy to see that
 
Ω1
M= .
Ω1 − Ω0

since this is the maximum number of copies of the Ω∆ -wide spectrum which fit with no
overlap in the [0, Ω0 ] interval.
◮ If Ω∆ > Ω0 we cannot hope to reduce the sampling frequency and we have to use normal
sampling. This artificial change of frequency will leave a small empty “gap” in the new
bandwidth [Ω′0 , Ω1 ], but that’s no problem.

11
Solution of question 2

◮ Now we use the previous result and sample with Ts = π/(Ω1 − Ω′0 ) with no overlap.
Since (Ω1 − Ω′0 ) = Ω1 /M, we have that, in conclusion, the maximum sampling period is
 
π Ω1
Tmax =
Ω1 Ω1 − Ω0

i.e. we obtain a sampling frequency reduction factor of ⌊Ω1 /(Ω1 − Ω0 )⌋.

12
Question 3
Consider a bandlimited continuous-time signal x(t) with a spectrum X (jΩ) as sketched in the
following figure:

Sketch the DTFT of the output signal for each of the following systems, where the frequency
of the raw sampler is indicated on each line:

◮
13
Solution of question 3

14