EE314: Digital Signal Processing

Assignment No: 5

Course Instructor: Siddhartha Sarma Date: 24 Oct 2024

Submission deadline: 12 Nov 2024, 9 AM

Related topics:
• Polyphase decomposition
• Spectral analysis using DFT
• Window sequences

Notations:
• X(jΩ) represents the continuous-time Fourier transform (CTFT) of a signal x(t).
• X(ejω ) represents the discrete-time Fourier transform (DTFT) of a sequence x[n].
• δ[n], δ(t) denote discrete-time and continuous-time impulse functions, respectively.

1. (a) Consider the system in Fig. (1a) where a filter H(z) is followed by a
compressor. Suppose that H(z) has an impulse response given by:
(
0.5n , 0 ≤ n ≤ 11
h[n] = (1)
0, otherwise

x[n] H(z) ↓2 y[n] x[n] ↑3 H(z) y[n]

(a) Block diagram of the system (b) Block diagram of the system
mentioned in Q. 1a. mentioned in Q. 1b.

Figure 1: Block diagrams corresponding to Q.1

The efficiency of the system can be improved by implementing the filter

H(z) and the compressor using a polyphase decomposition. Draw an ef-
ficient polyphase structure for this system. Please specify the filters you
use.
(b) Now consider the system in Fig. (1b) where a filter H(z) is preceded by an
expander. Suppose that H(z) has the same impulse response given in Eq.
(1). Similar to the previous question draw an efficient polyphase structure
for this system and specify the filters you use.

1
2. For the system shown in Fig. 2 and 3, determine whether or not it is possible to
specify a choice for H2 (z) in System 2 so that y2 [n] = y1 [n] when x2 [n] = x1 [n]
and H1 (z) is as specified. If possible, specify H2 (z). If it is not possible clearly
explain.
(
x1 [n/2], n/2 integer
Note that for w1 [n] =
0, otherwise
(
w2 [n/2], n/2 integer
and y2 [n] =
0, otherwise

System 1:
H1 (z)
w1 [n]
x[n] ↑2 1 + z3 y[n]

Figure 2: Block diagram of the system mentioned in Q. 2.

System 2:
w2 [n]
x[n] H2 (z) ↑2 y[n]

Figure 3: Block diagram of the system mentioned in Q. 2.

3. A continuous-time signal xc (t) is bandlimited to 5 KHz, i.e., Xc (jΩ) = 0 for

|Ω| ≥ 2π(5000). xc (t) is sampled with period T , producing the sequence x[n] =
xc (nT ). To examine the spectral properties of the signal, we compute the N -
point DFT of a segment of N samples of x[n] using a computer program that
requires N = 2ν , where ν is a positive integer.
Determine the minimum value for N and the range of sampling rates, Fmin <
1/T < Fmax such that aliasing is avoided, and the effective spacing between
DFT values is less than 5 Hz, i.e., equivalent continuous-time frequencies at
which the Fourier transform is evaluated are separated by less than 5 Hz.

4. Consider estimating the spectrum of a discrete-time signal x[n] using the DFT
with a Hamming window applied to x[n]. A conservative rule of thumb for the
frequency resolution of windowed DFT analysis is that the frequency resolution
is equal to the width of the main lobe of W (ejω ). You wish to be able to resolve
sinusoidal signals that are separated by as little as π/100 in ω. In addition,
your window length L is constrained to be a power of 2. What is the minimum
length L = 2ν that will meet your requirement?

5. Let x[n] be a signal with single sinusoidal component. The signal is windowed
with an L-point Hamming window and an L-point rectangular window to obtain
v1 [n] and v2 [n], respectively. The DTFT of vi [n] is denoted by Vi (ejω ), where
i = 1, 2. Find the magnitude and width of the peak of |Vi (ejω )|, i = 1, 2.

2
6. Figure 4 shows the spectrogram of a chirp signal of the form

x[n] = sin ω0 n + λn2 /2 .

Note that the spectrogram is a representation of the magnitude of X[n, k],

Figure 4: Spectrogram mentioned in Q. 6

where the dark regions indicate large values of |X[n, k]|. Based on the figure,
estimate ω0 , λ.

7. It is desired to estimate the spectrum of x[n] by applying a 512-point Kaiser

window to the signal before computing X(ejω ).

(a) The requirement for the frequency resolution of the system specify that
the largest allowable mainlobe for the Kaiser window is π/100. What is
the best side-lobe attenuation possible under these constraints?
(b) Suppose that you know that x[n] contains two sinusoidal components at
least π/50 apart, and that the amplitude of the stronger component is 1.
Based on your answer to part (a), give a threshold on the smallest value
of the weaker component you would expect see over the side lobe of the
stronger sinusoid.

8. A real-valued continuous-time segment of a signal xc (t) is sampled at a rate of

20000 samples/s, yielding a 1000-point finite length discrete-time sequence x[n]
that is non-zero in the interval 0 ≤ n ≤ 999. It is known that xc (t) is also
bandlimited to 10000 Hz (sampling didn’t introduce any aliasing distortion).
X[k] denotes the 1000-point DFT of x[n]. X[800] is known to have value 1 + j.

3
 (a) From this information given, can you determine X[k] at any other values
of k? If so, state which value(s) of k and what corresponding value of X[k]
is. If not, explain why not.
(b) From the information given, state the value(s) of Ω for which Xc (jΩ) is
known and the corresponding value(s) of Xc (jΩ).

9. Sketch the spectrogram obtained by using a 256-point rectangular windows and

256-point DFT on the signal
π  πn 
x[n] = cos + 1000 sin
4 8000
for the interval 0 ≤ n ≤ 16000.

10. Let x[n] = cos(2πn/5) and v[n] be the sequence obtained by applying a 32-
point rectangular window to x[n] before computing V (ejω ). Sketch |V (ejω )| for
−π ≤ ω ≤ π, labeling the frequencies of all peaks and the first null on either side
of the peak. In addition, label the amplitudes of the peaks and the strongest
side lobe of each peak.

All questions were taken from the following books

• Oppenheim, Schafer, “Discrete-time Signal Processing”, Pearson, 2nd and 3rd Ed