Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
University of Manchester
Manchester School of Engineering
Electrical Engineering Programmes
________________________________________________________________________
{…, 0, …, 0, 1, 0, −1, 0, …, 0, … }
{…, 0, …, 1, 2, 3, 2, 1, …, 0, … }. [4 marks]
Show that the system's gain and phase responses are given, respectively, by:
G(Ω) = | 2sin(Ω) | and φ(Ω) = π/2 − Ω.
where Ω denotes frequency in radians per sampling interval. [4 marks]
Explain the term "linear phase" and state whether this system is linear phase. [2 marks]
(b) Design a sixth order FIR high-pass digital filter with cut-off frequency at one quarter of the
sampling frequency. [ 10 marks]
2. (a) Explain why analogue signals are generally low-pass filtered before they are
converted to digital form. For a given signal, why does increasing the sampling
rate simplify the analogue filter required? [8 marks]
(c) How and to what extent would the maximum achievable SQNR be affected by
increasing the sampling rate to 32 kHz and replacing the 12-bit analogue-to-
digital converter by a 16-bit converter? [4 marks]
3. (a) Explain how the poles and zeros of the discrete time system function H(z) affect the
stability, the gain response and the phase response of the system. [7 marks]
(b) Give the system function H(z) and a signal-flow-graph for the following difference equation:
H (e j Ω ) = 1 + cos (3 Ω ) − j sin (3 Ω )
what would be its output when the input signal is {x[n]} with
4. Given that the system function of a third order Butterworth type analogue low-pass
analogue filter with a 3 dB cut-off frequency of one radian/second is:
1
H(s) =
(1 + s ) ( 1 + s + s2 )
use the bilinear transformation to design a third order low-pass digital filter with a 3 dB cut-off
frequency at one quarter of the sampling frequency. [8 marks]
With the aid of simple sketch graphs explain how frequency warping affects the frequency
response of the digital filter. [4 marks]
Give a flow-diagram to indicate how the digital filter would be implemented on a DSP
microprocessor with floating point arithmetic. [4 marks]
EE3271 Jan03 Exam Solutions 3 11/11/04 BMGC
5.(a) Briefly outline four of the main advantages and one disadvantage of digital as opposed to
analogue signal processing. [5 marks]
(b) Given that the discrete time Fourier transform (DTFT) of the 20th order rectangular
window sequence, {w[n]}, is W( e j Ω ) with magnitude spectrum shown in figure 1,
show that the DTFT of a cosine sequence of frequency Ω 0 radians per sampling interval,
windowed by {w[n]}, is 0.5 W( ej(Ω - Ω 0) ) + 0.5 W( ej(Ω + Ω 0) ). [3 marks]
Sketch the modulus of the DTFT of the windowed cosine sequence in two cases:
In each case, sketch the expected form of the modulus of the 20th order discrete Fourier
transform (DFT) of the rectangularly windowed cosine sequence. [4 marks]
(c) Referring to your answer above, explain why non-rectangular windows (Hann, etc.) are
generally preferred to the rectangular window when using the DFT or FFT for spectral
estimation? [4 marks]
20
15
Modulus DTFT
10
0
−π −0.8π −0.6π −0.4π −0.2π 0 0.2π 0.4π 0.6π 0.8π π
Radians/sample
-5
Figure 1
EE3271 Jan03 Exam Solutions 4 11/11/04 BMGC
Solutions
∞ 2
1(a) y[n] = ∑ h[m]x[n − m] = ∑ h[m]x[n − m]
m = −∞ m=0
= x[n] - x[n-2]
Solution is:
{ …, 0, 1, 2, 2, 0, -2, -2, -1, 0, …, 0, …}
X(e j Ω ) = 1 - e - 2 j Ω
= e - j Ω (e j Ω - e - j Ω)
= 2 j e - j Ω sin (Ω)
Not linear phase as the straight line phase-response does not pass through the origin.
"Linear phase" means that −φ(Ω) / Ω = constant, which means that the delay is constant for all
frequencies
EE3271 Jan03 Exam Solutions 5 11/11/04 BMGC
1 π
h[n] =
2π ∫π
−
G (Ω)e jΩn dΩ
1 π 1 −π / 2
=
2π ∫π e jΩn dΩ
/2
+
2π ∫π −
e jΩn dΩ
1 π
=
2π ∫π /2
(e jΩn + e − jΩn )dΩ
1 π
=
2π ∫π /2
(cos(Ωn))dΩ
On rectangularly windowing and delaying by 3 samples, we obtain the causal finite impulse
response:
{h[n]} = { …, 0, 0, -1/(3π), 0, -1/π, 0.5, -1/π, 0, 1/(3π), 0, …, 0, …}
∞
1
X(e jωT ) =
T
∑
n = −∞
X a ( j (ω − nω 0 )) with ω 0 = 2π / T
Where Xa(jω) is not band-limited to the frequency range -π/T to π/T, overlap occurs.
If now we take Xs( ejωT ) to represent Xa( jω )/T for -π/T < ω< π/T, it will be distorted.
This is aliasing distortion.
To avoid aliasing distortion, low-pass filter xa( t ) to band-limit the signal to ±fS/2 Hz
before sampling at fs Hz. It then satisfies “ Nyquist sampling criterion ”.
(1) Need very sharp analogue anti-aliasing filter to remove everything above F Hz.
(2) Need very sharp analogue reconstruction filter to eliminate images (ghosts):
Xs(j6.284f)
-fs/2 fs/2
REMOVE REMOVE f
-F F 2F Hz
Increasing fS, e.g. to 44.1 kHz when F is fixed at 20 kHz modifies this diagram as follows:-
Xs(j6.28f)
fs/2
-fs/2
REMOVE REMOVE f
-F F 2F Hz
Analogue filtering is now easier. Need only remove everything above fS - F Hz. If fs is further
increased, and F does not change, removing spectrum above fs -F without affecting -F to F
becomes even easier.
EE3271 Jan03 Exam Solutions 7 11/11/04 BMGC
The quantisation noise spectrum may be assumed white in the frequency range 0 to fs / 2 Hz.
In the time-domain, the quantisation error samples may be assumed random and statistically
uniformly distributed between -∆/2 and ∆/2.
2 (c) Doubling fs does not change the SQNR directly, but spreads the quantisation noise across
spectrum -16 kHz to 16 kHz. We can filter off noise above 7 kHz thus removing approximately
half its power saving 3 dB in SQNR.
Replacing the 12-bit ADC by a 16-bit device gains 24 dB.
Therefore the SQNR increases to 97.7 + 3 dB. = 100.7 dB.
EE3271 Jan03 Exam Solutions 8 11/11/04 BMGC
4. (concluded)
L1: INPUT X
Set W = X/3 - W2 / 2
Set V = W + 2 * W1 + W2
Set W2 = W1
Set W1 = W
Set Y = 0.5 * V + W3
Output Y
Set W3 = 0.5 * V
Repeat indefinitely
EE3271 Jan03 Exam Solutions 9 11/11/04 BMGC
2π ∫ π (πδ (θ − Ω
−
0 ) + πδ (θ + Ω 0 ))W (e j ( Ω −θ ) ) dθ
(b) When Ω0 = 0.4π, the frequency of the sinusoid corresponds exactly with a frequency
sampling point. Hence the DFT graph will have a line of amplitude 10 at 0.4π, and a similar
line at –0.4π. All other DFT frequency-domain samples are zero since they coincide exactly
with zero crossings of the sincs function (see fig A below)
EE3271 Jan03 Exam Solutions 10 11/11/04 BMGC
When Ω0 = 0.55π, the frequency of the sinusoid lies mid way between two DFT frequency
sampling points. Therefore we see samples of the sincs function as illustrated in fig B below.
(c) Fig B shows X k for the 20-point DFT of a sinusoid of amplitude 10 and frequency 5.5π
radians/sample. Only the first 10 points need be plotted.
The frequency sampling points are 0, 0.1π, 0.2π, …, 0.9π, π
Note that 0.55π lies between two frequency sampling points.
Samples of the rectangular window's frequency response, with its side-lobes, are seen, and
frequency samples 5 and 6 are strongly affected as they are closest to the centre of the main lobe.
As there is no frequency sampling point at the peak of the main lobe, this is not seen.
The full height of the main lobe is not seen making the sinusoid look lower in magnitude than it
really is.
When the sampled sinusoid is of relative frequency 0.4 π , the effect of the rectangular window is
no longer seen because the sampling points happen to coincide exactly with the zero crossings of
the sincs function.(Fig A). Now the full height of the main lobe is seen (amplitude 10). This
always occurs when the frequency of the cosine wave coincides exactly with a frequency
sampling point.
The difference between Figs A and B is undesirable for estimating the amplitudes of signals,
especially sinusoids.
The solution is to use non-rectangular window functions such as the Hann window defined for
0 ≤ n < N as follows:
EE3271 Jan03 Exam Solutions 11 11/11/04 BMGC
0 .9
0 .8
0 .7
0 .6
0 .5
H a n n w in d o w
0 .4
0 .3
0 .2
0 .1
n
0
0 2 4 6 8 10 12 14 16 18 20
The DTFT of a 20th order Hann window compared with 20th order rectangular:-
25
W(exp(jw))
20 Rect
R2
15 R3
Hann
10
Radians/sample
5
0
-3.14 -2.355 -1.57 -0.785 0 0.785 1.57 2.355 3.14
-5