SEL4223 Digital Signal Processing

Phase System

Musa Mohd Mokji

 Introduction

the signal cos( ) where

• In time-domain, phase in a signal can be seen as a time delay from

cos + = cos +

− /2
sin + = cos − /2 + = cos +

• Thus, in LTI system, phase on the impulse response ℎ will delay

the input signal at the output
 Introduction
• Prove:

• = .

• = + +

• = ℯ ∠
. ℯ ∠

• =ℯ ∠ !∠
.
 Introduction
• Equations above shows that:

– frequency of the output signal is obtain by multiplying

magnitude response of input signal and the impulse response

– phase of the output signal is the sum of the input signal phase
and the impulse response phase.

• Thus, if the impulse response has a phase, the input signal will be
delayed at the output.
 Phase Distortion
• Signal can be understood as the combination of sinusoids of
different frequencies. If the time delay of these frequencies is not
consistent, then it can be said that the output signal has a phase
distortion.
 Phase Quantification
• Because phase is closely related to time delay, two measures are
used to evaluate the time delay as follows:

∠
"ℎ#$% &%'# ⟹ )* =

&
+,-./ &%'# ⟹ )0 = − ∠
&

• Both of the Phase delay and the Group delay are referring to

is, Phase delay is referring to delay compare to cos

amount of delay in time-domain. The difference between the two

Group delay is referring to both cos and − cos

while

delay is also identified as the slope of the phase response ∠

. Group
 Phase System

3 types of phase response

o Zero phase system
o Linear phase system
o Non-linear phase system

From the 3 phase response types, zero phase and linear phase
do not have phase distortion while the non-linear phase has the
phase distortion.
 Zero Phase System 0
• ℎ[ ] must be symmetry (even signal) or anti-symmetry (odd signal)
– Symmetry (even): ℎ =ℎ −
– Anti-symmetry (odd): ℎ = −ℎ −
 Zero Phase System (cont.)
• ∠ for even and odd signal are constant values as follow:

0 89 ≥0
∠ 3435 = 6 89 <0
− 89 − <0

− /2 89 >0
∠ =>
<==
/2 89 <0
 Zero Phase System (cont.)
• As the ∠ is a constant value, differentiation to it results a zero
value, means that all frequencies convolve with the zero phase
system will have no time delay at the output.
• In other words, no phase distortion.

ℎ ≠ 0 for < 0. Thus, the system is not practical.

• The problem with zero phase system is that it is not causal where
 Example 1
• Find and plot = ∗ℎ where
B
ℎ = C D +1 +D +D − 1 and =
0.5 sin 0.25 + sin 0.02 . −. − 150

1.5
x[n]

-1.5

02 10 18 26 34 42 50 150
n
 Example 1 (cont.)

1.5
y[n]

-1.5

02 10 18 26 34 42 50 150
n

• It can be seen from Figure above that no delay occurs in the output
signal.
 Linear Phase System 0 =

• ℎ is a delayed symmetry or anti-symmetry signal

 Linear Phase System (cont.)
• To make sure that ℎ[ ] is causal, the signal is delayed so that it will
have nonzero values starts at = 0. Thus, the causality problem in
zero phase system is solved.
• For this system,

∠ = = and )* = )0 = =

• Because = is a constant, there will be no phase distortion.

• This system is only possible for FIR system. It is not possible to have
a delayed symmetry or anti-symmetry signal for IIR system.
Linear phase system variation

Type Signal Length G + 1

I ℎ = ℎ[G − ] Odd

II ℎ = ℎ[G − ] even

III ℎ = −ℎ[G − ] odd

IV ℎ = −ℎ[G − ] even
 Example 2
• Find ∠ for ℎ =D +D −4 .

Solution:
• ℎ
ℎ< = D + 2 + D − 2 where ℎ = ℎ< − 2 .
can be seen as a delayed symmetry signal of

• < = ℎ< 0 + 2 ∑B5JB ℎ cos = 2 cos .

• From the time shifting property,

• = ℯK L
< = 2ℯ K L
cos
• = 2 cos cos 2 − M2 cos sin 2
 Example 2 (cont.)
• Thus,
L PQR L
• ∠ = tanKB − L STP L = −2

• This shows that the system is linear phase system with

∠ = =

• Where = = −2
 Example 3
• Repeat Example 1 but with impulse response below:
• ℎ = [0.0036, 0, −0.0123, 0, 0.0344, 0, −0.0860, 0,
0.3111, 0.5, 0.3111, 0, −0.0860, 0, 0.0344, 0,
−0.0123, 0, 0.0036]

• In this example, ℎ[ ] is a delayed symmetry signal with 9 samples

delay. Basically ℎ[ ] will remove all frequencies greater than 0.5 .
Thus, both frequencies in will be preserved at the output.
• Because ℎ
∠ = −9
is delayed 9 samples from its symmetry version,
 Example 3 (cont.)

1.5
y[n]

-1.5

0 11 19 27 35 43 51 59
n

frequencies, = 0.25 and = 0.02 were both delayed by 9

• By looking at peaks of the output signal, it can be seen that the two

samples.
 Non-Linear Phase System
• ℎ[ ] can be both FIR and IIR
• ℎ[ ] can also be both stable and causal
• Phase delay and Group delay vary throughout the entire frequency
band.
• Other than zero-phase and linear phase systems, the system is
considered as non-linear phase system.
• The problem is, it suffers phase distortion.
 Example 4
• Repeat Example 3 but with 4th order IIR Butterworth filter as below:
• =
]
B!ℯ Z[\
^.__`aKBB.Ba_Bℯ Z[\ !_.bL`Cℯ Z[c\ BL.CC`KBB.Ba_Bℯ Z[\ !L.adaBℯ Z[c\

• From the , difference equation of the system is

• = 0.0085( +4 −1 +6 −2 +4 −3
• + − 4 + 244.89 − 1 − 221.75 −2
• +94.8 − 3 − 16.03 −4

• Similar to filter in Example 3, this filter will also pass through both
frequencies at = 0.02 and = 0.25 .
 Example 4 (cont.)
• Phase response ∠ of the filter is shown in next figure where
∠ 0.02 = −0.0672 and ∠ 0.25 = 0.928. Other than that,
the figure obviously shows a non-linear plot as the slope of the plot
varies.
• Phase delay at the two input frequencies can be computed as:

)* 0.25
f.fb`Lg f.^Lag
• )* 0.02 = − f.fLg = −3.36, = =
f.L_g
3.712

• This means that signal with frequency 0.02 is shift 3.36 samples to
the right while signal with frequency 0.25 is shift 3.712 samples to
the left. Because the delays are different between the two
frequencies, thus phase distortion occurs.
 Example 4 (cont.)
0.928
∠H(ω) x π

-0.0672

-1
-1 0.02 0.25 1
ω (π rad)

• Figure above shows the results where peaks for frequency 0.25
positions in the input signal have moved significantly while delay for
frequency 0.02 is unnoticeable.
 Example 4 (cont.)
• Delay of the frequency 0.25 is obvious because it only need 8
samples to complete one period while the delay for the frequency is
almost half of it which is 3.712 samples.
• For frequency 0.02 , although the delay is almost the same with
delay of the frequency 0.25 which is 3.36 samples, this is still too
small compare to the 100 samples needed to complete its one period.
1.5
y[n]

-1.5
02 10 18 26 34 42 50 150
n
 Quiz
Determine whether the signal has zero phase, linear phase or non-
linear phase.
1. h n = u n
2. h n = u n + u −n
3. h n = u n + u −n − 1
4. h n = δ n + 2δ n
5. h n = δ n − 2δ n − 1 + 2δ n + 1
6. h n = a R , a < 1
7. hn = a R u n ,a < 1
8. yn = y n−1 +x n
9. hn = u n − u −n
10. yn = 2y n − 1 + x n − 2x n − 1
Plot the pole-zero plot, magnitude response and frequency response
of these signals
 Sampling Revisited
• A process of converting analog signal to discrete signal
• ) = m no , ) = no
Lgp
• In frequency domain, =2 9= .
pq

• At = , where it is the end frequency for discrete signal, r =

pq
sL. Thus,

Frequency components preserved after the sampling

are the frequencies less than tusL
Sampling Revisited (cont.)
 Example 5
• ) = cos 2 1000) for 0 ≤ ) < 10w$

• If ro = 2000Hz
LgBfff5 Bfxo
• = cos
Lfff
9-, 0 ≤ < f._xo

• = cos = cos m 9-, 0 ≤ < 20

• m =

• Thus, when ro = 2000Hz, the signal frequency of Ω = 2 1000 is

mapped at =
 Example 5 (cont.)

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

x[n]
x(t)

0 0

-0.2 -0.2

-0.4 -0.4

-0.6 -0.6

-0.8 -0.8

-1 -1
0 1 2 3 4 5 6 7 8 9 10 0 2 4 6 8 10 12 14 16 18 20
t (ms) n

Continuous signal Discrete signal sampled at

ro = 1000 z
 Example 5 (cont.)
20

18

16

14

12
|H(ω)|

10

0
-π or 0 π or
-Fs/2 ω (π rad) Fs/2

Magnitude response - frequency { = 2 1000 ,#&$ KB is mapped at

= ,#&
 Example 5 (cont.)
• When ro = 6000Hz, the signal frequency Ω = 2 1000 is mapped
B
at = C

30

25

20
|H(ω)|

15

10

0
-pi -pi/3 0 pi/3 pi
ω (rad) (1000Hz) (3000Hz)
 Example 6
• ) = cos 2 1000) + sin 2 4000) for 0 ≤ ) < 10w$

0.366
x(t)

-1

-1.366

-2
0 0.33 0.67 1 1.33 1.67 2 2.33 2.67 3 3.33 3.67 4 4.33 4.67 5 5.33 5.67 6 6.33 6.67 7 7.33 7.67 8 8.33 8.67 9 9.33 9.67 1
t (ms)
 Example 6 (cont.)

For ro = 3000Hz,
Bfxo Bfxo
Number of sample in [ ] is | = = f.CCCCxo = 30
}q

0.366

0
x[n]

-1.366

0 2 4 6 8 10 12 14 16 18 20 22 24 26 28
n
 Example 6 (cont.)
• In equation, the discrete signal is written as:
LgBfff5 Lgdfff5
• = cos Cfff
+ sin Cfff
9-, 0 ≤ < 10w$ ×
3000 z

Lg5 ag5
• [ ] = cos C
+ sin
C

Lg5 L
• = cos + sin 2C
C

Lg5 L
• = cos + sin 9-, 0 ≤ < 30
C C
 Example 6 (cont.)

• It is shown that when ro = 3000Hz, both frequency of the 1000Hz

L
and the 4000Hz of the continuous signal are mapped to = C .

• In other words, although there are two frequencies exist in the

continuous signal, only one frequency appears in its discrete form.
• The following figures show the frequency response of the discrete
signal.
 Example 6 (cont.)
25

20

15
|H(ω)|

10

0
-pi -2pi/3 0 2pi/3 pi
ω (rad)

pi

pi/4
∠ H(ω)

0
-pi/4

-pi
-pi -2pi/3 0 2pi/3 pi
ω (rad)
 Example 6 (cont.)
• Thus, when [ ] is converted back to its continuous signal, the
frequency component of 4000Hz is missing as shown in equation
and figure below. This phenomenon is called ‘aliasing’.
• ) = cos 2 1000) + sin 2 1000)
2

1
x(t)

-1

-2
0 1 2 3 4 5 6 7 8 9 10
t (ms)
 Example 6 (cont.)
• To obtain a good sampling output, r$ ≥ 2r| , where r| is the
maximum frequency in the original continuous signal. This is called the
Nyquist Theorem
• If we now choose r$ = 10000Hz ≥ 2r| ≥ 8000Hz ,
2 1000 2 4000
• = cos + sin 9-, 0 ≤ < 10w$ × 10• z
10000 10000

4
• = cos 5
+ sin 5
9-, 0 ≤ < 100

• Now, frequency of 1kHz is mapped to = 5 and frequency of 4kHz is

4
mapped to = 5
. This
shows that both of the frequencies in the
continuous-time signal are preserved where no aliasing occur.
 Example 6 (cont.)

• The following figures show the new discrete time domain signal and
its frequency spectrum respectively.

1
x[n]

-1

-2
0 10 20 30 40 50 60 70 80 90 100
n
 Example 6 (cont.)
60

40
|H(ω)|

20

0
-pi -4pi/5 -pi/5 0 pi/5 4pi/5 pi
.
(1000Hz) (4000Hz) (10000Hz)
ω (rad)

pi

pi/2
∠ H(ω)

-pi/2

-pi
-pi -4pi/5 -pi/5 0 pi/5 4pi/5 pi
ω (rad)
 References
1) John G. Proakis, Dimitris K Manolakis, “Digital Signal Processing:
Principle, Algorithm and Applications”, Prentice-Hall, 4th edition
(2006).
2) Sanjit K. Mitra, “Digital Signal Processing-A Computer Based
Approach”, McGraw-Hill Companies, 3rd edition (2005).
3) Alan V. Oppenheim, Ronald W. Schafer, “Discrete-Time Signal
Processing”, Prentice-Hall, 3rd edition (2009).