Review of LTI Systems:

Impulse Response, Di↵erence Equation,

System Function, and Frequency Response
Convolution Sum, System Function, and Frequency Response
LTI systems can be completely characterized in the time domain by its impulse
response, h[n]. Given and input, x[n], the output is found with the convolution
sum:
X1
y[n] = x[k]h[n k]
k= 1

The z-transform of the output of an LTI system is related to the z-transform of

the impulse response and input:

Y (z) = H(z)X(z)

Since the system function, H(z), is the z-transform of the impulse response, it also
completely characterizes an LTI system. The frequency response is H(z) evaluated
on the unit circle (assuming ROC includes z = ej! ):

H(ej! )
Linear Di↵erence Equations and System Function
N
X M
X
y[n] = ak y[n k] + bk x[n k]
k=1
| {z } |k=0 {z }
past outputs current and past inputs
N
X M
X
=) ak y[n k] = bk x[n k] (a0 = 1)
k=0 k=0
N
X M
X
k k
=) ak z Y (z) = bk z X(z)
k=0 k=0
M
X
k
bk z
Y (z)
=) H(z) = = k=0
X(z) XN
k
ak z
k=0
Poles and Zeros of the System

Factored form of the system function:

QM Q
Y (z) b0 k=0 1 ck z 1
b0 z N M M (z ck )
H(z) = = Q = QN k=0
X(z) a0 Nk=0 (1
1
dk z ) a 0 k=0 (z dk )

I roots of the numerator polynomial are zeros of the system

I roots of the denominator polynomial are poles of the system
I causal and stable system has poles inside the unit circle (with ROC outward)
Frequency Response
Since a causal an stable system has poles inside the unit circle, the system function
evaluated on the unit circle is the frequency response:
M
X
j!k
bk e
k=0
H(z) = H ej! = N
z=ej! X
j!k
ak e
k=0

I Real coefficients {ak , bk } =) implies a real impulse response h[n]

I Real h[n] =) hermitian frequency response H ej!
(real part even, imaginary part odd).
I Magnitude response: H ej! has even symmetry.
I Phase response: arg H ej! or \H ej! has odd symmetry
Example: Magnitude and Phase of Frequency Response
Frequency Response: Phase Response and Group Delay
Phase Angle

Relationship between the Fourier transform of the system input and output:

Y (ej! ) = H(ej! )X(ej! )

Since, in general, the Fourier transform is complex number at each frequency, we

often investigate the magnitude and phase:

|Y (ej! )| = |H(ej! )||X(ej! )|

\Y (ej! ) = \H(ej! ) + \X(ej! )

Note that the phase angle is not uniquely defined, since any integer of 2⇡ can be
added to the phase without a↵ecting the complex number.
Phase Angle: Wrapped and Unwrapped

The continuous phase curve is

known as unwrapped.

The principal, or wrapped, phase

of the of H(ej! ) is:

⇡ < \H(ej! )  ⇡

Notice that this phase response

is nonlinear.
Linear Phase Response
If the input to a LTI system is

x[n] = cos (!1 n) + cos (!2 n)

and the phase response is

\H(ej! ) = !nd
Then, ignoring the magnitude response, the output is:

cos (!1 n !1 nd ) + cos (!2 n !2 nd ) = cos (!1 (n nd )) + cos (!2 (n nd ))

where each component is delayed by nd samples.

If linear phase, then each frequency component will have identical delay.
Ideal Frequency Characteristics: Constant-Gain and Linear Phase
Ideal response for frequency selective filters,
constant-gain and linear phase:
(
j! Ce j!n0 , !1 < ! < !2
H(e ) =
0, otherwise

Signal at the filter output:

Y (ej! ) = H(ej! )X(ej! )

= CX(ej! )e j!n0
, !1 < ! < ! 2

Using the scaling and time-shifting properties of

the Fourier transform, the time-domain output is:

y[n] = Cx[n n0 ]
Group Delay
The group delay of a filter is the derivative of the phase with respect to frequency
and has the units of delay:
d
⌧ (!) = \(H(ej! )
d!
If the phase response is linear:

\H(ej! ) = !n0

Then the the group delay is constant and all frequencies have same time delay:

⌧ (!) = n0

Nonlinearity of the phase results in time dispersion.

Example: Attenuation and Group Delay of Frequency Selective Filter

x1 [n] = w[n] cos(0.2⇡n)

x2 [n] = w[n] cos(0.4⇡n ⇡/2)
x3 [n] = w[n] cos(0.8⇡n + ⇡/5)

Filter input:

x[n] = x3 [n]+x1 [n M 1]+x2 [n 2M 2]

with M = 60
Example: Attenuation and Group Delay of Frequency Selective Filter
Input Spectrum:

What frequency components

Filter Magnitude Response: exist in the output signal?
Example: Attenuation and Group Delay of Frequency Selective Filter
Filter Phase Response (principal phase, wrapped):

Filter Phase Response (continuous phase, unwrapped):

Example: Attenuation and Group Delay of Frequency Selective Filter
Input Spectrum:

What is the delay for each

Filter Group Delay: frequency component in the
output signal?
Example: Attenuation and Group Delay of Frequency Selective Filter
Filter Input:

Filter Output:
 More Insight into LTI Systems:
Rational System Functions, Inverse Systems
System Function: Algebraic and Factored Forms
P
M
k
bk z
Y (z)
H(z) = = k=0
X(z) PN
k
ak z
k=0

Q
M
1)
✓ ◆ (1 ck z
b0 k=1
H(z) =
a0 QN
1)
(1 dk z
k=1

I Each factor (1 ck z 1 ) in the numerator contributes a zero at z = ck

(and pole at z = 0)
I Each factor (1 dk z 1 ) in the denominator contributes a pole at z = dk
(and zero at z = 0)
Example: ROC, Stability, Causality

Given LTI system with di↵erence equation:

5
y[n] 2 y[n 1] + y[n 2] = x[n]

Has the system function:

Y (z) 1 1
H(z) = = 5 1 2
= 1 1 1)
X(z) 1 2z +z (1 2 z )(1 2z
I Three possible choices for region of convergence, but if the system is causal,
then ROC is outside of outermost pole (|z| > 2).
I However, for system to be stable, the ROC must include the unit circle
( 12 < |z| < 2).
Inverse Systems
For a given system function, H(z), the corresponding inverse system, Hi (z), has an
overall e↵ective system of unity when the two systems are cascaded:

G(z) = H(z)Hi (z) = 1

Which implies:
1 1
Hi (z) = Hi (ej! ) = g[n] = h[n] ⇤ hi [n] = [n]
H(z) H(ej! )

I Not all systems have inverses, for those that do the poles of Hi (z) are the
zeros of H(z) and vice versa
I The regions of convergence of H(z) and Hi (z) must overlap
I If H(z) is causal, its ROC is |z| > max |dk | and ROC for Hi (z) must overlap
Example: Inverse System

1 0.5z 1
H(z) = 1
ROC: |z| > 0.9
1 0.9z
Then, the inverse system is
1 0.9z 1
Hi (z) = 1
1 0.5z
Hi (z) as one pole at 0.5 so it has two possibilities of ROC. The only one that
overlaps with the |z| > 0.9 is |z| > 0.5 and the inverse z-transform gives the
impulse response of the inverse system:

hi [n] = (0.5)n u[n] 0.9(0.5)n 1

u[n 1]

Note that this inverse system is causal and stable.

Example: Another Inverse System

z 1 0.5
H(z) = ROC: |z| > 0.9
1 0.9z 1
Then, the inverse system is

1 0.9z 1 2 + 1.8z 1
Hi (z) = 1
=
z 0.5 1 2z 1
Since both possible ROC of the inverse system, |z| < 2 and |z| > 2, overlap the
ROC of the original system, both lead to valid inverse systems:

ROC: |z| < 2 hi1 = 2(2)n u[ n 1] 1.8(2)n 1

u[ n] stable and noncausal
n n 1
ROC: |z| > 2 hi1 = 2(2) u[n] + 1.8(2) u[n 1] unstable and causal
Causal and Stable Inverse Systems

If an LTI system is causal and stable, when will its corresponding inverse system
be causal and stable?
I If H(z) is causal with zeros at ck then its inverse will be causal if:

Hi (z) ROC is |z| > max |ck |

I If the inverse system is stable, then ROC for Hi (z) must include unit circle:

max |ck | < 1

I Therefore, a LTI causal and stable system has causal and stable inverse if
both the poles and the zeros of H(z) are inside the unit circle. These systems
are called minimum-phase systems.
More Insight into LTI Systems:
Finite Impulse Response (FIR)
Finite Impulse Response (FIR) Systems
H(z) has no poles except at z = 0 and is a polynomial in z 1 (since N = 0)
P
M
k
bk z M
X
Y (z)
H(z) = = k=0 =) bk z k
X(z) PN
k
ak z k=0
k=0

So the impulse response is the b coefficients:

M
(
X bn , 0nM
h[n] = bk [n k] =
k=0
0, otherwise
The di↵erence equation is the same as the convolution sum:
M
X
y[n] = bk n[n k]
k=0
Example: FIR System Pole-Zero Plot, Di↵erence Equation
Consider truncation of the impulse response from the IIR system:
1
G(z) = 1
|z| > |a|
1
( az
an , onM
h[n] =
0, otherwise

The system function of this FIR system is

M
X M
X M 1
1 (az 1 )M +1 z z M +1 aM +1
H(z) = an z n
= (az 1 n
) = 1
=
1 az z 1 (z a)
n=0 n=0
z M +1 aM +1
= poles: z = 0 and z = a
z M (z a)
Example: FIR System Pole-Zero Plot, Di↵erence Equation (Continued)

z M +1 aM +1
H(z) =
z M (z a)
Finding zeros: z M +1 = aM +1

zeros: zk = aej2⇡k/(M +1) , k = 0,1,...,M

Pole at z = a cancelled by z0 = a

Two equivalent di↵erence equations for this system:

M
X
y[n] = ak x[n k]
k=0
1 (az 1 )M +1
H(z) = =) y[n] ay[n 1] = x[n] aM +1 x[n M 1]
1 az 1
Frequency Response From Rational System Functions:
Impact of Pole and Zero Locations
Frequency Response, Phase Response, Magnitude Response
The frequency response is the system function, H(z), evaluated on the unit circle:

P
M
j!k
Q
M
j!)
bk e ✓ ◆ (1 ck e
k=0 b0 k=1
H(ej! ) = H(ej! ) =
PN
j!k
a0 QN
j! )
ak e (1 dk e
k=0 k=1

Magnitude and the magnitude-squared function:

Q
M Q
M
|1 ck e j! |
✓ ◆2 (1 ck e j! )(1 c⇤k ej! )
b0 k=1 b0 k=1
|H(ej! )| = |H(ej! )|2 =
a0 QN a0 QN
|1 dk e j! | (1 dk e j! )(1 d⇤k ej! )
k=1 k=1
Gain in dB and Phase Response
Gain in dB:
XM M
X
b0 j! j!
20 log10 + 20 log10 |1 ck e | 20 log10 |1 dk e |
a0
k=1 k=1

Phase response:

X ⇥ M N
X
b0 ⇤ ⇥ ⇤
\H(e ) = \
j!
+ \ 1 ck e j!
\ 1 dk e j!
a0
k=1 k=1

I Magnitude in dB and phase are a sum of the contributions from each of the
poles and zeros of the system function
I Understanding the impact of a single pole/zero location on frequency response
provides insight into higher-order systems
Frequency Response: 1st-Order Systems
Single factor of H(ej! ): (1 rej✓ e j! ) the pole or zero has radius r and angle ✓
Magnitude-squared of factor:

|1 rej✓ e j! 2
| = (1 rej✓ e j!
)(1 re j✓ j!
e ) = 1 + r2 2r cos(! ✓)

Contribution to gain in dB (where + for a zero and for a pole):

⇥ ⇤
±20 log10 |1 rej✓ e j! | = ±10 log10 1 + r2 2r cos(! ✓)

Contribution to phase:
h i 
j✓ j! r sin(! ✓)
±\ 1 re e = ± arctan
1 r cos(! ✓)

I Note how magnitude in dB and phase change based on pole/zero: r and ✓

Example: Single Zero at Di↵erent Angles

Single Zero at r = 0.9 and ✓ = 0, ⇡/2, or ⇡

Log magnitude response (minimized at ! = ✓):

⇥ ⇤
+10 log10 1 + r2 2r cos(! ✓)

Phase Response (zero at ! = ✓):


r sin(! ✓)
+ arctan
1 r cos(! ✓)
Vector Diagram Perspective of Frequency Response
z rej✓
H(z) = (1 rej✓ z 1
)= r<1
z
Geometric interpretation:

v~1 = ej! location on unit circle

j✓
v~2 = re zero
v~3 = v~1 v~2 = ej! rej✓

Magnitude of H(z):

ej! rej✓ |v~3 |

|1 rej✓ e j!
| = |e j!
(ej! rej✓ )| = =
ej! |v~1 |

Phase: \(1 rej✓ e j! ) = \v~3 \v~1 = 3 !

Example: Single Zero at Di↵erent Radii with ✓ = ⇡
Magnitude of H(z) |1 rej✓ e j! | = |v~3 |
Phase: 3 !
Example: Second-Order IIR System
|v3 ||v3 |
|H(ej! )| =
Complex-conjugate pair or poles: |v1 ||v2 |
r = 0.9, ✓ = ⇡/4
Magnitude-Phase Relationships and All-Pass Systems
Exploring Possible System Functions
If given the square of the magnitude of the system frequency response, what are
possible choices of the system function?
|H(ej! )|2 = H(ej! )H ⇤ (ej! ) = H(z)H ⇤ (1/z ⇤ )|z=ej!
=) h[n] ⇤ h⇤ [ n]
Therefore, |H(ej! )|2 is the evaluation of this z-transform on the unit circle:
Q
M
✓ ◆2 (1 ck z 1 )(1 c⇤k z)
b0 k=1
C(z) = H(z)H ⇤ (1/z ⇤ ) =
a0 QN
(1 dk z 1 )(1 d⇤k z)
k=1

I For each pole dk of H(z) there is a pole of C(z) at (d⇤k ) 1

I For each zero ck of H(z) there is a zero of C(z) at (c⇤k ) 1

Di↵erent Systems With Same Magnitude Response
Figure (a): H1 (z) Figure (b): H2 (z)
Figure (c): C(z)=H1 (z)H1⇤ (1/z ⇤ )=H2 (z)H2⇤ (1/z ⇤ )
I H1 (z) and H2 (z) are di↵erent systems with same
magnitude response but di↵erent phase response
I Poles and zeros of C(z) occur in conjugate
reciprocal pairs (one inside / one outside unit
circle, or both on unit circle)
I Determining H(z) from C(z): Casual system:
poles can be identified from C(z) (inside unit
circle) but zeros can’t be identified
I Unknown phase but finite choices due to possible
locations of zeros
All-Pass Factors

Assume H(z) can be factored:

z 1 a⇤
H(z) = H1 (z) 1
1 az
Factors in this form are all-pass factors (unity magnitude) and they cancel in C(z):

C(z) = H(z)H ⇤ (1/z ⇤ ) = H1 (z)H1⇤ (1/z ⇤ )

Given C(z), choices for H(z) can be cascaded with an arbitrary number of all-pass
factors if the total number of poles/zeros of H(z) were unknown.
All-Pass Systems: Constant Gain or Attenuation
z 1 a⇤
Hap (z) = 1
|a| < 1
1 az
e j! a⇤ 1 a⇤ ej!
Hap (ej! ) = j!
=e j!
1 ae 1 ae j!
Constant Magnitude (independent of frequency):

|Hap (ej!) | = 1 (log magnitude in dB is zero) :

General form of system function (dk are real poles and ek are complex poles)
Mr
Y M
z 1 dk Yc (z 1 e⇤k )(z 1 ek )
Hap (z) = A ⇤
1 dk z 1 (1 ek z 1 )(1 ek z 1 )
k=1 k=1

For stable/causal systems: |dk | < 1 and |ek | < 1; M = N = 2Mc + Mr poles/zeros.
Typical Pole-Zero Plot for All-Pass System
Mr
Y M
z 1 dk Yc (z 1 e⇤k )(z 1 ek )
Hap (z) = A ⇤
1 dk z 1 (1 ek z 1 )(1 ek z 1 )
k=1 k=1

For stable/causal systems: |dk | < 1 and |ek | < 1; M = N = 2Mc + Mr poles/zeros.

Mr = 2 and Mc = 1
Example 1st and 2nd Order All-Pass Systems

Pole at 0.9 (solid), Pole at -0.9 (dashed) Poles at z = 0.9e±j⇡/4

I phase (unwrapped) is always nonpositive for 0 < ! < ⇡

Minimum-Phase Systems
Minimum-Phase Systems
Introduction to Minimum Phase Systems
I Inverse Systems
I h[n] ⇤ hinv [n] = [n]
I Cascade of system and its inverse is the identity system: H(z)Hinv (z) = 1
I Restrict inverse system, 1/H(z), to be stable and causal
I Therefore, all zeros (not just poles) must be inside the unit circle
I A causal and stable LTI system with a causal and stable inverse is called a
minimum-phase system

Given a magnitude-squared function of a minimum-phase system:

C(z) = H(z)H ⇤ (1/z ⇤ )

then, H(z) is uniquely determined by choosing all poles/zeros inside unit circle.
Minimum-Phase and All-Pass Decomposition

Assume H(z) has one zero at z = 1/a⇤ outside the unit circle (|a| < 1), and all
other poles/zeros are inside the unit circle:

H(z) = H1 (z) z 1
a⇤

Therefore, H1 (z) is minimum-phase. We can continue to decompose with an

all-pass factor:
z 1 a⇤
H(z) = H1 (z) 1 az 1
| {z } (1 az 1 )
minimum-phase | {z }
all-pass

Repeat this for every zero outside the unit circle for any arbitrary rational system
function.
Minimum-Phase and All-Pass Decomposition (Continued)

Any rational system can be factored into a minimum-phase system and an all-pass
system:
H(z) = Hmin (z)Hap (z)

I Hmin (z) contains all poles/zeros of H(z) that are inside unit circle
I Hmin (z) also contains conjugate reciprocals of zeros of H(z) outside unit circle
I Hap (z) contains all zeros of H(z) that are outside unit circle
I Hap (z) contains poles to cancel conjugate reciprocals zeros in Hmin (z)
Example: Minimum-Phase/All-Pass Decomposition
Find the minimum-phase all-pass decomposition of this stable and causal system:

1 + 3z 1
1
H(z) = pole at z = 2 and zero at z = 3
1 + 12 z 1

I Hmin (z) contains all poles/zeros of H(z) that are inside unit circle
I Hmin (z) also contains conjugate reciprocals of zeros of H(z) outside unit circle
I Hap (z) contains all zeros of H(z) that are outside unit circle
I Hap (z) contains all poles to cancel conjugate reciprocals zeros in Hmin (z)
! ! !
1
1 + 3z 1 z 1+ 3 1 + 13 z 1 z 1+ 1
3
H(z) = =3 =3
1 + 12 z 1 1 + 12 z 1 1 + 12 z 1 1 + 13 z 1
| {z }| {z }
Hmin (z) Hap (z)
Application: Frequency Response Compensation (Inverse Filters)

For a nonminimum phase system, we can form minimum phase system, Hd,min (z):
Hd (z) = Hd,min Hap (z)
Then the compensating filter is stable and causal:
1 Hd (z)
Hc (z) = where Hd,min (z) =
Hd,min (z) Hap (z)
Overall system function:
G(z) = Hd (z)Hc (z) = Hap (z)
Property of Minimum-Phase Systems
Using minimum-phase and all-pass decomposition:

H(z) = Hmin (z)Hap (z) =) \H(z) = \Hmin (z) + \Hap (z)

For all systems that have the same magnitude response, the system with all poles
and zeros inside the unit circle has:

1. minimum phase-lag (phase-lag is negative of the phase function)

I all-pass systems always have negative unwrapped phase curve
I therefore, the all-pass system increases the negative of the phase (phase-lag)
I leaving Hmin (z) to have minimum phase-lag
2. minimum group delay
I all-pass systems always have a positive group delay
I therefore, the all-pass system increases group delay
I leaving Hmin (z) to have minimum group delay
Systems With Linear Phase
Ideal Delay System: Linear Phase and Constant Magnitude
Consider an LTI system with frequency response, an ideal delay system:
H(ej! ) = e j!↵
|!| < ⇡
where ↵ is a real number but not necessarily an integer. For this system:
|H(ej! )| = 1 constant magnitude response
\H(ej! ) = !↵ linear phase
⌧gd = ↵ constant group delay
The inverse Fourier transform provides the impulse response:
sin (⇡(n ↵))
h[n] = = sinc (n ↵) 1<n<1
⇡(n ↵)
When, ↵ = nd , and nd is an integer, then
h[n] = [n nd ] and y[n] = x[n] ⇤ [n nd ] = x[n nd ]
Ideal Delay System: Non-Integer Delay
What if the delay, ↵, is not an integer? Interpret as a continuous time system:

j⌦↵T
hc (t) = (t ↵T ) and Hc (j⌦) = e

Overall response: H(ej! ) = e j!↵ |!| < ⇡

Output: y[n] = xc (nT ↵T ) Time shift of ↵ samples (even non-integer).
Linear Phase With Non-constant Magnitude Response
A linear phase system has the frequency response

H(ej! ) = A(ej! )e j!↵

|!| < ⇡

where A(ej! ) is purely real and is called the zero phase response.
I With real-valued h[n], then |H(ej! )| = |A(ej! )| are even
I Since A is real, A is even and the inverse Fourier transform is real and even
I The impulse response h[n] is real and even and time shifted by ↵
I If ↵ is an integer: h[n] is symmetric about n = ↵
I If ↵ is an integer + 12 : h[n] is symmetric about n = ↵ + 12
I Both cases: if 2↵ is an integer then h[n] has symmetry: h[2↵ n] = h[n]

Linear phase and causal: must be FIR with M = 2↵ and

h[0] = h[M ], h[1] = h[M 1], etc.
Example: Ideal Lowpass Filter With Linear Phase
Linear phase lowpass filter with cuto↵ !c :

sin (!c (n nd ))
h[n] =
⇡(n nd )

I Figure (a): !c = 0.4⇡, ↵ = nd = 5

h[n] symmetric about n = 5
I Figure (b):
!c = 0.4⇡, ↵ = nd = 4.5
h[n] symmetric about ↵ = nd = 4.5
I Figure (c):
!c = 0.4⇡, ↵ = nd = 4.3
h[n] not symmetric, still linear phase
Generalized Linear Phase
Generalized Linear Phase System

Broader class to include more linear phase systems (including odd symmetry):

H(ej! ) = A(ej! )e j↵!+j

I A(ej! ) is real (possible bipolar) function of !

I ↵ and are constants
I Phase is the equation of a straight line ( ↵! + )
I Constant group delay, ⌧gd = ↵
Necessary Conditions For Constant Group Delay
For generalized linear phase systems

H(ej! ) = A(ej! )e j↵!+j

Leads to necessary conditions on real-valued h[n], and constants ↵, and :

1
X
h[n] sin [!(n ↵) + ] = 0, for all !
n= 1

This doesn’t tell us how to find a linear phase system but we are interested in
particular cases:

2↵ = M = even or odd integer, and = 0, ⇡ or = ⇡2 , 3⇡

2
Causal Generalized Linear Phase Systems

1
X
h[n] sin [!(n ↵) + ] = 0, for all !
n=0

Causality and symmetry conditions imply:

h[n] = 0, n<0 and n>M

Symmetric or antisymmetric impulse response (even, odd):

( (
h[M n], 0  n  M h[M n], 0nM
h[n] = h[n] =
0, otherwise 0, otherwise
H(ej! ) = Ae (ej! )e j!M/2
H(ej! ) = jAo (ej! )e j!M/2
= Ao (ej! )(ej!M/2+j⇡/2 )
⇡ 3⇡
where = 0 or =⇡ where = 2 or = 2
Four Types of FIR Generalized Linear Phase Systems

Type of symmetry and whether M is an even or odd integer lead to di↵erent

expressions for the frequency response of FIR linear phase systems:

Type I II III IV
M even odd even odd
h symmetry even even odd odd
A symmetry even even odd odd
A e±j⇡ 6= 0 0 0 6= 0
↵ integer half-integer integer half-integer
⇡ 3⇡ ⇡ 3⇡
0 or ⇡ 0 or ⇡ 2 or 2 2 or 2

Note: Frequency response constrained to zero at ! = ⇡ for types II and III.

Examples of FIR Linear Phase Impulse Response
Locations of Zeros for FIR Linear Phase Systems
Symmetric cases of the impulse response, h[n], puts constraints on the zeros of FIR
system. Here are some examples: