You are on page 1of 5

Summer 2008 Signals & Systems S.F.

Hsieh

II-A. DT LTI System Analysis: Impulse Response & Convolution

1 DT LTI system: Convolution Sum


1. Impulse:
(
1, n = 0
δ[n] =
0, n =
6 0

[Sifting property:] Any sequence x[n] can be represented as a sum of scaled/shifted impulse
sequences:
X
x[n] = x[k]δ[n − k] : convolution of x[n] and δ[n]
k
= · · · + x[−2]δ[n + 2] + x[−1]δ[n + 1] + x[0]δ[n] + x[1]δ[n − 1] + · · ·
: any sequence can be decomposed as a sum of scaled and shifted impulses.

(a) Scale (by x[k]) and shift (from δ[n] to δ[n − k]) imply linearity and time-invariance, respec-
tively.
(b) Envision x[n] : {..., x[−1], x[0], x[1], ...} as a signal vector, and
δ[n − k] : {..., 0, δ[n − k], 0, ...} is itself a signal vector, too.
(c) The sifting property is used to prove the convolution thm of an LTI system.
2. Impulse response is the system’s response(output) to an impulse input:

h[n] ≡ y[n]|when input x[n]=δ[n] δ[n] → LTI → h[n]

[Ex] y[n] = 0.9y[n − 1] + x[n]. The impulse response h[n] can be computed recursively from y[n]
by setting x[n] = δ[n]. If the input is an impulse x[n] = δ[n] and y[−1] = 0(zero-state), then
y[0] = x[0] = 1, y[1] = 0.9y[0] + 0 = 0.9, y[2] = 0.9y[1] = 0.92 , y[3] = 0.93 , . . .. Thus the impulse
response is h[n] = 0.9n u[n].
3. [Convolution Sum Thm] The output of any LTI system is the discrete convolution of the
input x[n] with the system’s impulse response h[n]:

X
x[n] → LTI: h[n] → y[n] = x[n] ∗ h[n] = x[m]h[n − m]
m=−∞

(Pf): [ convolution = con (put together) + volvere (to roll, rotate) ]

y[n] = LT I{x[n]} · · · · · · sifting


 

 X 
= LT I x[k]δ[n − k] · · · · · · linearity
 
k=−∞

X
= x[k] · LT I {δ[n − k]} · · · · · · time-invariancy
k=−∞
X∞
= x[k] · h[n − k]
k=−∞
≡ x[n] ∗ h[n]

II-A - 1
2 Discrete Convolution(Convolution Sum)
P
1. y[n] = ∞ k=−∞ x[k]h[n − k]: the output is the superposition of a shifted and scaled impulse
responses.
[Ex] If h[n] = [3, 2, 1], and x[n] = [1, 2, 0, −1], then y[n] = 1 · h[n] + 2 · h[n − 1] − h[n − 3] =
[3, 2, 1, 0, 0, 0] + [0, 6, 4, 2, 0, 0] − [0, 0, 0, 3, 2, 1] = [3, 8, 5, −1, −2, −1].
(MATLAB) conv([3,2,1],[1,2,0,-1])

2. Graphical procedure(evaluation of y[n] at some specific n = n0 ): (Lathi Fig 3.17, p 205)

(a) Fix x[k]

(b) Flip h[k] to get h[−k]

(c) Right shift h[−k] by different n to get h[−(k − n)] = h[n − k]

(d) Multiply x[k]h[n − k]

P
(e) Sum up k and get y[n]

(f) Change n, redo (c)-(e)

3. Examples

(a) Convolution of two one-sided infinite sequences: x[n] = αn u[n] and h[n] = u[n].

X
y[n] ≡ αk u[k] · u[n − k], note u[k] = 0 if k < 0
k=−∞
X∞
k
= α · u[n − k], note that u[n − k] = 0, if n < k
k=0
(
0, n<0
= Pn 1−αn+1
k=0 αk = 1−α , n≥0
1− αn+1
= · u[n]
1−α
(b) Convolution of two finite-length sequences: x[n] = 1 for 0 ≤ n ≤ 4 and h[n] = αn for
0 ≤ n ≤ 6.
i. For n < 0, y[n] = 0, because x[k] and h[n − k] are disjoint.
Pn n−k = 1−αn+1 .
ii. For 0 ≤ n ≤ 4, y[n] = k=0 α 1−α
P4 n−k
iii. For 4 < n ≤ 6, y[n] = k=0 α = · · ·.
P4 n−k
iv. For 6 < n ≤ 10, y[n] = k=n−6 α = · · ·.
v. For n > 10, y[n] = 0.
0
( P
0
2k = 2,

n
X
k
X
k n≥0
(c) 2 u[−n] ∗ u[n] = 2 u[−k]u[n − k] = 2 u[n − k]. = Pnk=−∞ k n+1
k=−∞ k=−∞ k=−∞ 2 =2 n < 0.

II-A - 2
4. Sliding-tape method for discrete convolution:

multiplication(sum-of-indices) ··· x[−2] x[−1] x[0] x[1] x[2] ···


.. .. .. .. .. .. .. ..
. . . . . . . .
h[−2] ··· (-4) (-3) (-2) (-1) (0) ···
h[−1] ··· (-3) (-2) (-1) (0) (1) ···
h[0] ··· (-2) (-1) (-0) (1) (2) ···
h[1] ··· (-1) (0) (1) (2) (3) ···
h[2] ··· (0) (1) (2) (3) (4) ···
.. .. .. .. .. .. .. ..
. . . . . . . .
x[n] 1 2 3 h[n] 1 2 1
[Ex] and
n -1 0 1 n 0 1 2

x[−1] x[0] x[1]


1 2 3
y[n] 1 4 8 8 3
h[0] 1 1 2 3 Thus,
n -1 0 1 2 3
h[1] 2 2 4 6
h[2] 1 1 2 3

5. Polynomial Method(similar to Z-transform): Express x[n] and h[n] as



X
X(z) ≡ x[n]z n = · · · x[−2]z −2 + x[−1]z −1 + x[0]z 0 + x[1]z + x[2]z 2 + · · ·
n=−∞
X∞
H(z) ≡ h[n]z n = · · · h[−2]z −2 + h[−1]z −1 + h[0]z 0 + h[1]z + h[2]z 2 + · · ·
n=−∞
X∞
Y (z) ≡ y[n]z n = · · · y[−2]z −2 + y[−1]z −1 + y[0]z 0 + y[1]z + y[2]z 2 + · · ·
n=−∞

If y[n] = x[n] ∗ h[n], then Y (z) = X(z) · H(z), and y[n] can be found from the coefficients of the
polynomial Y (z) = X(z) · H(z).
[Previous Ex] X(z) = 1 · z −1 + 2 · z 0 + 3 · z 1 and H(z) = 1 · z 0 + 2 · z 1 + 1 · z 2 , multiply together,
Y (z) = X(z)H(z) = 1 · z −1 + (2 + 2) · z 0 + (3 + 4 + 1) · z 1 + (6 + 2) · z 2 + 3 · z 3 , thus y[n] = 1, 4, 8, 8, 3
for n = −1, 0, 1, 2, 3, respectively.
6. Vector-wize: y = Hx
    
. . . . . . . . . . .

 . 


 . . . . . . . . . 
 . 

    
 .   . . . . . . . . .  . 
    

 y[0] 


 . . . h[0] h[−1] h[−2] . . . 
 x[0] 

 y[1]  =  . . . h[1] h[0] h[−1] . . .  x[1] 
    

 y[2] 


 . . . h[2] h[1] h[0] . . . 
 x[2] 

    
 .   . . . . . . . . .  . 
    
 .   . . . . . . . . .  . 
. . . . . . . . . . .

II-A - 3
3 Properties of LTI Systems
1. Commutative: h1 [n] ∗ h2 [n] = h2 [n] ∗ h1 [n]
X
h1 [n] ∗ h2 [n] ≡ h1 [k]h2 [n − k] · · · · · · let l = n − k
k
X
= h1 [n − l]h2 [l]
l
X
= h2 [l]h1 [n − l]
l
≡ h2 [n] ∗ h1 [n]

Cascaded LTI systems are interchangeable.

2. Distributive: f [n] ∗ (g[n] + h[n]) = f [n] ∗ g[n] + f [n] ∗ h[n].

3. Associative: (f [n] ∗ g[n]) ∗ h[n] = f [n] ∗ (g[n] ∗ h[n]).


" #
X
f [n] ∗ [g[n] ∗ h[n]] ≡ f [n] ∗ g[k]h[n − k]
k
" #
X X
= f [l] g[k]h[n − l − k] · · · trick: let k = m − l
l k
XX
= f [l]g[m − l]h[n − m]
l m
" #
X X
= f [l]g[m − l] h[n − m]
m l
X
= [f [m] ∗ g[m]]h[n − m]
m
≡ [f [n] ∗ g[n]] ∗ h[n]

4. Convolution of x[n] with δ[n − n0 ] is x[n − n0 ].

5. If y[n] = x[n] ∗ h[n], then

• time-reversal: y[−n] = x[−n] ∗ h[−n]


• time-delay: y[n − n1 − n2 ] = x[n − n1 ] ∗ h[n − n2 ].
P∞ P∞ P∞
• DC: n=−∞ y[n] = n=−∞ x[n] · n=−∞ h[n], DCx × DCh = DCy .
• Define the center of gravity (or delay time) of a signal x[n] as
P∞
nx[n]
Dx ≡ Pk=−∞

k=−∞ x[n]
then Dy = Dx + Dn .

6. Edge and Width properties: If x[n] and y[n] are of finite duration with edges [n1 , n2 ] and [n3 , n4 ],
and widths N1 = n2 − n1 + 1 and N2 = n4 − n3 + 1, then z[n] = x[n] ∗ y[n] is of finite duration
with edges [n1 + n3 , n2 + n4 ] and width N = (n2 + n4 ) − (n1 + n3 ) + 1 = N1 + N2 − 1.

7. A DT LTI system with impulse response h[n] is causal if and only if h[n] = 0, ∀ n < 0.
Any noncausal FIR(Finite-duration Impulse Response) system can be made causal by cascading
it with a sufficiently long delay.

II-A - 4
(Pf) For an LTI system,

y[n] = x[n] ∗ h[n] = h[n] ∗ x[n]



X
= h[k]x[n − k]
k=−∞
= · · · + h[−2]x[n + 2] + h[−1]x[n + 1] + h[0]x[n] + h[1]x[n − 1] + h[2]x[n − 2] + · · ·
| {z } | {z }
noncausl part causal part

Recall that ”for a causal system, y[n] must be independent of the future input x[n + 1], x[n +
2], . . . .” Therefore, the impulse response of a causal system: h[n] = 0, ∀n < 0.
As a result, the i/o convolution of a causal LTI system can be written as
n
X ∞
X
y[n] = x[k]h[n − k] or h[k]x[n − k]
k=−∞ k=0

P
8. A DT LTI system is BIBO-stable iff its impulse response h[n] is absolutely summable: n |h[n]| <
∞.
(Pf of the sufficient condition)
Let the input is bounded: |x[n]| ≤ Bx , for all n,

X

|y[n]| = h[k]x[n − k]

k
X
≤ |h[k]| |x[n − k]|
k
X
≤ Bx |h[k]|
k
X
< ∞ if |h[k]| is bounded
k

(Pf of the necessary condition) If h[n] is not absolutely summable, consider the bounded input
x[k] = ±Bx for all k, where the sign of x[n0 − k] is chosen the same as h[k] then y[n] is not
P
bounded because |y[n0 ]| = Bx k |h[k]| = ∞.
Z Z Z

|y(t)| = x(τ )h(t − τ )dτ ≤ |x(τ )|h(t − τ )|dτ ≤ max |x(τ )| |h(λ)|dλ < ∞
τ

9. The inverse system, if exists, of an LTI system with impulse response h[n], has impulse response
hi [n] such that h[n] ∗ hi [n] = δ[n]. The inverse system of an accumulator system h[n] = u[n] is
a backward-difference system hi [n] = δ[n] − δ[n − 1].

10. The everlasting exponential z n is an eigenwaveform for a DT LTI system.


X
y[n] = h[n] ∗ z n = h[m]z n−m
m
X
n
= z h[m]z −m
≡ z n · H(z) k z n
m

P
where H(z) ≡ m h[m]z −m is called the transfer function of the LTI system, which is the
Z-transform of its impulse response.

II-A - 5

You might also like