Lecture 4

DT LTI Systems and

Convolution Sum
Representation of DT Signals
▪ Decomposition of an arbitrary signal 𝒙 𝒏 into basic signals

▪ Express a DT signal 𝑥 𝑛 in terms of Kronecker delta functions

▪ Any discrete-time signal can be decomposed into a weighted sum
of shifted unit (Kronecker delta) impulses
∞

𝑥 𝑛 = ෍ 𝑥 𝑘 𝛿 𝑛−𝑘
𝑘=−∞

▪ This follows from the sifting property of the Kronecker delta.

Example
𝑥 𝑛 = 2𝛿 𝑛 + 3𝛿 𝑛 − 1 + 𝛿 𝑛 − 2 + 4𝛿 𝑛 − 3
4
3
2 1
𝑥𝑛
𝑛
0 1 2 3 4
2
2𝛿 𝑛
𝑛
0 1 2 3 4
3
3𝛿 𝑛 − 1
𝑛
0 1 2 3 4
𝛿 𝑛−2 1
𝑛
0 1 2 3 4
4
4𝛿 𝑛 − 3
𝑛
0 1 2 3 4
@Prof. A. Murat Tekalp Koç University Digital Signal Processing

Representation in the Standard Basis

2 The signals 1, 0, 0, 0 , 0, 1, 0, 0 , 0, 0, 1, 0 and
0 1 2 3 𝑛 0, 0, 0, 1 constitute the standard basis vectors for
3 the 4-D signal space
+ 1 0 0 0
0 1 2 3 𝑛 𝐞0 = 0 , 𝐞1 = 1 , 𝐞2 = 0 , 𝐞3 = 0
0 0 1 0
1 0 0 0 1
+
𝑛
0 1 2 3
The signal 2, 3,1, 4 can be expressed in
+ 4 the standard basis as
𝑛 2 1 0 0 0
0 1 2 3
𝐱= 3 =2 0 +3 1 + 0 +4 0
4 5 0 0 1 0
= 3
2 1 1 0 0 0 1
𝑛 The components 2, 3, 1, 4 are projections of 𝐱 onto
0 1 2 3 the standard orthonormal basis vectors.
Time-Invariance
▪ If we shift the input by 𝑁 samples, the output will also shift
by 𝑁 samples

LTI
𝛿𝑛 ℎ𝑛
System

LTI
𝛿 𝑛−1 ℎ 𝑛−1
System
Linear Time-Invariant Systems
▪ Let’s use linearity and time-invariance together
LTI
2𝛿 𝑛 System 2ℎ 𝑛
ℎ𝑛

LTI
3𝛿 𝑛 − 1 System 3ℎ 𝑛 − 1
ℎ𝑛

LTI
2𝛿 𝑛 + 3𝛿 𝑛 − 1 System 2ℎ 𝑛 + 3ℎ 𝑛 − 1
ℎ𝑛
 Convolution Summation
▪ The output of LTI system to any input 𝑥 𝑛 is given by the convolution summation
is defined as
∞

𝑦 𝑛 = ෍ 𝑥 𝑘 ℎ 𝑛−𝑘
𝑘=−∞

▪ Derivation: Express the input in terms of weighted sum of shifted Kronecker deltas
▪ Compute the output for the input x 0 𝛿 𝑛 as x 0 ℎ 𝑛
▪ Use time invariance to write the output for the input x 𝑘 𝛿 𝑛 − 𝑘 as x 𝑘 ℎ 𝑛 − 𝑘
▪ Use linearity to write
∞ ∞

𝑥 𝑛 = ෍ 𝑥 𝑘 𝛿 𝑛−𝑘 LTI 𝑦 𝑛 = ෍ 𝑥 𝑘 ℎ 𝑛−𝑘

𝑘=−∞ 𝑘=−∞
System
ℎ𝑛
▪ That is, the output is given by the weighted sum of shifted impulse responses
Convergence
▪ The convolution sum
∞

𝑦 𝑛 = ෍ 𝑥 𝑘 ℎ 𝑛−𝑘
𝑘=−∞
is an infinite sum.
▪ Whenever we have an infinie sum, we have to worry about its convergence.

▪ The convolution summation converges if 𝑥 𝑛 is bounded and ℎ 𝑛 is absolutely

summable; i.e,
∞

𝑥 𝑛 < 𝐵 and ෍ ℎ 𝑘 < ∞

𝑘= −∞
▪ OR if ℎ 𝑛 is bounded and 𝑥 𝑛 is absolutely summable; i.e,
∞

ℎ 𝑛 < 𝐵 a𝑛𝑑 ෍ 𝑥 𝑘 < ∞

𝑘=−∞
 Stability of DT LTI Systems
▪ A DT LTI system is BIBO stable if and only if
∞

෍ ℎ𝑛 <∞
𝑛=−∞
▪ Proof:
▪ By definition, we need to show 𝑦 𝑛 < ∞ given 𝑥 𝑛 < 𝐵 < ∞ for all 𝑛.
▪ Express
∞ ∞

𝑦𝑛 = ෍ ℎ 𝑘 𝑥 𝑛−𝑘 ≤ ෍ ℎ 𝑘 𝑥 𝑛−𝑘
𝑘=−∞ 𝑘=−∞
▪ Replace 𝑥 𝑛 − 𝑘 by its upper bound 𝐵. This preserves the inequality.
∞

𝑦𝑛 ≤ ෍ ℎ𝑘 𝐵
𝑘=−∞
▪ Hence, 𝑦 𝑛 < ∞ if and only if σ∞
𝑘=−∞ ℎ𝑘 <∞
 Causality of DT LTI Systems
▪ A DT LTI system is causal if and only if
ℎ 𝑛 = 0 for 𝑛 < 0
▪ Proof:
▪ For a causal system, the output 𝑦 𝑛 can only depend on inputs for 𝑘 = −∞, ⋯ , 𝑛.
Hence,
𝑛

𝑦 𝑛 = ෍ 𝑥 𝑘 ℎ 𝑛−𝑘
𝑘=−∞

▪ Let’s make a change of variables 𝑚 = 𝑛 − 𝑘

∞

𝑦 𝑛 = ෍ 𝑥 𝑛−𝑚 ℎ 𝑚
𝑚=0
▪ Limits of the sum: 𝑘 = −∞ → 𝑚 = ∞, and 𝑘 = 𝑛 → 𝑚 = 0
▪ Hence, we have ℎ 𝑛 = 0 for 𝑛 < 0.
Properties of Convolution Sum
1) It is commutative

𝑦 𝑛 =ℎ 𝑛 ∗𝑥 𝑛 =𝑥 𝑛 ∗ℎ 𝑛

2) It is associative

𝑦 𝑛 = ℎ1 𝑛 ∗ ℎ2 𝑛 ∗ 𝑥 𝑛 = ℎ1 𝑛 ∗ ℎ2 𝑛 ∗ 𝑥 𝑛

3) It is distributive over addition

𝑦 𝑛 = ℎ1 𝑛 + ℎ2 𝑛 ∗ 𝑥 𝑛 = ℎ1 𝑛 ∗ 𝑥 𝑛 + ℎ2 𝑛 ∗ 𝑥 𝑛
Commutativity
▪ The convolution summation is commutative, i.e.,
𝑦 𝑛 =ℎ 𝑛 ∗𝑥 𝑛 =𝑥 𝑛 ∗ℎ 𝑛

LTI LTI
𝑥𝑛
ℎ𝑛 𝑦𝑛 ≡ ℎ𝑛
𝑥𝑛 𝑦𝑛

▪ Proof:
▪ From the definition:
∞

𝑦 𝑛 =𝑥 𝑛 ∗ℎ 𝑛 = ෍ 𝑥 𝑘 ℎ 𝑛−𝑘
𝑘=−∞
▪ Let’s do a change of variables 𝑚 = 𝑛 − 𝑘
∞

𝑦 𝑛 = ෍ ℎ 𝑚 𝑥 𝑛−𝑚 =ℎ 𝑛 ∗𝑥 𝑛
𝑚=−∞
Associativity
▪ The convolution summation is associative.
𝑦 𝑛 = ℎ1 𝑛 ∗ 𝑥 𝑛 ∗ ℎ2 𝑛 = ℎ1 𝑛 ∗ ℎ2 𝑛 ∗ 𝑥 𝑛
= ℎ1 𝑛 ∗ ℎ2 𝑛 ∗ 𝑥 𝑛
▪ The following series connection of systems are all equivalent.
LTI LTI
𝑥𝑛 𝑦𝑛
ℎ1 𝑛 ℎ2 𝑛

LTI LTI
𝑥𝑛 𝑦𝑛
ℎ2 𝑛 ℎ1 𝑛

LTI
𝑥𝑛 𝑦𝑛
ℎ1 𝑛 ∗ ℎ2 𝑛
Distributivity over Addition
▪ The convolution summation is distributive with addition
𝑦 𝑛 = ℎ1 𝑛 + ℎ2 𝑛 ∗ 𝑥 𝑛 = ℎ1 𝑛 ∗ 𝑥 𝑛 + ℎ2 𝑛 ∗ 𝑥 𝑛
▪ The following systems are equivalent

LTI
ℎ1 𝑛
𝑥𝑛 + 𝑦𝑛
LTI
ℎ2 𝑛

LTI
𝑥𝑛 𝑦𝑛
ℎ1 𝑛 + ℎ2 𝑛
Computation of Convolution Sum
▪ The convolution sum
∞ ∞

𝑦 𝑛 = ෍ 𝑥 𝑘 ℎ 𝑛−𝑘 = ෍ ℎ 𝑘 𝑥 𝑛−𝑘
𝑘=−∞ 𝑘=−∞

▪ CASE 1: Both 𝑥 𝑛 and ℎ 𝑛 are finite-extent signals

– Table method
▪ CASE 2: Either 𝑥 𝑛 or ℎ 𝑛 is infinite-extent or both are infinite-extent
– Compute geometric series
▪ CASE 3: Either 𝑥 𝑛 or ℎ 𝑛 is a periodic signal
– The output is periodic
▪ CASE 4: Both 𝑥 𝑛 and ℎ 𝑛 are periodic signals
– Periodic convolution – the output is periodic
CASE 1- Both Finite-Extent
▪ Convolution of an 𝑀-point sequence with an 𝑁-point sequence results
in a 𝑀 + 𝑁 − 1 point sequence.
▪ Example: ℎ[𝑛] = 1,2,0, −1,1 𝑀 = 5 point sequence
▪ 𝑥 𝑛 = 1,3, −1, −2 𝑁 = 4 point sequence
𝑀−1

𝑦 𝑛 = ෍ ℎ 𝑘 𝑥 𝑛−𝑘
𝑘=0
▪ 𝑘=0 𝑛 𝑦𝑛
▪ ℎ 𝑘 1 2 0 –1 1
▪ 𝑥 𝑘 1 3 –1 -2
▪ 𝑥 −𝑘 -2 –1 3 1 0 1
▪ 𝑥 1−𝑘 -2 –1 3 1 1 5
▪ 𝑥 2−𝑘 -2 –1 3 1 2 5
▪ 𝑥 3−𝑘 -2 –1 3 1 3 -5
▪ 𝑥 4−𝑘 -2 –1 3 1 4 -6
▪ 𝑥 5−𝑘 -2 –1 3 1 5 4
▪ 𝑥 6−𝑘 -2 –1 3 1 6 1
▪ 𝑥 7−𝑘 -2 1 3 1 7 -2
▪ 𝑦 𝑛 is 8 points
CASE 2: At least one Infinite-Extent
▪ EXAMPLE: Given 𝛼 < 1 and 𝛽 < 1
𝑥 𝑛 = 𝛼𝑛𝑢 𝑛
ℎ 𝑛 = 𝛽𝑛 𝑢 𝑛
▪ Find 𝑦 𝑛
▪ Solution:
∞ ∞

𝑦 𝑛 = ෍ 𝑥 𝑘 ℎ 𝑛 − 𝑘 = ෍ 𝛼 𝑘 𝑢 𝑘 𝛽𝑛−𝑘 𝑢 𝑛 − 𝑘
𝑘=−∞ 𝑘=−∞
∞

= ෍ 𝛼 𝑘 𝛽 𝑛−𝑘 𝑢 𝑘 𝑢 𝑛 − 𝑘
𝑘=−∞
𝑢𝑘 ⋯
𝑘
0 1 2 3 4 ⋯
𝑢 −𝑘 ⋯ 𝑘
⋯ −3 −2 −1 0 1 2 ⋯
𝑢 2−𝑘 ⋯
𝑘
0 1 2 3 4 ⋯
CASE 2: Example (cont’d)
▪ If 𝛼𝛽 −1 ≠ 1
𝑛 𝑛 𝑛+1
1 − 𝛼𝛽 −1
𝑦 𝑛 = ෍ 𝛽 𝑛 𝛼 ⋅ 𝛽 −1 𝑘
= 𝛽 𝑛 ෍ 𝛼𝛽 −1 𝑘 𝑛
=𝛽 ⋅
1 − 𝛼𝛽 −1
𝑘=0 𝑘=0
𝛼 𝑛+1 − 𝛽 𝑛+1
=
𝛼−𝛽
▪ If 𝛼𝛽 −1 = 1
𝑛

𝑦 𝑛 = 𝛽𝑛 ෍ 1 = 𝑛 + 1 𝛽𝑛
𝑘=0
▪ Thus,
𝛼 𝑛+1 − 𝛽 𝑛+1
𝛼≠𝛽
𝑦 𝑛 =൞ 𝛼−𝛽
𝑛 + 1 ⋅ 𝛽𝑛 𝛼=𝛽
 CASE 3: Periodic Input
▪ The input is periodic with period 𝑁, impulse response is not

▪ Then, the output is given by

∞ ∞

𝑥෤ 𝑛 = ෍ 𝑥 𝑛 − 𝑖𝑁 𝑦 𝑛 = ෍ 𝑥෤ 𝑘 ℎ 𝑛 − 𝑘
LTI
𝑖=−∞ 𝑘=−∞
System
ℎ𝑛

▪ Hence, the convolution summation is given by

∞

𝑦෤ 𝑛 = ෍ 𝑥෤ 𝑘 ℎ 𝑛 − 𝑘
𝑘=−∞
 CASE 4: Periodic Convolution
▪ Also known as Circular Convolution
▪ Both signals are periodic with a common period 𝑁

▪ The output is given by the convolution summation over the common period
𝑁

𝑦෤ 𝑛 = ෍ 𝑥෤ 𝑘 ℎ෨ 𝑛 − 𝑘 , 𝑛 = 0, ⋯ , 𝑁 − 1
𝑘=0

∞ 𝑁
𝑥෤ 𝑛 = ෍ 𝑥 𝑛 − 𝑖𝑁 LTI 𝑦෤ 𝑛 = ෍ 𝑥෤ 𝑘 ℎ෨ 𝑛 − 𝑘
𝑖=−∞ 𝑘=0
System
ℎ෨ 𝑛
 Circular (Periodic) Convolution
▪ Convolution of two periodic sequences with a common period 𝑁 samples
results in a periodic sequence with period 𝑁 samples.
▪ Example: ℎ[𝑛] = 1,2,0, −1 𝑁 = 4 point sequence
▪ 𝑥 𝑛 = 1,3, −1, −2 𝑁 = 4 point sequence
𝑁−1

𝑦෤ 𝑛 = ෍ ℎ෨ 𝑘 𝑥෤ 𝑛 − 𝑘
𝑘=0

▪ 𝑘=0 𝑛 𝑦𝑛
▪ ℎ 𝑘 1 2 0 –1
▪ 𝑥 𝑘 1 3 –1 -2
▪ 𝑥 −𝑘 -2 –1 3 1 -2 -1 3 0 -6
▪ 𝑥 1−𝑘 -2 –1 3 1 -2 -1 1 6
▪ 𝑥 2−𝑘 -2 –1 3 1 -2 2 7
▪ 𝑥 3−𝑘 -2 –1 3 1 3 -5