20CYS111 Digital Signal Processing

Time Domain Representation of LTI Systems:

Impulse Response and Convolution

Dr. J. Aravinth (Mentor)

Linear and Time-Invariant (LTI) Systems

We classi ed systems in accordance with a number of

characteristic properties or categories, namely, linearity, causality,
stability, and time-invariance, etc.

In the remainder of this course, we will be mostly concerned with an

important class of systems, namely, the Linear and Time-Invariant
(LTI) systems.

LTI systems are mathematically easy to characterize and analyze,

and consequently, easy to design.

Highly useful signal processing algorithms have been developed for

this class of systems over the last several decades.
Linear and Time-Invariant (LTI) Systems

In this module, we are interested in characterizing the input-output

relationship of LTI systems in time domain, i.e., when both the input
and output signals are represented as functions of time.

We shall see that:

An LTI system is characterized in the time domain simply by

its response to an unit impulse function, called its impulse
response.
The output of an LTI system is given by the convolution of
the input and the impulse response of the LTI system.
LTI Systems in Discrete Time
Consequence of Linearity

Consider an arbitrary input signal, 𝑥[𝑛], applied to an LTI system

.

Suppose the input signal can be decomposed as a weighted sum of

elementary signals, 𝑥𝑘 [𝑛], 𝑘 = 1, 2, … , as

𝑥[𝑛] = 𝑎𝑘 𝑥𝑘 [𝑛] ,
∑
𝑘

where 𝑎𝑘 , 𝑘 = 1, 2, … , are the weighting coef cients.

Consequence of Linearity

Let the response of the LTI system to the elementary input 𝑥𝑘 [𝑛]
be 𝑦𝑘 [𝑛], i.e.,

{𝑥𝑘 [𝑛]} = 𝑦𝑘 [𝑛] .

Then, the output of the LTI system to the input 𝑥[𝑛] is given by

𝑦[𝑛] = {𝑥[𝑛]} =  𝑎𝑘 𝑥𝑘 [𝑛]

{∑ }
𝑘

= 𝑎𝑘 {𝑥𝑘 [𝑛]} = 𝑎𝑘 𝑦𝑘 [𝑛].

∑ ∑
𝑘 𝑘
Choice of Elementary Signals

The choice of elementary signals is dependent on the class of input

signals that we wish to consider.

Unit impulses as elementary signals: are mathematically

convenient and completely general (i.e., they apply to any
input signal).
Complex exponentials as elementary signals: are more
convenient mathematically if the input signal is periodic.

Example: Consider a nite-duration input sequence, given by

𝑥[𝑛] = {2, 4, 0, 3} , where the location of the time origin has
↑
been indicated with an upward arrow. Decompose 𝑥[𝑛] as a
weighted sum of impulses.
Decomposition of Input into Unit Impulses
Decomposition of Input into Unit Impulses

For any discrete-time signal 𝑥[𝑛], we have

𝑥[𝑛] = 𝑥[𝑘]𝛿[𝑛 − 𝑘] ,
∑
𝑘=−∞

which can be veri ed by substituting 𝑛 = 0, ±1, ±2, … .

Here,

The signals, 𝛿[𝑛 − 𝑘] , 𝑘 = 0, ±1, ±2, …, can be viewed

as the elementary signals.
The values, 𝑥[𝑘] , 𝑘 = 0, ±1, ±2, …, can be viewed as the
weighting coef cients.
Impulse Response of Discrete-Time LTI Systems

The output of a discete-time LTI system , when a discrete-time

unit impulse 𝛿[𝑛] is applied as input, is called its impulse response
ℎ[𝑛] , i.e.,

The impulse response ℎ[𝑛] of a discete-time system  is

de ned by

{𝛿[𝑛]} = ℎ[𝑛] .

Since the discrete-time LTI system  is time-invariant, the output

of the system, when a time-shifted unit impulse 𝛿[𝑛 − 𝑘] is
applied as input, is given by

{𝛿[𝑛 − 𝑘]} = ℎ[𝑛 − 𝑘] .

Output = Convolution of the Input and Impulse Response

The output 𝑦[𝑛] of the discrete-time LTI system  for any

arbitrary discrete-time input signal 𝑥[𝑛] is given by

𝑦[𝑛] = {𝑥[𝑛]} =  𝑥[𝑘]𝛿[𝑛 − 𝑘]

{ ∑ }
𝑘=−∞

∞ ∞

= 𝑥[𝑘]{𝛿[𝑛 − 𝑘]} = 𝑥[𝑘]ℎ[𝑛 − 𝑘].

∑ ∑
𝑘=−∞ 𝑘=−∞

⇒ 𝑦[𝑛] = 𝑥[𝑘]ℎ[𝑛 − 𝑘] = 𝑥[𝑛] ∗ ℎ[𝑛] ,

∑
𝑘=−∞

where '∗ ' denotes the convolution operation.

Finding the Impulse Response of an LTI System

Example: Let the input-output relation of a discrete-time LTI system

be given by

𝑦[𝑛] = 𝛼0 𝑥[𝑛] + 𝛼1 𝑥[𝑛 − 1] + 𝛼2 𝑥[𝑛 − 2] + 𝛼3 𝑥[𝑛 − 3].

To obtain the impulse response ℎ[𝑛] of the system, set

𝑥[𝑛] = 𝛿[𝑛] . Then, the impulse response of the system is

ℎ[𝑛] = 𝛼0 𝛿[𝑛] + 𝛼1 𝛿[𝑛 − 1] + 𝛼2 𝛿[𝑛 − 2] + 𝛼3 𝛿[𝑛 − 3].

Hence, the impulse response of the system is a sequence of length

four, given by

{𝛼0 , 𝛼1 , 𝛼2 , 𝛼3 },
↑

where the location of the time origin has been indicated with an
upward arrow.
Algorithm to Compute Convolution Sum

The output at a particular time 𝑛 = 𝑛0 is

∞
𝑦[𝑛0 ] = 𝑥[𝑘]ℎ[𝑛0 − 𝑘].
∑
𝑘=− ∞

Since the index in the summation is 𝑘, both the input signal

𝑥[𝑘] and the impulse response ℎ[𝑛0 − 𝑘] are functions of

𝑘.

Hence, follow the procedure outlined below:

Obtain the sequence ℎ[𝑛0 − 𝑘] from ℎ[𝑘] by

Folding or re ecting ℎ[𝑘] about 𝑘 = 0 (the time
origin), which results in the sequence ℎ[−𝑘]
Shifting the sequence ℎ[−𝑘] to the right by 𝑛0 to
yield ℎ[𝑛0 − 𝑘].
Multiply the sequences 𝑥[𝑘] and ℎ[𝑛0 − 𝑘] to form a
product sequence 𝑤𝑛 [𝑘] = 𝑥[𝑘]ℎ[𝑛0 − 𝑘] .
0

Sum the product sequence for all 𝑘 to get the output 𝑦[𝑛0 ]
at time 𝑛 = 𝑛0 .
Example to Compute Convolution Sum
Problem: Let the impulse response of a discrete-time LTI system be ℎ[𝑛] = 𝛼 𝑛 𝑢[𝑛],
where 0 < 𝛼 < 1 . Find the output 𝑦[𝑛] of the system in response to an input
𝑥[𝑛] = 𝑢[𝑛] .

Solution: We have
∞ ∞

𝑛−𝑘
𝑦[𝑛] = 𝑥[𝑛] ∗ ℎ[𝑛] = 𝑥[𝑘]ℎ[𝑛 − 𝑘] = 𝑢[𝑘]𝛼 𝑢[𝑛 − 𝑘]
∑ ∑
𝑘=−∞ 𝑘=−∞

𝑛+1
⎧ 1 − 𝛼
𝑛 𝑛−𝑘 𝑛 𝑘
⎪∑ 𝛼 = ∑ 𝛼 = for 𝑛 ≥ 0
𝑘=0 𝑘=0
= ⎨ 1 − 𝛼
⎪
⎩0 for 𝑛 < 0
LTI Systems in Continuous Time
Decomposition of Input into Unit Impulses

In the continuous time, any arbitrary input signal 𝑥(𝑡) can be re-
written as

∞
′ ′ ′
𝑥(𝑡) = 𝑥(𝑡 )𝛿(𝑡 − 𝑡 )𝑑𝑡 ,
∫
−∞

which is the convolution property satis ed by the delta function.

Remarks: We can interpret this property just as in the

discrete-time case by applying the following approximation:

∞ ∞

′ ′ ′
𝑥(𝑡) = 𝑥(𝑡 )𝛿(𝑡 − 𝑡 )𝑑𝑡 ≈ 𝑥(𝑘Δ𝑡)𝛿Δ𝑡 (𝑡 − 𝑘Δ𝑡)Δ𝑡
∫ ∑ 
−∞
𝑘=−∞

unit area

⇒ 𝑥(𝑡 0 ) ≈ 𝑥(𝑘0 Δ𝑡), where 𝑘0 Δ𝑡 ≤ 𝑡 < (𝑘0 + 1)Δ𝑡

Consequences of Linearity and Time-Invariance

Consider a continuous-time LTI system  to which we apply the

∞
input 𝑥(𝑡) = ∫ −∞ 𝑥(𝑡′ )𝛿(𝑡 − 𝑡′ )𝑑𝑡′ .

By linearity, we have
∞ ∞
′ ′ ′ ′ ′ ′
𝑦(𝑡) = {𝑥(𝑡)} =  𝑥(𝑡 )𝛿(𝑡 − 𝑡 )𝑑𝑡 = 𝑥(𝑡 ) {𝛿(𝑡 − 𝑡 )} 𝑑𝑡
{∫ } ∫
−∞ −∞

By time-invariance, we have

′ ′
{𝛿(𝑡)} = ℎ(𝑡) ⇒ {𝛿(𝑡 − 𝑡 )} = ℎ(𝑡 − 𝑡 )
Output = Convolution of the Input and Impulse Response

In summary, for the LTI system, we have

∞
′ ′ ′
𝑦(𝑡) = 𝑥(𝑡 )ℎ(𝑡 − 𝑡 )𝑑𝑡 = 𝑥(𝑡) ∗ ℎ(𝑡) ,
∫
−∞

where '∗ ' denotes convolution.

The convolution operation is a summation in the discrete-

time, whereas it is an integration in the continuous time.
Similar to the discrete-time case, the convolution in the
continuous-time case can be obtained by performing the
folding (or re ecting), shifting, multiplying and integrating
operations, in that order.
Example to Compute Convolution Integral
Example to Compute Convolution Integral

Radar Range Measurement: Let ℎ(𝑡) = 𝑎𝛿(𝑡 − 𝛽) . Find the

output 𝑟(𝑡) if the input is given by

sin(𝜔 𝑐 𝑡) 0 ≤ 𝑡 ≤ 𝑇0
𝑥(𝑡) = .
{0 otherwise
Example to Compute Convolution Integral

Solution: We have ℎ(𝑡′ ) = 𝑎𝛿(𝑡

′
.
− 𝛽)

′ ′ ′
⇒ ℎ(−𝑡 ) = 𝑎𝛿(−𝑡 − 𝛽) = 𝑎𝛿(𝑡 + 𝛽).

′ ′ ′ ′
⇒ ℎ(𝑡 − 𝑡 ) = ℎ(−(𝑡 − 𝑡)) = 𝑎𝛿((𝑡 − 𝑡) + 𝛽) = 𝑎𝛿(𝑡 − (𝑡 − 𝛽))

∞
′ ′ ′
⇒ 𝑟(𝑡) = 𝑥(𝑡) ∗ ℎ(𝑡) = 𝑥(𝑡 )ℎ(𝑡 − 𝑡 )𝑑𝑡
∫
−∞

∞
′ ′ ′
= 𝑥(𝑡 )𝑎𝛿(𝑡 − (𝑡 − 𝛽))𝑑𝑡 = 𝑎𝑥(𝑡 − 𝛽)
∫
−∞

𝑎 sin(𝜔 𝑐 (𝑡 − 𝛽)) 0 ≤ 𝑡 ≤ 𝑇0
= .
{0 otherwise
Interconnection of LTI Systems
Block Representation
Parallel Connection of LTI Systems

𝐃𝐢𝐬𝐭𝐫𝐢𝐛𝐮𝐭𝐢𝐯𝐞 𝐏𝐫𝐨𝐩𝐞𝐫𝐭𝐲: 𝑥(𝑡) ∗ ℎ1 (𝑡) + 𝑥(𝑡) ∗ ℎ2 (𝑡) = 𝑥(𝑡) ∗ {ℎ1 (𝑡) + ℎ2 (𝑡)} .
Series Connection of LTI Systems

𝐀𝐬𝐬𝐨𝐜𝐢𝐚𝐭𝐢𝐯𝐞 𝐏𝐫𝐨𝐩𝐞𝐫𝐭𝐲: {𝑥(𝑡) ∗ ℎ1 (𝑡)} ∗ ℎ2 (𝑡) = 𝑥(𝑡) ∗ {ℎ1 (𝑡) ∗ ℎ2 (𝑡)} .

𝐂𝐨𝐦𝐦𝐮𝐭𝐚𝐭𝐢𝐯𝐞 𝐏𝐫𝐨𝐩𝐞𝐫𝐭𝐲: ℎ1 (𝑡) ∗ ℎ2 (𝑡) = ℎ2 (𝑡) ∗ ℎ1 (𝑡) .

Example to Find Equivalent System
Properties of LTI Systems
Memoryless LTI Systems

For a discrete-time LTI system, we have

∞ ∞

𝑦[𝑛] = 𝑥[𝑘]ℎ[𝑛 − 𝑘] = ℎ[𝑘]𝑥[𝑛 − 𝑘]

∑ ∑
𝑘=−∞ 𝑘=−∞

⇒ 𝑦[𝑛0 ] = ℎ[0]𝑥[𝑛0 ] + ℎ[1]𝑥[𝑛0 − 1] + ℎ[2]𝑥[𝑛0 − 2] + …

+ ℎ[−1]𝑥[𝑛0 + 1] + ℎ[−2]𝑥[𝑛0 + 2] + …

If the system is memoryless, then 𝑦[𝑛0 ] can depend only on 𝑥[𝑛0 ]

and no other value of 𝑥[𝑛] ⇒ 𝑦[𝑛0 ] = ℎ[0]𝑥[𝑛0 ].

Therefore, for a discrete-time memoryless LTI system, we

must have

⇒ ℎ[𝑛] = 0, ∀𝑛 ≠ 0 ⇒ ℎ[𝑛] = 𝑐𝛿[𝑛] .

Memoryless LTI Systems

Similarly, for a continuous-time memoryless LTI system, we must

have

ℎ(𝑡) = 𝑐𝛿(𝑡) .

Remarks: The memoryless condition places severe restrictions on

the form of the impulse response.

All memoryless LTI systems simply perform scalar

multiplication of the input.
Causal LTI Systems

For a discrete-time LTI system, we have

∞ ∞

𝑦[𝑛] = 𝑥[𝑘]ℎ[𝑛 − 𝑘] = ℎ[𝑘]𝑥[𝑛 − 𝑘]

∑ ∑
𝑘=−∞ 𝑘=−∞

⇒ 𝑦[𝑛0 ] = ℎ[0]𝑥[𝑛0 ] + ℎ[1]𝑥[𝑛0 − 1] + ℎ[2]𝑥[𝑛0 − 2] + …

+ ℎ[−1]𝑥[𝑛0 + 1] + ℎ[−2]𝑥[𝑛0 + 2] + …

If the system is causal, then 𝑦[𝑛0 ] can depend only on 𝑥[𝑛],

𝑛 ≤ 𝑛0 .

Therefore, for a discrete-time causal LTI system, we must

have

⇒ ℎ[𝑛] = 0, ∀𝑛 < 0
Causal LTI Systems

The output of a discrete-time causal LTI system is given by

𝑛 ∞

⇒ 𝑦[𝑛] = 𝑥[𝑘]ℎ[𝑛 − 𝑘] = ℎ[𝑘]𝑥[𝑛 − 𝑘] .

∑ ∑
𝑘=−∞ 𝑘=0

Similarly, for a continuous-time memoryless LTI system, we must

have

ℎ(𝑡) = 0, ∀𝑡 < 0 .

𝑡 ∞
′ ′ ′ ′ ′ ′
⇒ 𝑦(𝑡) = 𝑥(𝑡 )ℎ(𝑡 − 𝑡 )𝑑𝑡 = ℎ(𝑡 )𝑥(𝑡 − 𝑡 )𝑑𝑡 .
∫ ∫
−∞ 0
Stable LTI Systems

Assume that the input 𝑥[𝑛] applied to a discrete-time LTI system is

bounded, i.e., suppose
|𝑥[𝑛]| ≤ 𝑀𝑥 < ∞, ∀𝑛.

Let the impulse response of the discrete-time LTI system be ℎ[𝑛] .

Then, we have
∞
|𝑦[𝑛]| = | ℎ[𝑘]𝑥[𝑛 − 𝑘]|
∑
𝑘=− ∞
∞ ∞
≤ |ℎ[𝑘]𝑥[𝑛 − 𝑘]| = |ℎ[𝑘]||𝑥[𝑛 − 𝑘]|
∑ ∑
𝑘=− ∞ 𝑘=− ∞
∞
≤ 𝑀𝑥 |ℎ[𝑘]|.
∑
𝑘=− ∞
Stable LTI Systems

Therefore, the discrete-time LTI system is BIBO stable, i.e., its

output is bounded for bounded input if the impulse response is
absolutely summable, that is, if

|ℎ[𝑘]| < ∞ .
∑
𝑘=−∞

Similarly, the continuous-time LTI system is BIBO stable, i.e., its

output is bounded for bounded input if the impulse response is
absolutely summable, that is, if

∞
′ ′
|ℎ(𝑡 )|𝑑𝑡 < ∞ .
∫
−∞
Invertible LTI Systems and Deconvolution

An LTI system may or may not have an inverse system, e.g., a

lowpass lter.

An inverse of an LTI system, if it exists, may or may not be an LTI

system itself.

Here, we restrict our discussion to inverse systems that are LTI.

For physical realizability and practical considerations, one

should restrict to inverse systems that are also causal and
stable.

Consider a continuous-time LTI system with impulse response ℎ(𝑡).

Assume that the inverse system exists and is itself an LTI system
with impulse response ℎ𝑖𝑛𝑣 (𝑡) .

This inverse LTI system may be noncausal or unstable.

Invertible LTI Systems and Deconvolution

By the de nition of the inverse system, we must have

𝑖𝑛𝑣
{𝑥(𝑡) ∗ ℎ(𝑡)} ∗ ℎ (𝑡) = 𝑥(𝑡)
𝑖𝑛𝑣
⇒ 𝑥(𝑡) ∗ {ℎ(𝑡) ∗ ℎ (𝑡)} = 𝑥(𝑡)

𝑖𝑛𝑣
⇒ ℎ(𝑡) ∗ ℎ (𝑡) = 𝛿(𝑡) .

For a discrete-time LTI system with impulse response ℎ[𝑛] , if the

inverse exists and is itself an LTI system with impulse response
[𝑛] , then we must have
𝑖𝑛𝑣
ℎ

𝑖𝑛𝑣
ℎ[𝑛] ∗ ℎ [𝑛] = 𝛿[𝑛] .
Example of Finding Inverse LTI System

Problem: Consider a multipath communication system given by

the input-output relationship in the discrete-time
𝑦[𝑛] = 𝑥[𝑛] + 𝑎𝑥[𝑛 − 1] . Find a causal LTI inverse system.

Check if the inverse system is stable.

Solution: Setting 𝑥[𝑛] = 𝛿[𝑛] , we nd the impulse response as

⎧1 𝑛 = 0
⎪
ℎ[𝑛] = ⎨ 𝑎 𝑛 = 1 .
⎪
⎩0 otherwise

Assume that an LTI inverse system exists whose impulse response is

[𝑛] . Then, we must have
𝑖𝑛𝑣
ℎ

𝑖𝑛𝑣 𝑖𝑛𝑣 𝑖𝑛𝑣

ℎ[𝑛] ∗ ℎ [𝑛] = 𝛿[𝑛] ⇒ ℎ [𝑛] + 𝑎ℎ [𝑛 − 1] = 𝛿[𝑛].
Example of Finding Inverse LTI System

For 𝑛 = 0 , we have 𝛿[𝑛] = 𝛿[0] = 1 , which implies that

𝑖𝑛𝑣 𝑖𝑛𝑣
ℎ [0] + 𝑎ℎ [−1] = 𝛿[0] = 1.

If the LTI inverse system has to be causal, we must have

[𝑛] = 0, ∀𝑛 < 0 . This implies that ℎ [−1] = 0, and
𝑖𝑛𝑣 𝑖𝑛𝑣
ℎ

hence ℎ [0] = 1 .
𝑖𝑛𝑣

For 𝑛 ≠ 0, we have 𝛿[𝑛] = 0 , which implies that

𝑖𝑛𝑣 𝑖𝑛𝑣 𝑖𝑛𝑣 𝑖𝑛𝑣

ℎ [𝑛] + 𝑎ℎ [𝑛 − 1] = 0 ⇒ ℎ [𝑛] = −𝑎ℎ [𝑛 − 1], ∀𝑛 ≠ 0.

Therefore, we have ℎ𝑖𝑛𝑣 [𝑛] .

𝑛
= (−𝑎) 𝑢[𝑛]
Example of Finding Inverse LTI System

To check if the inverse system is stable, we need to check if ℎ𝑖𝑛𝑣 [𝑛]

is absolutely summable.

Since ℎ𝑖𝑛𝑣 [𝑛] is causal, it is stable if

∞ ∞

𝑖𝑛𝑣 𝑘
|ℎ [𝑘]| = |𝑎| < ∞,
∑ ∑
𝑘=−∞ 𝑘=0

which is a geometric series that converges if |𝑎| < 1 .

References:
[1] Simon Haykin and Barry Van Veen, Signals and Systems, Second
Edition, John Wiley and Sons, 2003.

[2] Lecture Notes by Michael D. Adams.

https://www.ece.uvic.ca/~frodo/sigsysbook/downloads/lecture_slides_for_signals_and_systems
2.0.pdf
(https://www.ece.uvic.ca/~frodo/sigsysbook/downloads/lecture_slides_for_signals_and_system
2.0.pdf)

[3] Lecture Notes by Richard Baraniuk.

https://www.di.univr.it/documenti/OccorrenzaIns/matdid/matdid018094.pdf
(https://www.di.univr.it/documenti/OccorrenzaIns/matdid/matdid018094.pdf)