23ECE112

Signal Processing I
Latha Srinivasan
Assistant Professor
Department of ECE
ASE, Bengaluru
 What we have learned last few classes
• Different properties of continuous time and discrete time systems

• Among these properties linearity and time invariance play important role in analysis of signals and systems

• Many physical systems possesses both linearity property and time invariance property.

• These systems can be modeled as a linear, time invariant system

• A system that possesses linearity and time invariance property is called a Linear Time Invariant (LTI) system.

• Linearity and time invariance properties allow the mathematical analysis of LTI systems to be much simpler
when compared to non linear time variant systems
 Why LTI systems are easier to analyse?

• Satisfy linearity property

• If input to an LTI system -> linear combination of several signals then output -> same linear combination of
the individual outputs

• Let us assume that there is a set of elementary signals, and we know the output for these elementary signals.

• Suppose that any arbitrary signal x(t) can be represented as a linear combination of these elementary signals.

• Then the output corresponding to ANY signal x(t) is easy.

• Output will be the same linear combination of the outputs of these elementary signals
 Why LTI systems are easier to analyse?
• Suppose the elementary signals are given by 𝑒1 𝑡 , 𝑒2 𝑡 , … , 𝑒𝑁 𝑡

• Let the output for the LTI system corresponding to 𝑒1 𝑡 , 𝑒2 𝑡 , … , 𝑒𝑁 𝑡 be 𝑟1 𝑡 , 𝑟2 𝑡 , … , 𝑟𝑁 𝑡 respectively.

• We assumed that any arbitrary signal x(t) can be represented using these elementary signals i.e,

• 𝑥 𝑡 = 𝑎1 𝑒1 𝑡 + 𝑎2 𝑒2 𝑡 + ⋯ + 𝑎𝑁 𝑒𝑁 𝑡

• If x(t) is applied as the input to the LTI system what would be the output?

• 𝑦 𝑡 = 𝑎1 𝑟1 𝑡 + 𝑎2 𝑟2 𝑡 + ⋯ + 𝑎𝑁 𝑟𝑁 𝑡

• You will be able to find the output for any x(t).

• You need to know how to represent x(t) as linear combination of 𝑒1 𝑡 , 𝑒2 𝑡 , … , 𝑒𝑁 𝑡 .

• For any x(t), the set of elementary signals remain same but the coefficients 𝑎1 , 𝑎2 , … , 𝑎𝑁 will be different
 Why LTI systems are easier to analyse?
• The same holds true for discrete time systems also

• Suppose the elementary signals are given by 𝑒1 [𝑛], 𝑒2 [𝑛], … , 𝑒𝑁 [𝑛]

• Let the output for the LTI system corresponding to 𝑒1 [𝑛], 𝑒2 [𝑛], … , 𝑒𝑁 [𝑛] be 𝑟1 [𝑛], 𝑟2 [𝑛], … , 𝑟𝑁 [𝑛] respectively.

• We assumed that any arbitrary discrete time signal x[n] can be represented using these elementary signals i.e,

• 𝑥[𝑛] = 𝑎1 𝑒1 [𝑛] + 𝑎2 𝑒2 [𝑛] + ⋯ + 𝑎𝑁 𝑒𝑁 [𝑛]

• If x[n] is applied as the input to the LTI system the output will be 𝑦[𝑛] = 𝑎1 𝑟1 [𝑛] + 𝑎2 𝑟2 [𝑛] + ⋯ + 𝑎𝑁 𝑟𝑁 [𝑛]

• But again, you need to know how to represent x[n] as linear combination of 𝑒1 [𝑛], 𝑒2 [𝑛], … , 𝑒𝑁 [𝑛].

• You should be able to calculate 𝑎1 , 𝑎2 , … , 𝑎𝑁

 Why LTI systems are easier to analyse?
• All the discussions till now revolved around linearity property. What about time invariance?

• Assume that all elementary signals 𝑒1 𝑡 , 𝑒2 𝑡 , … , 𝑒𝑁 𝑡 are shifted versions of a single elementary signal e(t).

• i.e, 𝑒1 𝑡 = 𝑒(𝑡 − 𝜏1 ), 𝑒2 𝑡 = 𝑒 𝑡 − 𝜏2 , 𝑒3 𝑡 = 𝑒 𝑡 − 𝜏3 , … 𝑒𝑁 𝑡 = 𝑒 𝑡 − 𝜏𝑁

• Suppose we know the output of the system when e(t) is given as input signal. Let it be r(t).

• What is output corresponding to 𝑒1 𝑡 = 𝑒(𝑡 − 𝜏1 )?

• Since system is time invariant, the output of 𝑒1 𝑡 = 𝑒(𝑡 − 𝜏1 ) will be 𝑟(𝑡 − 𝜏1 )

• 𝑟1 𝑡 = 𝑟(𝑡 − 𝜏1 ), 𝑟2 𝑡 = 𝑟 𝑡 − 𝜏2 , 𝑟3 𝑡 = 𝑟 𝑡 − 𝜏3 , … 𝑟𝑁 𝑡 = 𝑟 𝑡 − 𝜏𝑁

• If you know the output r(t) corresponding to the input as e(t), we can find out the outputs corresponding to 𝑟1 𝑡 , 𝑟2 𝑡 , … , 𝑟𝑁 𝑡
 Why LTI systems are easier to analyse?
• The same holds true for discrete time systems also

• Suppose the elementary signal is e[n] whose output r[n] is known to us.

• All the elementary signals given by 𝑒1 [𝑛], 𝑒2 [𝑛], … , 𝑒𝑁 [𝑛] are shifted versions of e[n].

• i.e, 𝑒1 [𝑛] = 𝑒[𝑛 − 𝑛1 ], 𝑒2 [𝑛] = 𝑒[𝑛 − 𝑛2 ], … 𝑒𝑁 [𝑛] = 𝑒[𝑛 − 𝑛𝑁 ]

• If the system is time invariant (aka shift invariant) the system outputs corresponding to 𝑒1 [𝑛], 𝑒2 [𝑛], … , 𝑒𝑁 [𝑛]
are 𝑟1 𝑛 = 𝑟 𝑛 − 𝑛1 , 𝑟2 𝑛 = 𝑟 𝑛 − 𝑛2 , … , 𝑟𝑁 𝑛 = 𝑟[𝑛 − 𝑛𝑁 ]respectively.

• What is the advantage?

 What is the advantage?
• For an arbitrary input signal x(t) given by 𝑥 𝑡 = 𝑎1 𝑒1 𝑡 + 𝑎2 𝑒2 𝑡 + ⋯ + 𝑎𝑁 𝑒𝑁 𝑡

• 𝑥 𝑡 = 𝑎1 𝑒(𝑡 − 𝜏1 ) + 𝑎2 𝑒(𝑡 − 𝜏2 ) + ⋯ + 𝑎𝑁 𝑒(𝑡 − 𝜏𝑁 )

• Output y(t) can be written as 𝑦 𝑡 = 𝑎1 𝑟1 𝑡 + 𝑎2 𝑟2 𝑡 + ⋯ + 𝑎𝑁 𝑟𝑁 𝑡

• 𝑦 𝑡 = 𝑎1 𝑟(𝑡 − 𝜏1 ) + 𝑎2 𝑟(𝑡 − 𝜏2 ) + ⋯ + 𝑎𝑁 𝑟(𝑡 − 𝜏𝑁 )

• To find y(t) you only need to know the output r(t) corresponding to a single elementary signal e(t)

• If you know how to obtain linear combination coefficients 𝑎1 , 𝑎2 , … , 𝑎𝑁 for any arbitrary signal x(t), you need to
know only r(t) to find the corresponding output.
 What is the advantage?
• Similarly in the discrete case suppose we can write any arbitrary signal x[n] as 𝑥[𝑛] = 𝑎1 𝑒1 [𝑛] + 𝑎2 𝑒2 [𝑛] +
⋯ + 𝑎𝑁 𝑒𝑁 [𝑛]

• 𝑥[𝑛]= 𝑎1 𝑒[𝑛 − 𝑛1 ] + 𝑎2 𝑒[𝑛 − 𝑛2 ] + ⋯ + 𝑎𝑁 𝑒[𝑛 − 𝑛𝑁 ]

• The output y[n] when the input is x[n] can be obtained as 𝑦 𝑛 = 𝑎1 𝑟1 𝑛 + 𝑎2 𝑟2 𝑛 + ⋯ + 𝑎𝑁 𝑟𝑁 𝑛

• y[n] = 𝑎1 𝑟[𝑛 − 𝑛1 ] + 𝑎2 𝑟[𝑛 − 𝑛2 ] + ⋯ + 𝑎𝑁 𝑟[𝑛 − 𝑛𝑁 ]

• We can find the output of any arbitrary signal x[n] if:

• We know how to express x[n] in terms of shifted versions of e[n]

• We know the output of the system r[n] when the input is e[n]
 Summary
• Assumptions: Any arbitrary signal x(t)/x[n] can be expressed as linear combination of elementary signals
𝑒1 𝑡 /𝑒1 [𝑛], 𝑒2 𝑡 /𝑒2 [𝑛], … , 𝑒𝑁 𝑡 /𝑒𝑁 [𝑛]

• The elementary signals 𝑒1 𝑡 /𝑒1 [𝑛], 𝑒2 𝑡 /𝑒2 [𝑛], … , 𝑒𝑁 𝑡 /𝑒𝑁 [𝑛]are generated as time shifted versions of a basic
signal e(t)/e[n].

• The output of LTI system when the input is e(t)/e[n] is known and is r(t)/r[n]

• Conclusions: Under these assumptions the output of the LTI system due to any arbitrary input x(t)/x[n] can be
computed.

• Are all our assumptions valid/possible?

• Do we have such e(t)/e[n] that generates 𝑒1 𝑡 /𝑒1 [𝑛], 𝑒2 𝑡 /𝑒2 [𝑛], … , 𝑒𝑁 𝑡 /𝑒𝑁 [𝑛]?

• If so, what is the value of N? How many elementary signals do we need to represent any arbitrary signal x(t)/x[n]

• Is it finite or infinite?
 Answers for the discrete case
• The unit impulse signal 𝛿 𝑛 can be used as e[n].

• Elementary signals will be shifted versions of 𝛿 𝑛

• Any arbitrary signal x[n] can be represented as a linear combination of the shifted unit impulse signals.

• Why and How?

• Consider an example signal x[n]

 Answers for discrete case

• What happens if we add 𝑧1 𝑛 , 𝑧2 𝑛 , … , 𝑧5 𝑛 ?

• We obtain 𝑥 𝑛 = 𝑧1 𝑛 + 𝑧2 𝑛 + 𝑧3 𝑛 + 𝑧4 𝑛 + 𝑧5 𝑛

• 𝑧1 𝑛 = −2𝛿 𝑛 + 2

• 𝑧2 𝑛 = 𝛿 𝑛 + 1

• 𝑧3 𝑛 = 2𝛿[𝑛]

• Similarly 𝑧4 𝑛 = −1𝛿 𝑛 − 1 and 𝑧5 𝑛 = 2𝛿[𝑛 − 2]

• 𝑥 𝑛 = −2𝛿 𝑛 + 2 + 𝛿 𝑛 + 1 + 2𝛿[𝑛] − 𝛿 𝑛 − 1 +
2𝛿[𝑛 − 2]
 Another example
• Express the given signal as linear combination of shifted impulse signal
• 𝑥 𝑛 = 3𝛿 𝑛 + 2𝛿 𝑛 − 1 + 𝛿 𝑛 − 2 + 0.5𝛿 𝑛 − 3
• In this way we can represent any x[n] as linear
combination of shifted impulse signals.
• So 𝛿 𝑛 works as basic elementary signal e[n] in discrete
case.
• 𝑒1 [𝑛], 𝑒2 [𝑛], … , 𝑒𝑁 [𝑛] are shifted impulse signals.
• What is N?

• N depends on length of the signal

• If x[n] extends from -∞ to ∞ then we need infinitely many shifted versions of 𝛿 𝑛 .
• If length of x[n] is finite, we need only a finite number of shifted versions of 𝛿 𝑛 .
 Example
• 𝑥 𝑛 = 2𝛿 𝑛 + 2 − 2𝛿 𝑛 + 1 + 3𝛿 𝑛 +
4𝛿 𝑛−1

• 𝑥 𝑛 = 𝑎−2 𝛿 𝑛 + 2 + 𝑎−1 𝛿 𝑛 + 1 + 𝑎0 𝛿 𝑛 +
𝑎1 𝛿 𝑛 − 1

• What are linear combination coefficients 𝑎−2 ,

𝑎−1 , 𝑎0 , 𝑎1 in terms of x[n]?

• 𝑎−2 = 𝑥 −2 , 𝑎−1 = 𝑥 −1 , 𝑎0 = 𝑥 0 , 𝑎1 = 𝑥[1]

• For any arbitrary signal x[n] we can write as 𝑥 𝑛 = σ∞
𝑘=−∞ 𝑥[𝑘] 𝛿 𝑛 − 𝑘

• Linear combination coefficients are values of the signal itself.

• Sifting property: Unit impulse signal sifts out the value of x[n]
• What will be the output of the LTI system if this signal is applied at input?
 Output of LTI system
• 𝑥 𝑛 = σ∞
𝑘=−∞ 𝑥[𝑘] 𝛿 𝑛 − 𝑘 = x −∞ 𝛿 𝑛 + ∞ + ⋯ + x −2 𝛿 𝑛 + 2 + x −1 𝛿 𝑛 + 1 + x 0 𝛿 𝑛 +

x 1 𝛿 𝑛 − 1 + x 2 𝛿 𝑛 − 2 + ⋯+ x ∞ 𝛿 𝑛 − ∞

• We know that system is linear and time invariant. Let us consider just the linearity aspect for the time being.

• Then output 𝑦 𝑛 = x −∞ 𝑜𝑢𝑡𝑝𝑢𝑡(𝛿 𝑛 + ∞ ) + ⋯ + x −2 𝑜𝑢𝑡𝑝𝑢𝑡(𝛿 𝑛 + 2 ) + x −1 𝑜𝑢𝑡𝑝𝑢𝑡(𝛿 [𝑛 +

1]) + x 0 𝑜𝑢𝑡𝑝𝑢𝑡(𝛿 𝑛 ) + x 1 𝑜𝑢𝑡𝑝𝑢𝑡(𝛿 𝑛 − 1 ) + x 2 𝑜𝑢𝑡𝑝𝑢𝑡(𝛿 𝑛 − 2 ) + ⋯ + x ∞ 𝑜𝑢𝑡𝑝𝑢𝑡(𝛿 𝑛 − ∞ )

• For computing y[n] we need to know the output of the system when the input is 𝛿 𝑛 + ∞ …, 𝛿 𝑛 + 2 ,
𝛿 𝑛 + 1 ,𝛿 𝑛 , 𝛿 𝑛 − 1 ,𝛿 𝑛 − 2 ,…𝛿 𝑛 − ∞

• In other words we need to know the output of the system corresponding to any 𝛿 𝑛 − 𝑘 .
 Output of LTI system
• Let the output corresponding to 𝛿 𝑛 − 𝑘 be ℎ𝑘 [𝑛]

• 𝛿[𝑛 − 1] → ℎ1 𝑛 , 𝛿[𝑛 − 2] → ℎ2 [𝑛], 𝛿 𝑛 + 1 → ℎ−1 𝑛 , 𝛿[𝑛 + 2] → ℎ−2 [𝑛]

• Then 𝑦 𝑛 = x −∞ ℎ−∞ 𝑛 + ⋯ + x −2 ℎ−2 𝑛 + x −1 ℎ−1 𝑛 + x 0 ℎ0 𝑛 + x 1 ℎ1 𝑛 + x 2 ℎ2 𝑛 + ⋯ + x ∞ ℎ∞ 𝑛

• What is ℎ0 [𝑛]?

• Output of system due to 𝛿[𝑛] at input

• Now we bring in time invariance property.

• If 𝛿[𝑛] produces output ℎ0 𝑛 , then what is output due to 𝛿[𝑛 − 1]?

• 𝛿[𝑛 − 1] → ℎ0 [𝑛 − 1]

• In general, 𝛿[𝑛 − 𝑘] → ℎ0 [𝑛 − 𝑘]
 Output of LTI system
• 𝑦 𝑛 = x −∞ ℎ−∞ 𝑛 + ⋯ + x −2 ℎ−2 𝑛 + x −1 ℎ−1 𝑛 + x 0 ℎ0 𝑛 + x 1 ℎ1 𝑛 + x 2 ℎ2 𝑛 + ⋯ +
x ∞ ℎ∞ 𝑛

• Output y[n] can be written as 𝑦 𝑛 = σ∞

𝑘=−∞ 𝑥[𝑘] ℎ𝑘 [𝑛]

• After applying time invariance we get

• 𝑦 𝑛 = σ∞
𝑘=−∞ 𝑥[𝑘] ℎ0 [𝑛 − 𝑘]

• Subscript 0 of h can be omitted since it is 0 for all values of k. So we can rewrite y[n] as

• 𝑦 𝑛 = σ∞
𝑘=−∞ 𝑥[𝑘] ℎ[𝑛 − 𝑘]

• Where h[n]= ℎ0 [𝑛] is the output of the system when the input is 𝛿[𝑛]

• h[n] is called impulse response of the system as it is the response of the system when a unit impulse signal is
applied at input.
 Output of LTI system
• 𝑦 𝑛 = σ∞
𝑘=−∞ 𝑥[𝑘] ℎ[𝑛 − 𝑘] provides the output of the LTI system when the input is any arbitrary discrete time

signal x[n].

• As long as you know the impulse response of the system.

• Different LTI systems will have different impulse responses.

• Initially when we studied discrete time systems we represented them using input-output relation. For e.g: 𝑦 𝑛 =
2𝑥 𝑛 + 3, y n = 5x n − 2 + 4x n − 1 + 3x[n]

• This is called direct input-output representation of a system

• 𝑦 𝑛 = σ∞
𝑘=−∞ 𝑥[𝑘] ℎ[𝑛 − 𝑘] is another way of representing input output relation of the LTI system. It

characterises system using impulse response.

• Hence also known as impulse response characterization

 Convolution sum
• 𝑦 𝑛 = σ∞
𝑘=−∞ 𝑥[𝑘] ℎ 𝑛 − 𝑘

• Above equation is called Convolution sum. It is a mathematical operation between x[n] and h[n]

• Can be seen as an operation that takes x[n] and h[n] and produces y[n]

• Convolution operator is denoted by a *

• 𝑦 𝑛 = 𝑥 𝑛 ∗ ℎ[𝑛]

• Note that convolution is a general mathematical operation that provides a discrete sequence from 2 other
discrete sequence

• For 2 general discrete time sequences 𝑥1 [𝑛] and 𝑥2 [𝑛] the convolution can be written as 𝑥3 𝑛 = 𝑥1 𝑛 ∗
𝑥2 [𝑛]
 Convolution sum
• As far as LTI systems are concerned, output y[n] of the system due to input x[n] is obtained by convolving
x[n] with impulse response h[n]

• 𝑦 𝑛 = 𝑥 𝑛 ∗ ℎ 𝑛 = σ∞
𝑘=−∞ 𝑥[𝑘] ℎ 𝑛 − 𝑘

• Find system output y[n] for the given impulse response and input signal for a LTI system?

• 𝑦 𝑛 = σ∞
𝑘=−∞ 𝑥[𝑘] ℎ 𝑛 − 𝑘 = ⋯ + x −2 ℎ 𝑛 + 2 + x −1 ℎ 𝑛 + 1 + x 0 ℎ 𝑛 + x 1 ℎ 𝑛 − 1 +

x 2 ℎ 𝑛−2 +⋯
• 𝑦 𝑛 =x 0 ℎ 𝑛 +x 1 ℎ 𝑛−1 +x 2 ℎ 𝑛−2
Numerical Problem
 Numerical Problem
• Obtain the output of the LTI system when the input signal and impulse response are given below:

• 𝑦 𝑛 = σ∞
𝑘=−∞ 𝑥[𝑘] ℎ 𝑛 − 𝑘 = ⋯ + x −2 ℎ 𝑛 + 2 + x −1 ℎ 𝑛 + 1 + x 0 ℎ 𝑛 + x 1 ℎ 𝑛 − 1 +

x 2 ℎ 𝑛−2 +⋯
• 𝑦 𝑛 = x −1 ℎ 𝑛 + 1 + x 0 ℎ 𝑛 + x 1 ℎ 𝑛 − 1
Numerical Problem

𝑦 𝑛 = x −1 ℎ 𝑛 + 1 + x 0 ℎ 𝑛 + x 1 ℎ 𝑛 − 1
Another way of interpreting convolution sum
• 𝑦 𝑛 = σ∞ ∞
𝑘=−∞ 𝑥[𝑘] ℎ 𝑛 − 𝑘 = σ𝑘=−∞ 𝑔[𝑘, 𝑛] where 𝑔[𝑘, 𝑛]= x k ℎ 𝑛 − 𝑘

• To find the output at time instance n=𝑛0 , substitute n=𝑛0 in the above equation

• 𝑦 𝑛0 = σ∞
𝑘=−∞ 𝑔[𝑘, 𝑛0 ]

• The output at time instance n=𝑛0 is obtained by substituting n=𝑛0 and then adding all values of 𝑔[𝑘, 𝑛0 ] for
k=-∞ to ∞

• We can interpret 𝑔[𝑘, 𝑛0 ] as a function of k and we have to add all values of 𝑔[𝑘, 𝑛0 ] for k=-∞ to ∞

• What is 𝑔[𝑘, 𝑛0 ]?

• 𝑔 𝑘, 𝑛0 = x k ℎ 𝑛0 − 𝑘 = x k ℎ −𝑘 + 𝑛0
Another way of interpreting convolution sum

• 𝑔 𝑘, 𝑛0 = x k ℎ 𝑛0 − 𝑘 = x k ℎ −𝑘 + 𝑛0

• What is x[k]?

• The same input signal x[n] with independent variable as k instead of n

Another way of interpreting convolution sum
• What is ℎ −𝑘 + 𝑛0 ?

• h[k] shifted by 𝑛0 and reflected

• Shift left or right?

• Depends on value of 𝑛0

• Case 1: 𝑛0 is negative (say -2)

• ℎ −𝑘 + 𝑛0 = ℎ −𝑘 − 2

• Right shift h[k] by 2 units and then do reflection.

• Case 2: 𝑛0 is positive (say 3)

• ℎ −𝑘 + 𝑛0 = ℎ −𝑘 + 3

• Left shift h[k] by 3 units and then do reflection.

Another way of interpreting convolution sum

• Case 3: 𝑛0 = 0

• ℎ −𝑘 + 𝑛0 = ℎ −𝑘

• Take h[k] and do reflection.

• Once we get x[k] and ℎ 𝑛0 − 𝑘 , find the product for each value of k.

• The product is g[k, 𝑛0 ] which is a function of k.

• Find the sum of g[k, 𝑛0 ] over various values of k. We get 𝑦 𝑛0 .

• Repeat the procedure to obtain 𝑦 𝑛 for all values of n

 Numerical Example
𝑦 𝑛 = σ∞ ∞
𝑘=−∞ 𝑥[𝑘] ℎ 𝑛 − 𝑘 = σ𝑘=−∞ 𝑔[𝑘, 𝑛]

• For n=-3, the product x[k] h[-3-k] is zero everywhere so y[-3]=0

• It can be observed that y[n]=0 for n=-4,-5,…
 Numerical Example

• Sum over k from -∞ to ∞ we get y[-2]=3

• Now let us try to calculate y[-1]

 Numerical Example

• Sum over k from -∞ to ∞ we get y[-1]=8

• Now let us try to calculate y[0]

 Numerical Example

• Sum over k from -∞ to ∞ we get y[0]=14

• Now let us try to calculate y[1]

 Numerical Example

• Sum over k from -∞ to ∞ we get y[1]=8

• Similarly, when we calculate y[2], we get y[2]=3.

• For all values of n≥3 we see that there is no overlap between x[k] and h[n-k]. So y[n]=0 for n≥3.
Numerical Example
Simplified graphical method for convolution sum

• To calculate y[-3] , we need x[k] and h[-k-3],get the product and then add over all values of k.
• To get h[-k-3] we need to first shift h[k] by 3 units to right and then reflect it.
• Can we find h[-3-k] in a single step?
Simplified graphical method for convolution sum

x[k] h[-k]

• h[-k-3] can also be obtained by shifting h[-k] to the left by 3 units

• Now y[-3] can be found by multiplying above h[-3-k] with x[k] and add the values.

• Similarly for y[2] we can find it by adding the product of h[2-k] with x[k] for all values of k.

• h[2-k] is obtained by shifting h[-k] to the right by 2 units.

Simplified graphical method for convolution sum
• Given x[n] and h[n]

• Draw x[k] by replacing n with k

• Draw h[-k] by replacing n with k and then

reflecting it

• If you want to find the output at n= 𝑛0 , shift h[-

k] by 𝑛0 units.

• If 𝑛0 is negative shift h[-k] to the left and if 𝑛0

is positive shift h[-k] to the right.

• Find x[k] h[𝑛0 -k] and sum the resultant signal

for all values of k

• Repeat the above procedure for finding y[n] for

any value of n
 Numerical Example
• Find x[n]*h[n] where x[n] and h[n] are given below:
 Numerical Example
• Step 2: Identify the value of n from where y[n] has non-zero value starts. Let us denote it as 𝑛1 . y[n]=0 for all
n< 𝑛1 . We have to find it by trial and error. Check for n=-2

• There is overlap at k=-1 and -2. Therefore y[-2]≠0.

• What about n=-3?

• The overlap will be only at k=-2. Therefore y[-3]≠0.

 Numerical Example
• But if we shift further by one more unit to the left for finding y[-4] there will be no overlap.

• So, product will be zero and hence y[-4]=0

• If you shift h[-k] further to the left, there will be no overlap. Hence y[n]=0 for n<-3

• In other words, the non-zero values of y[n] starts from n=-3 onwards.

• Next step is to find the largest value of n for which y[n] is non zero. Let us denote it as 𝑛2 . y[n]=0 for all
n>𝑛2 . We have to find it by trial and error. Check for n=3
 Numerical Example

• There is an overlap at k=1 and 2. Therefore y[3]≠0.

• What about n=4?

• There is an overlap at k=2. Therefore y[4]≠0.

 Numerical Example
• But if we shift further by one more unit to the right for finding y[5] there will be no overlap.

• So, product will be zero and hence y[5]=0

• If you shift h[-k] further to the right, there will be no overlap. Hence y[n]=0 for n>4

• In other words, the non-zero values of y[n] ends at n=4.

• Non zero values for y[n] exists for -3≤n≤4.

• Next step is to find the values of y[n] for -3≤n≤4.

Numerical Example
 Numerical Example

-4 -3 -2 -1 0
 Numerical Example

y[3]=4x1+3x1=7

1 2 3 4
 Numerical Example

y[4]=4x1=4

2 3 4 5

Similarly you will get y[-1]=6,y[0]=10, y[1]=10,y[2]=9

 Numerical Example
• For the following signals x[n] and h[n] calculate y[n]=x[n]*h[n]
Numerical Example
 Useful observations
• Given two sequences x[n] and h[n] of finite duration, can we quickly find the values of n for which the result
of convolution sum starts and stops?

• If x[n] has length N1 and h[n] has length N2, what is the length of x[n]*h[n]?

• Consider 2 sequence x[n] and h[n] where x[n] starts at n=p1 and ends at n=p2 and h[n] starts at n=q1 and ends
at n=q2.

• Therefore N1=p2-p1+1 and N2=q2-q1+1

 Useful observations
• x[n]*h[n] starts at n1=p1+q1 and ends at n2=p2+q2

• The length of the result of x[n]*h[n] will be N1+N2-1

 Convolution with infinite sequences
• Let 𝑥 𝑛 = 𝛼 𝑛 𝑢[𝑛] with 0<α<1. and ℎ 𝑛 = 𝑢 𝑛 . Find y[n]=x[n]*h[n]
 Convolution with infinite sequences

• We have 𝑝1 = 0 and 𝑞1 = 0. Hence x[n]*h[n] will start at 𝑛1 = 𝑝1 + 𝑞1 = 0

• Similarly 𝑝2 = ∞ and 𝑞2 = ∞. Hence x[n]*h[n] extends up to 𝑛2 = 𝑝2 + 𝑞2 = ∞

• For n<0, y[n]=0

• For n≥0

• 𝑦[0] = 1, 𝑦 1 = 1 + 𝛼, 𝑦 2 = 1 + 𝛼 + 𝛼 2 , 𝑦 1 = 1 + 𝛼 + 𝛼 2 +𝛼 3 , …

• 𝑦 𝑛 = 1 + 𝛼 + ⋯ + 𝛼 𝑛−1 + 𝛼 𝑛 for all n>0

• This is sum of a GP with common ratio as 𝛼

1−𝛼𝑛+1
• The resultant value will be 1−𝛼
 Convolution with infinite sequences

1−𝛼𝑛+1
• 𝑦𝑛 = for n≥0 and 0 otherwise
1−𝛼

1−𝛼𝑛+1
• We can rewrite it as 𝑦 𝑛 = 𝑢[𝑛]
1−𝛼

• Directly evaluates convolution sum without completely depending on graphical method

• Let 𝑥 𝑛 = 𝛼 𝑛 𝑢[𝑛] with 0<α<1. and ℎ 𝑛 = 𝑢 𝑛 . Find y[n]=x[n]*h[n]

AKA Direct Formula method

 Problems involving time interval
• In some problems, it may be advantageous to consider the evaluation of convolution sum separately for
different intervals of n

• In each interval, the convolution sum behaves in a specific way

• Find y[n]=x[n]*h[n]
 Problems involving time interval
• Interval 1: n<0, For n<0 we have y[n]=0

• Interval 2: 0≤n ≤2 We have y[0]=1,

y[1]=2,y[2]=3.

• The convolution sum increases by 1 as we

increase the value of n

• Interval 3: 3 ≤n ≤4 We have y[3]=3 and y[4]=3

• Convolution sum is constant in this interval

• Interval 4: 5 ≤n ≤7 We have y[5]=3 y[6]=2 and

y[7]=1

• Convolution sum is decreasing in this interval

• Interval 5: n>7 For n>7 we have y[n]=0

 Practice Problem
• Find y[n]=x[n]*h[n] for the following signals