You are on page 1of 35

Digital Signal Processing

Lecture 3 Contents (Convolution and its properties)


Follows Section 2.3 of the textbook (Proakis and
Manolakis, 4th ed.).

By
Dr. Muhammad Imran Farid
1
Lecture 3 Contents
• Review of impulse response
• Running example: computing a system's response to a signal in different
ways
• Direct computation
• Using the convolution sum: adding up shifted and scaled copies of the
impulse response
• Flipping and sliding one signal against the other
• Matlab example of flipping and sliding
• Understanding h[n-k]
• The convolution array (a fast method for convolving short signals)

2
Lecture 3 Contents
• Convolving infinite-length signals
• The sum of a finite geometric series
• Properties of convolution/LTI systems
• Commutative property
• Distributive property
• Associative property
• Causality and the impulse response
• The step response and its relationship to the impulse response
• Differential and difference equations

3
Digital Signal Processing
Lecture 3 (Convolution and its properties)
By
Dr. Muhammad Imran Farid

4
5
6
Given LTI system

• [1 -2 3] is the impulse response of the system which is the same as the coefficients of the LTI systems given. i.e.

7
8
9
Convolution GUI demo

10
Convolution GUI demo

11
Convolution GUI demo

12
Convolution GUI demo

13
Convolution GUI demo

14
Convolution GUI demo

15
Can we do shift and then flip? The answer is yes

16
EXAMPLE:

• We have to cover the whole signals. Shifting h[-k] not just to the right but to the left too, in
order to get the output

17
Example

18
Example

19
Example

• Cannot use this method when the input is infinitely long


20
• Usually we have finite impulse response. Lets for example we don’t have such case

What if: ∝ ∈ [0, 1]

∝ ∈ [0, 1)

∝ ∈ (0, 1]

∝ ∈ (0, 1)

21
• Can we do better than this expression?
• Yes! Using calculus we can. Lets see this in the next slide

22
Verify

Using this equivalent equation

Prove

෍ ∝𝑘 =∝𝑛+1 +∝𝑛+2 +∝𝑛+3 +∝𝑛+4 + ⋯


𝑘=𝑛+1
=∝𝑛+1 1 +∝ +∝2 +∝3 +∝4 + ⋯
= ∝𝑛+1 σ∞
𝑘=0 ∝
𝑘 23
• Graphically what does this output looks like ?

24
25
Asymptotically
1
reached at
1−∝

26
What if I told you to use convolution sum

GIVEN: 𝑥 𝑛 =∝𝑛 𝑢 𝑛
ℎ 𝑛 =𝑢 𝑛

All Function of k

• Same result as before using convolution sum formula but a


lot easier

27
• Real world systems are composed of many small systems and usually the output of some system can
be the input of another. These systems can be in parallel or series…..

Example of Nonlinear System

Input

• NOTE: Apart from impulse response, a system is also fully characterize by some other input, other than the
impulse response. e.g. Step Function
• If we know the response of a system from a step function and if we can decompose the new input into step
functions, then we can find the output of a system from the new input
28
flipping x instead of h

29
Graphically

30
We can put parenthesis
whenever we want

31
• Not true for nonlinear systems, as shown in the following example

• That’s why a lot of time we have impulse response looks like


something times the step function
32
• If I put LHS and RHS into the
system what I will get is

• If we know the step response then we can go back and calculate the impulse
response and from there we can calculate the response of any input

33
• Reason of studying the property of LTI is to modify your system which is composed of
many systems and combine them in such a way that one can simplify it to get the result

System whose impulse


response is step function

Signal

34
• Just some flavor of difference equations

Difference equation Called “linear constant coefficient


usually looks like this difference equation”

We solve them in the same way


we solve differential equations,
Homogeneous solution Particular solution
so that our solution be like this

• Here we only do convolution to get the answer


• Where as once we start combining the previous values of “Y” we will get the recursive solution to answer and that’s
where we will have to look for homogeneous and particular solutions, just like differential equation in analog domain
• Solving difference equations can be simplify with transform methods like Fourier Transform and Z-Transform etc.
• We will come back to difference equation when we will Insha'Allah covers the Z-Transform
35

You might also like