Notes 03
Notes 03
Notes 03
x[n]
x[0] δ [n] x[1] δ [n−1] x[2] δ [n−2]
... ... ... ... ... ... ... ...
n = n + n + n
(a) any signal can be decomposed into the sum of scaled and shifted impulses
δ [n] h[n]
... ... ... ...
n n
(c) scaling and shifting the impulse produces a scaled and shifted impulse response
x[n] y[n]
(d) superposition of the three signals on the left from (c) gives x[n];
likewise, superposition of the three signals on the right gives y[n];
so if x[n] is input into our system with impulse response h[n],
the corresponding output is y[n]
we have decomposed x[n] into the sum of x[0] [n], x[1] [n , 1], and x[2] [n , 2]. In general, any x[n] can be broken
up into the sum of x[k] [n , k], where x[k] is the appropriate scaling for an impulse [n , k] that is centered at n = k.
EE120: Signals and Systems; v5.0.0 2
X
In other words, we have:
1
x[n] = x[k] [n , k]
k=,1
Next, we are given h[n], the response of the system to an impulse centered at n = 0. In other words, if [n] is
the input into this system, the output is h[n]. Utilizing the homogeneity property of a linear system, if x[0] [n] is the
input, the output is x[0]h[n]. If x[1] [n , 1] is the input, the output is x[0]h[n , 1] by both the homogeneity and
time-invariant properties of H . A similar argument can be made for x[2] [n , 2] as input producing x[2]h[n , 2] as
output. This is illustrated in Figure 1(c).
Finally, the superposition property of a linear system permits us to add up the outputs to form the composite output
to the x[n] from Figure 1(a). The final output y[n] is shown in Figure 1(d). This final output is just:
X
1
y[n] = x[k]h[n , k]
k=,1
2 Superposition Integral
Assume we have a linear system which performs the operation H on its input x(t), giving the output y(t). In symbols,
y(t) H [x(t)]
=
Z1
= H[ x( ) (t , )d ]
Z1 ,1
= H [x( ) (t , )]d
Z1,1
= x( )H [ (t , )]d
,1
where we have used the sifting integral to substitute for x(t) in the second line, the additivity property of linearity
[x1 (t) + x2 (t) ! y1 (t) + y2 (t)] in the third, and the homogeneity property of linearity [x(t) ! y(t)] in the fourth.
What is H [ (t , )]? Well, first we’d better ask what (t , ) is; it’s an impulse function centered at t = .
H [ (t , )] is then the response of the system H at time t to an impulse centered at time . Let’s denote H [ (t , )]
by g(t; ). Note that g is a function of two variables. It’s a function of t because the response of the system is going
to be a function of time. It’s a function of , because the system may be time varying, so i need to know where the
impulse is centered. If the system is time varying, the response of the system to an impulse centered at t = t1 will
in general be different from the response of the system to an impulse centered at t = t2 . Anyway, the output of the
system y(t) to any input x(t) becomes Z1
y(t) = x( )g(t; )d
,1
But if the system is time-invariant, we can further massage g(t; ) into something more palatable. Consider a time
invariant system, which responds to an impulse centered at t = a as in Figure 2(b); this system response is denoted
g(t; a). Now, because the system is time invariant, if i shift the impulse by ∆, the system response should also shift by
∆, as in Figure 2(d). This new system response is denoted g(t , ∆; a , ∆). Why the second argument shifts is easy to
see; the second argument tells us where the impulse is centered, and since the center has been shifted by ∆, the new
center is a , ∆. Big deal. What about the first argument? The best way to see this is to note what happens to the point
(p; q ); it gets shifted by ∆ also. Now let’s make a = ∆. In other words, we shift the impulse to the origin by a units
and the system response gets shifted, by a total of a units, as in Figure 2(f).
So the output of the system y(t) to any input x(t) becomes
Z1
y(t) = x( )g(t , ; 0)d
,1
EE120: Signals and Systems; v5.0.0 3
δ (t−a) g(t,a)
q
t t
a b p
(a) (b)
δ (t) g(t−a,0)
q
t t
b−a p−a
(e) (f)
Figure 2: (a) impulse centered at t = a, (b) response of system to impulse centered at t = a, (c) impulse centered at
t = a , ∆, (d) response of system to impulse centered at t = a , ∆, (e) impulse centered at t = 0, (f) response of
system to impulse centered at t = 0. We assume that the system is time invariant, so that (c) and (d) are simply (a) and
(b) shifted. For the particular choice of a = ∆, we obtain (e) and (f) from (c) and (d). Note what happens to the point
(p; q ) in (b), (d), and (f).
We can then define a new impulse response h(t , ) = g(t , ; 0) and finally we have the superposition integral as
seen in lecture: 1 Z
y(t) = x( )h(t , )d
,1
We then note that the superposition integral conveniently corresponds to the convolution operation, as previously
seen in math classes. So, to summarize:
y(t)x(t) h(t)
Z1 =
5 An Example
Of course, theory is nice, but if you can’t apply it, it is not terribly useful.
This example is worked using a cookbook approach to convolution. There are more insightful ways of doing this
example. One method is hinted at in the exercise at the end of this section.
Anyway, the cookbook approach for LTI systems only:
1. Set up x( ).
2. Set up h(t , ) by flipping.
3. Determine regions of integration by dragging h(t , ).
4. Do the integration of the product of x( ) and h(t , ).
The choice of which function is x and and which is h is arbitrary. Since convolution is commutative, you’ll
probably want to choose the simpler one (such as the one symmetric about t = 0) to flip and shift. Too bad i haven’t
done that in my example (note also that i have used t1 instead of – it made creating the figure slightly easier).
To generate x( ), use in place of t.
To generate h(t , ) is slightly harder. First, create h( ) from h(t) by replacing t with . Then make h(, ) by
flipping h( ) about = 0. Finally, realize h(t , ) by relabeling the points on the axis; = a becomes = t + a.
Note that by increasing t, h(t , ) moves to the right.
At this point in time, reread the previous paragraph and try to figure out why you relabel the point = a as
= t + a (hint: draw h(0 , ) and h(1 , ), and then generalize). Try not to confuse with t. The difference between
t and is the index into h.
Determining the different regions of integration is the most difficult step. We drag h(t , ) over x( ) by changing
the value of t (not ). We then look for values of t where:
both h(t , ) and x( ) are nonzero (if one of them were zero, their product would be zero, and the integral
would be trivial).
the limits on the integral will change.
EE120: Signals and Systems; v5.0.0 5
x(t1) h(0−t1)
2 4 2 2
t=−1: t=3:
t1 t1 t1 t1
2 −2 2 2
h(1−t1)
4 2 2
t=0: t=4:
t1 t1 t1
−1 2 2
h(t−t1)
4 2
t=1:
t1 t1
t−2 t 2
In my example, h(t , ) and x( ) are both nonzero only between t = ,2 and t = 4. We have the following
regions of integration: 8> 0 if t < ,2
>> R t x( )h(t , )d if ,2 < t < 0
< R, 2
y(t) = > tt, x( )h(t , )d if 0 < t < 2
>: Rt, x( )h(t , )d
2
if 2 < t < 4
2
2
0 if t > 4
For ,2 < t < 0:
Zt
4( + 2) d = 2 2 + 8 jt,2
,2
= 2t2 + 8t , (,8)
= 2(t + 2)2
Does this make sense? Well, the product of x and h increases from 0 at t = ,2 to a maximum value at t = 1 and
then falls back to zero at t = 4. Note that there is no discontinuity at t = ,2, t = 0, t = 2, or t = 4. In general, if
your x and h are piecewise continuous, y should also be continuous.
Exercise Redo this example. Then try it for the same x(t), but with h(t) = 4 for only ,1 < t < 2, zero elsewhere.
Exercise Redo this example. However, consider the pulse as the sum of two steps, and the triangle as the sum of three
ramps. Do only one convolution of a step and a ramp, and then use linearity and time invariance to construct the whole
solution.
6 A Look Ahead
More convolution fun follows. If you are uncomfortable with how to do it, try working some of the examples in the
textbook or asking any of us for help.