1

Notes 04 largely plagiarized by %khc

1 Convolution Recap
Some tricks:
 x(t)   (t) = x(t)
 x(t)   (t , t0) = x(t , t0)

R
x(t)  u(t) = ,1 t
x( )d
 x(t)  ˙(t) = ẋ(t)
This then tells us that an integrator has impulse response h(t) = u(t), and that a differentiator has impulse response
h(t) = ˙(t).
Convolution is associative and commutative. Convolution also distributes over addition.
Exercise Prove it.
Convolution with h(t) is both linear and time invariant.

[x1(t) + x2 (t)]  h(t) = [x1(t)]  h(t) + [x2 (t)]  h(t)

= [x1(t)  h(t)] +  [x2 (t)  h(t)]
= y1 (t) + y2 (t)
x(t , t0 )  h(t) = [x(t)   (t , t0 )]  h(t)
= [x(t)  h(t)]   (t , t0 )
= y(t)   (t , t0 )
= y(t , t0)
This should not surprise you. If y(t) is the output of some LTI system with input x(t), then y(t) = x(t)  h(t).

2 More Convolution Fun

We now dig into the old ee120 archives and drag out some old problems to illustrate convolution. [Sketches of the
output for each of the parts are available. Please see Figure 1.]
P1
A. An LTI system has impulse response h(t) = e,t=2 u(t), input x(t) = x1 (t) n=,1  (t , n), and output y(t). If
x1(t) = u(t + 0:01) , u(t , 3:01), determine the output.
First determine x(t). Notice that x(t) is just an impulse train multiplied by a pulse from t = ,0:01 to t = 3:01.
This just picks out the four impulses at t = 0, t = 1, t = 2, and t = 3. Then use superposition. Convolution with
shifted impulses gives four shifted copies of h(t). So y(t) = e,t=2 u(t) + e,(t,1)=2u(t , 1) + e,(t,2)=2 u(t , 2) +
e,(t,3)=2u(t , 3).

B. x(t) = e,2tu(t). h(t) = 2 (t , 1) +  (t + 1).

Same tricks as above. y(t) = 2e,2(t,1)u(t , 1) + e,2(t+1)u(t + 1).

x(t) = 1
P
C. n=,1  (t , 0:01n). h(t) = Π(200t).
x(t) is a train of impulses, each separated from the next by 100 1
. h(t) is a pulse of height 1 from t = , 4001
to
t = 400 . Convolving h(t) with an impulse centered at t = t0 gives h(t , t0 ). Since there are infinitely many shifted
1

impulses, there P are infinitely many copies of h(t), whose centers are separated from that of their nearest neighbors by
1
n=,1 Π[200(t , 100 )].
1
. So y (t ) =
n
100
EE120: Signals and Systems; v5.0.0 2

x (t) impulse train x(t) y(t)

1

... ...
t t t
−0.01 3.01 t
1 2 3
(a) waveforms for A.

y(t) y(t) y(t)

−4
2+e 1
... ... 1/e
1
t t
t
1 2
−1 1 1/100 1/200

(b) output for B. (c) output for C. (d) output for D.

y(t) y(t)

1 1/2π
t
t
1 2
−1/2 1/2 7/2 9/2

(e) output for E. (f) output for F.

Figure 1: Sketches of the example problems. Knowing what the output looks like is definitely as good as having a
mathematical formula.

D. x(t) = e,t u(t). h(t) = r(t , 1)Π(t , 32 ).

Let’s choose x(t) to flip and shift. There are three places where the integral is different.
For t < 1, there is no overlap, so y(t) = 0.
For 1 < t < 2,
Zt
y ( t) = e,(t, ) ( , 1)d
1
Zt
= e,t e ( , 1)d
1
= e,t (e , 2e )j1t
= t , 2 + e1,t
For t > 2,
Z
e,(t, ) ( , 1)d
2
y(t) =
1
Z
e,t
2
= e ( , 1)d
1
= e,t (e , 2e )j21
= e,(t,1)

E. x(t) = Π(t). h(t) = Π( t,4 2 ).

You can do flip and shift, make the pulses into sums of unit steps and convolve, or just use your intuition.
EE120: Signals and Systems; v5.0.0 3

8> 0 for t < , 12

< for , 12 < t <
y(t) = > t1 +
1 1
2 2
for 12 < t < 72
: ,t 9
for 72 < t < 92
2

F. x(t) = u(t , 1). h(t) = cos(2t)u(t).

Convolve u(t) with h(t). This gives y(t) = 21 sin(2t)u(t). Now, since the unit step is actually delayed by 1,
delay y(t) by 1. So y(t) = 21 sin[2(t , 1)]u(t , 1).

3 System Interconnections
There are three major ways of putting systems together. Check out Figure 2 and see if the equivalents make sense to
you [two of them already should, since you saw them on ps2, problem 6]. The reason why we haven’t fully talked about
the feedback configuration is that the analysis becomes much easier when we hit the Fourier and Laplace transforms.
Wait 1.5 months.

g(t) h(t) g(t) h(t)

(a) two systems in series; (b) the equivalent of the series connection in (a);
impulse responses are g(t) and h(t) note the poor convolution sign

g(t)
+ g(t) h(t)
h(t)

(c) two systems in parallel (d) the equivalent of the parallel connection in (c)

+ g(t)

h(t)

(e) two systems in feedback connection (g) unfortunately, we’re not prepared to talk
about the equivalent of the feedback connection
yet in a simple fashion, but we will...

Figure 2: The three major ways we have of composing systems.

4 Eigenfunctions
Consider an LTI system with input x(t), impulse response h(t), and output y(t). What function can we put into the
system so that we will get the same function out, scaled by a constant? Such functions are called eigenfunctions and
their associated constants are called eigenvalues. In symbols, for a system performing operation H on its input, we
have H [f (
)] = f (
), where f (
) is the eigenfunction and  is its eigenvalue.1
Let’s try x(t) = ej!t. Then

y(t) = x(t)  h(t)

1 Eigen is German for self, i think. Well, on good days, that is.
EE120: Signals and Systems; v5.0.0 4

Z1
= x( )h(t ,  )d
Z,1
1
= x(t ,  )h( )d
Z,1
1
= ej!(t, ) h( )d
,1Z
1
= ej!t e,j! h( )d
,1
= ej!tH (!)
where H (!) is defined as
R1 ,j! ) h( )d .
,1 e
Note that H (!) could be complex. That means that we can also write:

y(t) = jH (!)jej (!t+6 H (!))

What if we tried a cosine? If we assume that the system performs operation H on its input:

y(t) = H [x(t)]
= H [cos(!t)]
H [ (ej!t + e,j!t)]
1
=
2
H [ej!t] + H [e,j!t]
1 1
=
2 2
j H (!)je +j 6 H (!) + jH (,!)je,j!t+j 6 H (,!)
1 j!t 1
=
2 2
6=  cos(!t)
Oh well. Because there is no guarantee that jH (!)j is even and 6 (H (!)) is odd, we have to delete cosine from the list
of candidates. Similar reasoning allows us to delete sine.
Exercise Prove that sin(!t) is not an eigenfunction.
Anyway, we still have an important result. If you gave me a system with impulse response h(t) and input x(t)
and told me to find y(t), i could always convolve and give you an answer. But since ej!t is an eigenfunction for the
convolution operator, if i can
1. represent the input x(t) as the sum of complex exponentials
2. determine H (!) for the impulse response h(t)
then i can give you the output y(t) as the sum of complex exponentials, scaled by H at the appropriate values of !.

X
That is, if
1
x(t) = Xn ejn! t
0

n=,1
then
X
1
y(t) = Xn H (n!0 )ejn! t 0

n=,1
X
1
= Xn jH (n!0)jej (n! t+6 0 H (n!0 ))
n=,1

Notice that i didn’t have to do any convolution. What a feature.

EE120: Signals and Systems; v5.0.0 5

5 Sinusoidal Steady-State Response of Real-World LTI Systems

If h(t) is real, then we can definitely say some useful things about cosine and sine being inputs into LTI systems.
If h(t) is real, then h(t) = h (t). So
Z1
H  (!) = [ h(t)e,j!tdt]
Z ,1
1
= h (t)ej!t dt
Z,1
1
= h(t)e,j (,!)tdt
,1
= H (,!)
Rewriting H  (!) and H (,!) in polar form gives:

H  (!) = [jH (!)jej 6 H (!) ]

= jH (!)je,j 6 H (!)
H (,!) = jH (,!)jej 6 H (,!)
If we equate the two polar forms,

jH (,!)j = jH (!)j
6 H (,!) = ,6 H (!)
In other words, for a real h(t) the magnitude of the frequency response is even and the phase of the frequency response
is odd.
Let’s now go back to trying a cosine as an input into an LTI system.

y(t) = H [cos(!t)]
=
1
2
jH (!)jej!t+j 6 H (!) + 12 jH (,!)je,j!t+j 6 H (,!)
=
1
2
jH (!)jej!t+j 6 H (!) + 12 jH (!)je,j!t,j 6 H (!)
= jH (!)j cos[!t + 6 H (!)]
So if our impulse response is real, then a cosine as input comes back out still looking like a cosine, but its amplitude
is scaled by the magnitude of the frequency response, and its phase is shifted by the phase of the frequency response.
But all real-world systems are real, so this should work on any system i care to take down to the lab and throw cosines
into. In fact, this gives me a good way to figure out what H (!) is; i can generate all sorts of cosines with amplitude 1
at various frequencies and record the output cosines’ amplitude and phase shift to construct H (!). This procedure is
actually used in the real world.
Exercise Prove that if h(t) is real, then sine as input gives a sine as output.

6 A Look Ahead
Fourier series you have already seen in differential equations. We’re going to use it as a stepping stone into the Fourier
transform. After developing the FT, we’ll find out that we have a shortcut to convolution. This will be exceedingly
cool.