ECEN 314: Signals and Systems

Lecture Notes 6: Continuous-Time Convolution

Reading:
• Current: SSOW 2.2
• Next: SSOW 2.5

1 Continuous-time convolution
For DT LTI systems, we have
∞
X ∞
X
x[n] = x[k]δ[n − k] =⇒ y[n] = x[k]h[n − k]
k=−∞ k=−∞

To achieve a similar result for CT LTI systems, we need to express x(t) as a linear combi-
nation of a “basic” signal and its time shifts. The problem is that we need an uncountably
infinite number of time shifts to cover the real line. Therefore, the decomposition must be
in terms of an integral. Can we use

1, t = 0
z(t) =
0, t 6= 0.
R
The problem with this is that z(t)x(t)dt = 0.
Instead, CT convolution can be derived by combining the sifting property of δ(t) with
the linearity and time-invariance of the system. First, we observe that the sifting property
allows us to write Z ∞
x(t) = x(τ )δ(t − τ )dτ
−∞

This shows that any signal x(t) can be be written as a weighted linear combination (via
integration) of delta functions.
S
Next, we consider a linear system x(t) → y(t) and define h(t; τ ) to be the output signal
S
(i.e., function of t) from S associated with the input δ(t − τ ) (i.e., δ(t − τ ) → h(t; τ )). We
note that linearity of a system extends to integrals because integrals are defined to be limits
S
of sums. Therefore, x(t; τ ) → y(t; τ ) for all a ≤ τ ≤ b implies
Z b Z b
S
x(t; τ )dτ → y(t; τ )dτ.
a a

1
Combining this with the previous observation shows that
Z ∞ Z ∞
S
x(t) = x(τ )δ(t − τ )dτ → y(t) = x(τ )h(t; τ )dτ.
−∞ −∞

Finally, the time-invariance of the system implies that

S
δ(t − τ ) → h(t − τ ; 0)

and h(t − τ ; 0) is exactly the impulse response h(t). Therefore, we find that
Z ∞
y(t) = x(t) ∗ h(t) , x(τ )h(t − τ )dτ.
−∞

2 CT convolution via approximation

The linearity argument used above can be treated in more detail by considering the piecewise
constant approximation of x(t). For x(t) and ∆ > 0, the piecewise constant approximation
is x∆ (t) = x(k∆) when t ∈ ((k − 1/2)∆, (k + 1/2)∆]. When x(t) is continuous and ∆ → 0,
we have x∆ (t) → x(t) for all t ∈ R.

x(t)

x∆ (t)

t
−3∆ −2∆ −∆ 0 ∆ 2∆ 3∆

Can we express x∆ (t) as a linear combination of a “basic” signal and its time shifts? Yes,
consider  1 ∆ ∆

∆
, t ∈ − ,
2 2
δ∆ (t) =
0, otherwise
Note that δ∆ (t − k∆) are non-overlapping for different k and together they “fill up” the
entire time axis. We thus have
∞
X
x∆ (t) = x(k∆)δ∆ (t − k∆)∆
k=−∞

2
 δ∆ (t)

δ∆ (t + ∆) δ∆ (t − ∆)

δ∆ (t + 2∆) δ∆ (t − 2∆)
1
δ∆ (t + 3∆) δ∆ (t − 3∆)
∆

t
−3∆ −2∆ −∆ 0 ∆ 2∆ 3∆

where ∆ here is used to compensate 1/∆ in the definition of δ∆ (t).

Let h∆ (t) be the response to δ∆ (t). By the LTI properties, the output signal
∞
X
y∆ (t) = x(k∆)h∆ (t − k∆)∆
k=−∞

When ∆ → 0, let δ(t) and h(t) be the limits of δ∆ (t) and h∆ (t), respectively. Then,
∞
X Z ∞
x(t) ←− x∆ (t) = x(k∆)δ∆ (t − k∆)∆ −→ x(τ )δ(t − τ )dτ
k=−∞ −∞

and ∞
X Z ∞
y(t) ←− y∆ (t) = x(k∆)h∆ (t − k∆)∆ −→ x(τ )h(t − τ )dτ
k=−∞ −∞

Conclusion, the convolution integral is given by

Z ∞
y(t) = x(τ )h(t − τ )dτ = x(t) ∗ h(t.)
−∞

But what is h(t)? Note that δ∆ (t) → h∆ (t) so δ(t) → h(t), i.e., h(t) is the response to
δ(t). But what is δ(t) anyway? Answer: It is a singularity function and it is operationally
defined via the following properties:
1) δ(t) = 0 for all t 6= 0; and
R 0+
2) 0− δ(t)dt = 1.

3
 δ∆ (t) δ(t)

1
∆

∆
t t
∆
− 0 0
2 2

Observation: If we know the unit impulse response h(t) of a LTI system, we can compute
the output y(t) of an arbitrary input x(t) as y(t) = x(t) ∗ h(t). In this sense, a LTI system
is fully determined by its unit impulse response.

3 Properties of CT convolution
Property 1 (The sifting property). x(t)∗δ(t−t0 ) = x(t−t0 ). In particular, x(t)∗δ(t) = x(t).

Proof. By definition, we have

Z ∞
x(t) ∗ δ(t − t0 ) = x(τ )δ(t − τ − t0 )dτ
−∞
= x(t − t0 ),

where the integral is computed using the sifting proprty of δ(t).

Property 2 (The commutative property). x(t) ∗ h(t) = h(t) ∗ x(t)

Property 3 (The distributive property). x(t) ∗ (h1 (t) + h2 (t)) = x(t) ∗ h1 (t) + x(t) ∗ h2 (t)

Property 4 (The associative property). x(t) ∗ (h1 (t) ∗ h2 (t)) = (x(t) ∗ h1 (t)) ∗ h2 (t)

Property 5 (Causality). A CT LTI system is causal if and only if its unit impulse response
h(t) = 0 for all t < 0.

Property 6 (Stability).
R∞ A CT LTI system is BIBO stable if and only if its unit impulse
response satisfies −∞ |h(t)|dt < ∞.

Proofs of Properties 2–6 are completely analogous to those for DT systems and hence
are omitted from this note.

4
 4 Calculation of convolution integrals
Convolution integrals can be calculated using a similar graphical procedure as that for cal-
culating convolution sums:
Step 1: Choose a value of t and consider it fixed.
Step 2: Plot x(τ ) as a function of τ .
Step 3: Plot the function h(t − τ ) (as a function of τ ) by first flipping h(τ ) and then shifting
to the right by t (if t is negative, this means a shift to the left by |t|).
Step 4: Compute the intermediate signal wt (τ ) via pointwise multiplication of x(τ ) and h(t−τ ).
Then, integrate over the entire real line to find y(t).
Example 1. Calculate y(t) = x(t) ∗ h(t) when x(t) = h(t) = u(t) − u(t − 1).
Example 2. Consider the CT system described by the LCCDE
dy(t)
2 + y(t) = x(t)
dt
Assuming the condition of initial rest, calculate the response to the input signal x(t) = et u(t).
Since the system is LTI, let us first calculate the unit impulse response h(t) of the system
and then use the convolution integral to calculate the response y(t) to the input signal x(t) =
et u(t). To calculate the unit impulse response, let us assume that x(t) = δ(t). Since x(t) = 0
for t > 0, the particular solution yp (t) = 0 for all t ∈ R. The homogeneous solution yh (t),
as calculated previously in Lecture 3, is given by
yh (t) = Ae−t/2
Thus, the overall solution
y(t) = yp (t) + yh (t) = Ae−t/2 , ∀t > 0
To determine the value of A, we may determine the value of y(t) at t = 0+ from the condition
of initial rest as follows. Integrating the LCCDE from 0− to 0+ , we have
Z 0+ Z 0+ Z 0+
dy(t)
2 dt + y(t)dt = x(t)dt
0− dt 0− 0−
giving
1 1
2(y(0+ ) − y(0− )) = 1 =⇒ y(0+ ) = y(0− ) + =
2 2
Substituting y(0+ ) = 1/2 into the overall solution, we can determine that A = 1/2, giving
the unit impulse response
1
h(t) = e−t/2 u(t)
2
Following a similar graphical procedure to DT convolution, we can calculate the output y(t)
corresponding to the input x(t) = et u(t) as given by
1
y(t) = x(t) ∗ h(t) = (et − e−t/2 )u(t)
3

5
 x(τ )

τ
0 1

h(−τ )

τ
−1 0

h(t − τ )

τ
t−1 t

x(τ )h(t − τ ) : 0 ≤ t ≤ 1

τ
0 t

x(τ )h(t − τ ) : 1 ≤ t ≤ 2

τ
0 t−1 1

y(t)

t
0 1 2