Lecture 5

CT LTI Systems and

Convolution Integral
Representation of CT Signals
▪ According to the sifting property of Dirac delta, any signal 𝑥(𝑡) can be
expressed as a continuous sum of weighted and shifted impulses (Dirac delta)
∞
𝑥 𝑡 = න 𝑥(𝜏)𝛿(𝑡 − 𝜏)𝑑𝜏
−∞
▪ Proof:
1
0<𝑡<𝛥
▪ Let’s start with the definition 𝛿𝛥 (𝑡) = ൝𝛥 1
0 otherwise
1 0<𝑡<𝛥
▪ Then, we have 𝛥 ⋅ 𝛿𝛥 (𝑡) = ቊ
0 otherwise
▪ Now, a staircase approximation to 𝑥(𝑡) can be defined by
∞

𝑥ො 𝑡 = ෍ 𝑥(𝑘𝛥) 𝛥 𝛿𝛥 (𝑡 − 𝑘𝛥)
𝑘= −∞
𝑥 𝑘∆

𝑡
(𝑘 − 1)∆ 𝑘∆ (𝑘 + 1)∆
Representation (cont’d)
▪ Now, take the limit as 𝛥 → 0, then
– the Riemann sum becomes an integral, such that 𝛥 → 𝑑𝜏
– the discrete dummy variable 𝑘𝛥 → continuous dummy variable 𝜏
– 𝑥 𝑘𝛥 → 𝑥 𝜏 , and
– 𝛿𝛥 (𝑡 − 𝑘𝛥) → 𝛿(𝑡 − 𝜏)

▪ Therefore, we have
𝑥 𝑡 = lim𝛥→0 𝑥ො 𝑡
∞

= lim𝛥→0 ෍ 𝑥 𝑘𝛥 𝛿𝛥 𝑡 − 𝑘𝛥 𝛥
𝑘=−∞
∞
=න 𝑥(𝜏) 𝛿(𝑡 − 𝜏) 𝑑𝜏
−∞
Linear Time-Invariant Systems
▪ Let’s use linearity and time-invariance together
LTI
𝛿 𝑡 System ℎ 𝑡
ℎ 𝑡

LTI
𝛿 𝑡−𝜏 System ℎ 𝑡−𝜏
ℎ 𝑡
∞ ∞
𝑥(𝑡) = න 𝑥(𝜏)𝛿(𝑡 − 𝜏) 𝑑𝜏 LTI 𝑦(𝑡) = න 𝑥 𝜏 ℎ(𝑡 − 𝜏) 𝑑𝜏
−∞ System −∞

ℎ 𝑡
 Convolution Integral
▪ 1) Represent an arbitrary input signal x(t) as a continuum sum (integral) of
weighted and shifted impulse signals.
∞
𝑥(𝑡) = න 𝑥(𝜏)𝛿(𝑡 − 𝜏) 𝑑𝜏
−∞
(This also follows from the sifting property of the unit impulse.)

▪ 2) Use the principles of superposition and time-invariance to represent the output

of the system as the sum of responses of the system to these shifted unit impulses.
∞
𝑦(𝑡) = 𝓛 𝑥(𝑡) = 𝓛 න 𝑥(𝜏) 𝛿(𝑡 − 𝜏) 𝑑𝜏
−∞
∞
= න 𝑥(𝜏)𝓛 𝛿(𝑡 − 𝜏) 𝑑𝜏 principle of superposition
−∞
∞
= න 𝑥(𝜏) ℎ(𝑡 − 𝜏) 𝑑𝜏 principle of time−invariance
−∞
where ℎ(𝑡 − 𝜏) is the response of the system to a shifted impulse (i.e., shifted
impulse response)
Existence and Convergence
▪ Definition: A signal is absolutely integrable over 𝑎, 𝑏 if
𝑏
න 𝑥(𝑡) 𝑑𝑡 < ∞
𝑎

▪ The convolution integral is an infinite integral, so we need to worry about

its convergence.

▪ The CONVOLUTION INTEGRAL

∞
𝑦(𝑡) = 𝑥(𝑡) ∗ ℎ(𝑡) = න 𝑥(𝜏)ℎ(𝑡 − 𝜏)𝑑𝜏
−∞
converges (i.e., the result is finite) if either 𝑥(𝑡) or ℎ(𝑡) is absolutely
integrable over (−∞, ∞)
Stable LTI Systems
▪ A C-T system is stable if every bounded input produces a bounded output.
𝑥(𝑡) < 𝐵1 implies 𝑦(𝑡) < 𝐵2
▪ This condition is difficult to check, since you either need to find a counter-
example or show this condition holds for all possible bounded inputs!!
▪ Alternatively, for LTI systems, we have
∞
𝑦(𝑡) = න ℎ(𝜏)𝑥(𝑡 − 𝜏)𝑑𝜏
−∞
▪ Thus,
∞ ∞
𝑦(𝑡) = න ℎ(𝜏)𝑥(𝑡 − 𝜏)𝑑𝜏 ≤ න ℎ(𝜏) 𝑥(𝑡 − 𝜏) 𝑑𝜏
−∞ −∞
▪ Since 𝑥(𝑡) < 𝐵1 , for all t
∞
𝑦(𝑡) ≤ 𝐵1 න ℎ(𝜏) 𝑑𝜏
−∞
∞
▪ Hence, 𝑦(𝑡) < ∞ if ‫׬‬−∞ ℎ(𝜏) 𝑑𝜏 < ∞. Therefore, we conclude that
▪ An LTI system is stable iff its impulse response is absolutely integrable.
 Causal LTI Systems
▪ Memory of the LTI system:
▪ A general system is memoryless if 𝑦(𝑡) = 𝐾𝑥(𝑡)
▪ For an LTI system, this implies that: ℎ(𝑡) = 𝐾𝛿(𝑡)

▪ The output of a causal system depends only on the present and past
values of the input. Hence, the convolution integral becomes
𝑡
𝑦(𝑡) = න 𝑥(𝜏)ℎ(𝑡 − 𝜏)𝑑𝜏
−∞
▪ Now, using the change of variables, 𝛼 = 𝑡 − 𝜏
0 ∞
𝑦(𝑡) = න ℎ(𝛼)𝑥(𝑡 − 𝛼)(−𝑑𝛼) = න ℎ(𝛼)𝑥(𝑡 − 𝛼)𝑑𝛼
∞ 0
which implies that for a causal LTI system ℎ(𝑡) = 0, 𝑡 < 0

▪ Note: For a causal LTI system, if 𝑥(𝑡) = 0 for 𝑡 < 0, then

𝑡
𝑥(𝑡) ∗ ℎ(𝑡) = න 𝑥(𝜏)ℎ(𝑡 − 𝜏)𝑑𝜏
0
Commutativity
▪ Convolution integral is commutative.
𝑦(𝑡) = 𝑥(𝑡) ∗ ℎ(𝑡) = ℎ(𝑡) ∗ 𝑥(𝑡)
▪ Proof:
𝑦(𝑡) = න𝑥(𝜏)ℎ(𝑡 − 𝜏)𝑑𝜏 = 𝑥(𝑡) ∗ ℎ(𝑡)

Let 𝛼 = 𝑡 − 𝜏 ; 𝜏 = 𝑡 − 𝛼
𝑦(𝑡) = න𝑥(𝑡 − 𝛼)ℎ(𝛼)𝑑𝛼 = නℎ(𝛼)𝑥(𝑡 − 𝛼)𝑑𝛼 = ℎ(𝑡) ∗ 𝑥(𝑡)

▪ Interchange the role of the input and the impulse response

LTI LTI
𝑥 𝑡 ℎ 𝑡 𝑦 𝑡 ≡ ℎ 𝑡
𝑥 𝑡
𝑦 𝑡
Associativity
▪ Convolution integral is associative.
𝑥(𝑡) ∗ ℎ1 (𝑡) ∗ ℎ2 (𝑡) = 𝑥(𝑡) ∗ ℎ1 (𝑡) ∗ ℎ2 (𝑡) = 𝑥(𝑡) ∗ ℎ1 (𝑡) ∗ ℎ2 (𝑡)

▪ Series Connection of LTI Systems

LTI LTI
𝑥(𝑡) 𝑦(𝑡)
ℎ1 (𝑡) ℎ2 (𝑡)

LTI LTI
𝑥(𝑡) 𝑦(𝑡)
ℎ2 (𝑡) ℎ1 (𝑡)

LTI
𝑥(𝑡) 𝑦(𝑡)
ℎ1 (𝑡) ∗ ℎ2 (𝑡)
Distributivity over addition
▪ Convolution integral is distributive over addition.
𝑦(𝑡) = 𝑥(𝑡) ∗ ℎ1 (𝑡) + ℎ2 (𝑡) = 𝑥(𝑡) ∗ ℎ1 (𝑡) + 𝑥(𝑡) ∗ ℎ2 (𝑡)

▪ Parallel Connection of LTI Systems

LTI
ℎ1 (𝑡)
𝑥(𝑡) + 𝑦(𝑡)
LTI
ℎ2 (𝑡)

LTI
𝑥(𝑡) 𝑦(𝑡)
ℎ1 (𝑡) + ℎ2 (𝑡)
 More Properties
▪ 1) The area under the convolution 𝑦(𝑡) = 𝑥(𝑡) ∗ ℎ(𝑡) is equal to the
product of the areas under the two signals 𝑥(𝑡) and ℎ(𝑡).
▪ Proof:
∞ ∞ ∞ ∞ ∞
න 𝑦(𝑡)𝑑𝑡 = න න 𝑥(𝜏)ℎ(𝑡 − 𝜏)𝑑𝜏 𝑑𝑡 = න 𝑥(𝜏) න ℎ(𝑡 − 𝜏)𝑑𝑡 𝑑𝜏
−∞ −∞ −∞ −∞ −∞

▪ 2) Combining commutativity with the shift-invariance of the system

▪ If 𝑦(𝑡) = 𝑥(𝑡) ∗ ℎ(𝑡), then
∞
𝑦(𝑡 − 𝑡0 ) = 𝑥(𝑡 − 𝑡0 ) ∗ ℎ(𝑡) = න 𝑥(𝜏 − 𝑡0 )ℎ(𝑡 − 𝜏)𝑑𝜏
−∞
∞
= ℎ(𝑡) ∗ 𝑥(𝑡 − 𝑡0 ) = න ℎ 𝜏 𝑥(𝑡 − 𝑡0 − 𝜏)𝑑𝜏
−∞
▪ In words: if we shift the input 𝑥(𝑡), the output is a shifted version of 𝑦(𝑡).
More Properties (cont’d)
▪ 3) Scaling Property:
𝑡 𝑡 𝑡
If 𝑦(𝑡) = 𝑥(𝑡) ∗ ℎ(𝑡), then 𝑥 ∗ℎ = 𝑎 𝑦( )
𝑎 𝑎 𝑎

▪ 4) Derivative Property:
∞
𝑑𝛿(𝑡) 𝑑
𝑦(𝑡) = 𝑥(𝑡) ∗ = න 𝑥(𝜏) 𝛿(𝑡 − 𝜏) 𝑑𝜏
𝑑𝑡 −∞ 𝑑𝑡
interchanging the order of integration and differentiation
𝑑 ∞
= න 𝑥(𝜏)𝛿(𝑡 − 𝜏)𝑑𝜏
𝑑𝑡 −∞
𝑑
= 𝑥(𝑡)
𝑑𝑡
Evaluation: Special Examples
▪ 1) Special case: If 𝑥(𝑡) = 0, for 𝑡 < 0, then
∞ ∞
𝑥(𝑡) ∗ ℎ(𝑡) = න 𝑥(𝜏)ℎ(𝑡 − 𝜏)𝑑𝜏 = න 𝑥(𝜏)ℎ(𝑡 − 𝜏)𝑑𝜏
−∞ 0
▪ 2) Evaluate 𝑥(𝑡) ∗ 𝛿(𝑡) and 𝑥 𝑡 ∗ 𝛿 𝑡 − 𝑡0
∞ ∞
𝑥 𝑡 ∗ 𝛿 𝑡 = න 𝑥 𝜏 𝛿 𝑡 − 𝜏 𝑑𝜏 = 𝑥 𝑡 න 𝛿 𝑡 − 𝜏 𝑑𝜏 = 𝑥 𝑡
−∞ −∞
∞ ∞
𝑥 𝑡 ∗ 𝛿 𝑡 − 𝑡0 = න 𝑥 𝜏 𝛿 𝑡 − 𝑡0 − 𝜏 𝑑𝜏 = 𝑥 𝑡 − 𝑡0 න 𝛿 𝑡 − 𝑡0 − 𝜏 𝑑𝜏
−∞ −∞
= 𝑥 𝑡 − 𝑡0
▪ 3) Evaluate 𝑥(𝑡) ∗ 𝑢(𝑡)
∞
𝑥(𝑡) ∗ 𝑢(𝑡) = න 𝑥(𝜏)𝑢(𝑡 − 𝜏)𝑑𝜏
−∞
We note that 𝑢(𝑡 − 𝜏) = 1 only if 𝑡 − 𝜏 > 0 or 𝜏 < 𝑡. Therefore,
𝑡
𝑥(𝑡) ∗ 𝑢(𝑡) = න 𝑥(𝜏)𝑑𝜏
−∞
This is called an integrator.
Evaluation of Convolution Integral
▪ Calculation of 𝑥(𝑡) ∗ ℎ(𝑡) can be facilitated by its graphical interpretation
when one or both of the signals is defined in a piecewise fashion.
▪
▪ Step 1: Break the time axis into a number of intervals 𝑡𝑖−1 ≤ 𝑡 ≤ 𝑡𝑖 ,
where each interval 𝑡𝑖−1 , 𝑡𝑖 is chosen such that the product 𝑥(𝜏)ℎ(𝑡 − 𝜏)
has the same analytical form.
▪
▪ Step 2: For an arbitrary but fixed value of 𝑡 in the interval [𝑡𝑖−1 , 𝑡𝑖 ],
plot 𝑥(𝜏), ℎ(𝑡 − 𝜏) and the product 𝑥(𝜏)ℎ(𝑡 − 𝜏). Note that ℎ(𝑡 − 𝜏) is
obtained by reflecting ℎ(𝜏) about the origin ( = 0) and then
appropriately shifting the reflected signal.

▪ Step 3: Integrate 𝑥(𝜏)ℎ(𝑡 − 𝜏) over . The result is 𝑦(𝑡), a function of 𝑡

Example 1
▪ Find the unit-step response of a first-order system with time constant 𝑎.
▪ Solution:
▪ Evaluate 𝑥(𝑡) ∗ ℎ(𝑡), where ℎ(𝑡) = 𝑒 −𝑎𝑡 𝑢(𝑡), 𝑎 > 0 and 𝑥(𝑡) = 𝑢(𝑡)
∞
𝑦 𝑡 = ℎ 𝑡 ∗ 𝑥 𝑡 = ‫׬‬−∞ 𝑒 −𝑎𝜏 𝑢 𝜏 𝑢 𝑡 − 𝜏 𝑑𝜏 𝑢(𝜏)
▪ Note that
0 𝜏
1 𝜏 > 0 and 𝑡 − 𝜏 > 0
𝑢(𝜏) 𝑢(𝑡 − 𝜏) = ቊ 𝑢(−𝜏)
0 otherwise
0
▪ Also, 𝑡 > 𝜏 and  > 0 implies 𝑡 > 0. 𝑢(𝑡 − 𝜏)
Hence, we have
𝑡 𝑡 𝜏
𝑦(𝑡) = න 𝑒 −𝑎𝜏 𝑑𝜏 , 𝑡>0
0
1
= 1 − 𝑒 −𝑎𝑡 𝑢(𝑡)
𝑎
▪ This equation corresponds to charging of a capacitor connected series with
a resistor using DC input applied at 𝑡 = 0.
 Example 2
𝑥(𝜏)
▪ 𝑥(𝑡) = 𝑢(𝑡) − 𝑢(𝑡 − 𝑇)
▪ ℎ 𝑡 = 𝑡 𝑢 𝑡 − 𝑢 𝑡 − 2𝑇
▪ We can express in the dummy variable 𝜏 𝜏
1 0<𝜏<𝑇
𝑥(𝜏) = ቊ h(𝜏)
0 otherwise

𝜏 0 < 𝜏 < 2𝑇
ℎ(𝜏) = ቊ 𝜏
0 otherwise

▪ Interval 1 𝑡<0
𝑦(𝑡) = 0 since 𝑥(𝜏) and ℎ(𝑡 − 𝜏) do not overlap for 𝑡 < 0.
𝑥(𝜏)

h(𝑡 − 𝜏) 𝜏

𝜏
𝑡 − 2𝑇 𝑡
Example 2 (cont’d)
▪ Interval 2 0<𝑡<𝑇
𝑡−𝜏 0 < 𝑡 − 𝜏 < 2𝑇 𝑡−𝜏
𝑡 − 2𝑇 < 𝜏 < 𝑡
ℎ 𝑡−𝜏 =൜ =ቊ
0 otherwise 0 otherwise
∞ 𝑡 𝑡
1 2 1 2
𝑦(𝑡) = න 𝑥(𝜏)ℎ(𝑡 − 𝜏)𝑑𝜏 = න (𝑡 − 𝜏)𝑑𝜏 = 𝑡𝜏 − 𝜏 = 𝑡
−∞ 0 2 0
2
𝑥(𝜏)

h(𝑡 − 𝜏)

𝜏
𝑡 − 2𝑇 0 𝑡
Example 2 (cont’d)
▪ Interval 3 𝑇 < 𝑡 < 2𝑇
∞ 𝑇
1 2
𝑦(𝑡) = න 𝑥(𝜏)ℎ(𝑡 − 𝜏)𝑑𝜏 = න (𝑡 − 𝜏)𝑑𝜏 = 𝑇𝑡 − 𝑇
−∞ 0 2

𝑥(𝜏)

𝜏
Example 2 (cont’d)
▪ Interval 4 2𝑇 < 𝑡 < 3𝑇
𝑇 1 3
𝑦(𝑡) = න (𝑡 − 𝜏)𝑑𝜏 = − 𝑡 2 + 𝑇𝑡 + 𝑇 2
𝑡−2𝑇 2 2

𝑥(𝜏)

h(𝑡 − 𝜏)

𝜏
0 𝑡 − 2𝑇 𝑡
Example 2 (cont’d) 𝑥(𝜏)
▪ Interval 5 𝑡 > 3𝑇
𝑦(𝑡) = 0

h(𝑡 − 𝜏)

0 𝑡 − 2𝑇 𝑡

▪ Final Solution:
0 𝑡<0
1 2
𝑡 0 < t < T
2
1
𝑦(𝑡) = 𝑇𝑡 − 𝑇 2 T < t < 2T
2
1 2 3 2
− 𝑡 + 𝑇𝑡 + 𝑇 2T < t < 3T
2 2
0 𝑡 > 3T
Try Yourself
▪ Convolve
𝑥(𝑡)
2

𝑡
-1 0 2

and
ℎ 𝑡 = 𝑢 𝑡 + 3 − 𝑢(𝑡 − 1)