Chapter Five: Time-Domain System Analysis

• The essential goal in designing systems is that they respond in the right way.
• Therefore, we must be able to calculate the response of a system to any arbitrary input signal.
• We have already seen how to find the response of a system described by differential or difference
equations by finding the total solution of the equations with boundary conditions.
• In this chapter we will develop another technique called convolution.
• We will show that, for an LTI system, if we know its response to a unit impulse occurring at 𝒕 = 𝟎
or 𝒏 = 𝟎, that response completely characterizes the system and allows us to find the response to
any input signal.

5.2 CONTINUOUS TIME

IMPULSE RESPONSE
• We have seen techniques for finding the solutions to differential equations that describe systems.
• The total solution is the sum of the homogeneous and particular solutions.
• The homogeneous solution is a linear combination of eigenfunctions. The particular solution depends
on the form of the forcing function.
• Although these methods work, there is a more systematic way of finding how systems respond to
input signals and it lends insight into important system properties. It is called convolution.
• The convolution technique for finding the response of a continuous-time LTI system is based on a
simple idea. If we can find a way of expressing a signal as a linear combination of simple functions,
we can, using the principles of linearity and time invariance, find the response to that signal as a
linear combination of the responses to those simple functions. If we can find the response of an LTI
system to a unit-impulse occurring at 𝒕 = 𝟎 and if we can express the input signal as a linear
combination of impulses, we can find the response to it.
• Therefore, use of the convolution technique begins with the assumption that the response to a unit
impulse occurring at 𝒕 = 𝟎 has already been found. We will call that response 𝒉(𝒕) the impulse
response.

CH05 - 1
CONTINUOUS-TIME CONVOLUTION
Derivation

• Once the impulse response of a system is known, we can develop a method for finding its response
to a general input signal.
• Let a system be excited by an arbitrary input signal 𝒙(𝒕). How could we find the response?
• We could find an approximate response by approximating this signal as a sequence of contiguous
rectangular pulses, all of the same width 𝑻𝒑.

An arbitrary signal Contiguous-pulse approximation to an arbitrary signal

• Now we can (approximately) find the response to the original signal as the sum of the responses to
all those pulses, acting individually.
• Since all the pulses are rectangular and the same width, the only differences between pulses are
when they occur and how tall they are.

• Let the response to a single pulse of width 𝑻𝑷 and unit area centered at 𝒕 = 𝟎 be a function 𝒉𝒑 (𝒕)
called the unit-pulse response. The unit pulse is (𝟏/𝑻𝑷 ) 𝒓𝒆𝒄𝒕(𝟏/𝑻𝑷 ).

CH05 - 2
• Due to linearity and time invariance, the response to each of these pulses must be the unit pulse
response 𝒉𝒑 (𝒕), amplitude scaled by the factor 𝑻𝒑 𝒙(𝒏𝑻𝒑 ) and time shifted from the time origin
the same amount as the pulse. Then the approximation to the response is:

• As 𝑻𝒑 approaches zero, the summation becomes an integral and the approximations become exact
(Riemann sum / integral), and the unit pulse (𝟏/𝑻𝑷 ) 𝒓𝒆𝒄𝒕(𝟏/𝑻𝑷 ) approaches a unit impulse.

• The unit-pulse response 𝒉𝒑 (𝒕) approaches the unit-impulse response 𝒉(𝒕) (more commonly called
just the impulse response) of the system. The integral is called the convolution integral. The
convolution of two functions is conventionally indicated by the operator *.

CH05 - 3
• Note that
❖ Convolution is a mathematical operation that combines two signals to produce a third one.
In the context of systems, it represents how an input signal is transformed by a system to
produce the output signal.
❖ Convolution describes how the system’s impulse response shapes or modifies the input signal
over time.
❖ We "slide" the impulse response across the input signal, multiplying and summing the
overlapping values at each point in time to get the final output.

Graphical and Analytical Examples of Convolution

CH05 - 4
• Extra examples on graphical convolution analysis can be found in the text book pages 173-176 and
page 178

Convolution Properties
• An operation that appears frequently in signal and system analysis is the convolution of a signal with
an impulse.
∞

𝒙(𝒕) ∗ 𝑨𝜹(𝒕 − 𝒕𝟎 ) = ∫ 𝒙(𝝉) 𝑨𝜹(𝒕 − 𝝉 − 𝒕𝟎 )𝒅𝝉.

−∞

• We can use the sampling property of the impulse to evaluate the integral. The variable of integration
is 𝝉. The impulse occurs in 𝝉 where 𝒕 − 𝝉 − 𝒕𝟎 = 𝟎 or 𝝉 = 𝒕 − 𝒕𝟎 . Therefore:

𝒙(𝒕) ∗ 𝑨 𝜹(𝒕 − 𝒕𝟎 ) = 𝑨 𝒙(𝒕 − 𝒕𝟎 ) .

• Note that:
❖ The impulse response 𝒉(𝒕) characterizes a Linear Time-Invariant (LTI) system.

❖ Convolution of 𝜹(𝒕)with 𝒉(𝒕 )gives 𝒉(𝒕 ) — this is how we define the impulse response of
a system.

𝜹(𝒕) ∗ 𝒉(𝒕 ) = 𝒉(𝒕)

❖ This means the system's response to an impulse input is its impulse response, and by linearity
and time-invariance, the output to any input can be derived via convolution with 𝒉(𝒕).

𝜹(𝒕) 𝒉(𝒕 ) 𝒉(𝒕)

• If we define a function 𝒈(𝒕) = 𝒈𝟎 (𝒕) ∗ 𝜹(𝒕), then a time-shifted version 𝒈(𝒕 − 𝒕𝟎 ) can be
expressed in either of the two alternate forms:

𝒈(𝒕 − 𝒕𝟎 ) = 𝒈𝟎 (𝒕 − 𝒕𝟎 ) ∗ 𝜹(𝒕) 𝒐𝒓 𝒈(𝒕 − 𝒕𝟎 ) = 𝒈𝟎 (𝒕) ∗ 𝜹(𝒕 − 𝒕𝟎 )

• Also, 𝒈𝟎 (𝒕 − 𝒕𝟎 ) ∗ 𝜹(𝒕 − 𝒕𝟎 ) = 𝒈(𝒕 − 𝟐 𝒕𝟎 ). This property is true not only when
convolving with impulses, but with any functions.

CH05 - 5
• The commutativity, associativity, distributivity, differentiation, area and scaling properties of the
convolution integral are summarized below.

• Note that:
❖ The commutative, associative, and distributive properties of convolution in linear time-
invariant systems are similar to those of algebraic multiplication.
❖ The differentiation property of convolution extends to second-order (or higher-order):

𝒅𝟐 𝒅 𝒅
𝒚(𝒕) = 𝒙(𝒕) ∗ 𝒉(𝒕)
𝒅𝒕𝟐 𝒅𝒕 𝒅𝒕
𝒅𝟐 𝒅𝟐 𝒅𝟐 𝒅𝟐
𝒚(𝒕) = 𝟐 𝒙(𝒕) ∗ 𝒉(𝒕) ≡ 𝒚(𝒕) = 𝒙(𝒕) ∗ 𝟐 𝒉(𝒕)
𝒅𝒕𝟐 𝒅𝒕 𝒅𝒕𝟐 𝒅𝒕

1
❖ In the scaling property of convolution, both input and system functions must be scaled in time
first before performing convolution; the resulting output is multiplied by the factor |𝒂|.
2 3

∞
• Let the convolution of 𝒙(𝒕) with 𝒉(𝒕) be 𝒚(𝒕) = ∫−∞ 𝒙(𝒕 − 𝝉)𝒉(𝝉)𝒅𝝉 . Let 𝒙(𝒕) be bounded.
Then |𝒙(𝒕 − 𝝉)| < 𝑩, for all 𝝉 where 𝑩 is a finite upper bound. The magnitude of the
convolution integral is:
∞
|𝒚(𝒕)| = |∫−∞ 𝒙(𝒕 − 𝝉)𝒉(𝝉)𝒅𝝉|

CH05 - 6
• Using the principles that the magnitude of an integral of a function is less than or equal to the integral
𝜷 𝜷
of the magnitude of the function | ∫𝜶 𝒈(𝒙)𝒅𝒙 | ≤ ∫𝜶 |𝒈(𝒙)|𝒅𝒙
• And that the magnitude of a product of two functions is equal to the product of their magnitudes,
|𝒈(𝒙)𝒉(𝒙)| = |𝒈(𝒙)||𝒉(𝒙)| , we can conclude that:
∞

|𝒚(𝒕)| ≤ ∫ |𝒙(𝒕 − 𝝉)||𝒉(𝝉)| 𝒅𝝉.

−∞

System Connections
• Two very common connections of systems are the cascade connection and the parallel connection.
• Using the associativity property of convolution we can show that the cascade connection of two
systems can be considered as one system whose impulse response is the convolution of the
impulse responses of the two systems.

• Using the distributivity property of convolution we can show that the parallel connection of two
systems can be considered as one system whose impulse response is the sum of the impulse
responses of the two systems.

CH05 - 7
Step Response and Impulse Response
• In actual system testing, a system is often tested using some standard signals that are easy to
generate and do not drive the system into a nonlinearity. One of the most common signals of this
type is the step function.
• The response of an LTI system to a unit step is:
∞ ∞

𝒉−𝟏 (𝒕) = 𝒉(𝒕) ∗ 𝒖(𝒕) = ∫ 𝒉(𝝉) 𝒖(𝒕 − 𝝉)𝒅𝝉 = ∫ 𝒉(𝝉)𝒅𝝉

−∞ −∞

• This proves that the response of an LTI system excited by a unit step is the integral of the impulse
response. Therefore, we can say that just as the unit step is the integral of the impulse, the unit-
step response is the integral of the unit-impulse response.

CH05 - 8
Stability and Impulse Response
• Stability was generally defined in Chapter 4 by saying that a stable system has a bounded output
signal in response to any bounded input signal. We can now find a way to determine whether a
system is stable by examining its impulse response.

• The response 𝒚(𝒕) of a system to 𝒙(𝒕) is 𝒚(𝒕) = 𝒙(𝒕) ∗ 𝒉(𝒕). Then, if 𝒙(𝒕) is bounded we
can say that 𝒚(𝒕) is 𝒃𝒐𝒖𝒏𝒅𝒆𝒅 𝒊𝒇 𝒉(𝒕) 𝒊𝒔 𝒂𝒃𝒔𝒐𝒍𝒖𝒕𝒆𝒍𝒚 𝒊𝒏𝒕𝒆𝒈𝒓𝒂𝒃𝒍𝒆. That is, if:
∞

∫ |𝒉(𝒕)|𝒅𝒕 𝒊𝒔 𝒇𝒊𝒏𝒊𝒕𝒆 ≡ 𝒊𝒏𝒕𝒆𝒈𝒓𝒂𝒃𝒍𝒆 ≡ 𝒃𝒐𝒖𝒏𝒅𝒆𝒅

−∞

• A continuous-time system is BIBO stable if its impulse response is absolutely integrable.

Complex Exponential Excitation and the Transfer Function

• Let a stable LTI system be described by a differential equation of the form:
𝑁 𝑀

∑ 𝑎𝑘 𝑦 (𝑘) (𝑡) = ∑ 𝑏𝑘 𝑥 (𝑘) (𝑡)

𝑘=0 𝑘=0

Where
1. The 𝒂’𝒔 and 𝒃’𝒔 are constants, and
2. The notation 𝒙(𝒌) (𝒕) means the 𝒌𝒕𝒉 derivative of 𝒙(𝒕) with respect to time and, if 𝒌 is negative,
that indicates integration instead of differentiation.

• Then let the excitation be in the form of a complex exponential 𝒙(𝒕) = 𝑿𝒆𝒔𝒕 where 𝑿 and 𝒔 are,
in general, complex valued.
• This description of the excitation is valid for all time.
• The solution of the differential equation is the sum of the homogeneous and particular solutions.
• The system is stable, so the eigenvalues have negative real parts and the homogeneous solution
approaches zero as time passes.
• The system has been operating with this excitation for a semi-infinite time, so the homogeneous
solution has decayed to zero and the total solution now is the particular solution.

CH05 - 9
• The response 𝒚(𝒕) must have the form 𝒚(𝒕) = 𝒀𝒆𝒔𝒕 where 𝒀 is a complex constant. Then, in
the differential equation, the 𝒌𝒕𝒉 derivatives take the forms 𝒙(𝒌) (𝒕) =
𝒔𝒌 𝑿𝒆𝒔𝒕 and 𝒚(𝒌) (𝒕) = 𝒔𝒌 𝒀𝒆𝒔𝒕
𝑵 𝑴

∑ 𝒂𝒌 𝒔𝒌 𝒀𝒆𝒔𝒕 = ∑ 𝒃𝒌 𝒔𝒌 𝑿𝒆𝒔𝒕
𝒌=𝟎 𝒌=𝟎

• The equation is no longer a differential equation with real coefficients. It is now an algebraic equation
with complex coefficients. The factors 𝑿𝒆𝒔𝒕 and 𝒀𝒆𝒔𝒕 can be factored out, leading to:
𝑵 𝑴
𝒔𝒕 𝒌
𝒀𝒆 ∑ 𝒂𝒌 𝒔 = 𝑿𝒆 ∑ 𝒃𝒌 𝒔𝒌
𝒔𝒕

𝒌=𝟎 𝒌=𝟎

• The ratio of the response to the excitation is then:

𝒀𝒆𝒔𝒕 𝒀 ∑𝑴
𝒌=𝟎 𝒃𝒌 𝒔
𝒌
= =
𝑿𝒆𝒔𝒕 𝑿 ∑𝑵
𝒌=𝟎 𝒂𝒌 𝒔
𝒌

• A ratio of two polynomials in s called a rational function. This is the system’s transfer function.

• The response is therefore 𝒀𝒆𝒔𝒕 = 𝐇(𝐬)𝑿𝒆𝒔𝒕 𝒐𝒓 𝒚(𝒕) = 𝐇(𝐬) 𝒙(𝒕).

PROBLEMS
1. Graph these functions.

a. 𝒈 (𝒕) = 𝒓𝒆𝒄𝒕 (𝟒𝒕)

b. 𝒈 (𝒕) = 𝒓𝒆𝒄𝒕(𝟒𝒕) ∗ 𝟒𝜹 (𝒕)
c. 𝒈 (𝒕) = 𝒓𝒆𝒄𝒕 (𝟒𝒕) ∗ 𝟒𝜹 (𝒕 − 𝟐)
d. 𝒈(𝒕) = 𝒓𝒆𝒄𝒕 (𝟒𝒕) ∗ 𝟒𝜹 (𝟐𝒕)

CH05 - 10
2. Find the following values. 𝒈 (𝟑) 𝒊𝒇 𝒈 (𝒕) = 𝟑 𝒆−𝟐𝒕 𝒖(𝒕) ∗ 𝟒𝜹 (𝒕 − 𝟏)

3. If 𝒙(𝒕) = −𝟓 𝒓𝒆𝒄𝒕 (𝒕/𝟐) ∗ [𝜹(𝒕 + 𝟏) + 𝜹(𝒕)], find the values of:

a. x (1/2)
b. x(−1/2)
c. x(−5/2)
4. Evaluate 𝒚(𝒕) = 𝒙(𝒕) ∗ 𝒉(𝒕), where 𝒙(𝒕) and 𝒉(𝒕) are shown below,

𝒙(𝒕) = 𝒖(𝒕) − 𝒖(𝒕 − 𝟑) 𝒉(𝒕) = 𝒖(𝒕) − 𝒖(𝒕 − 𝟐)

a. By an analytical technique
b. By a graphical method.

CH05 - 11
Frequency Response
• The variable 𝒔 in the complex exponential 𝒆𝒔𝒕 is, in general, complex valued. Let it be of the form
𝒔 = 𝝈 + 𝒋𝝎 where 𝝈 is the real part and 𝝎 is the imaginary part.

• For the special case 𝝈 = 𝟎, 𝒔 = 𝒋𝝎, the complex exponential 𝒆𝒔𝒕 becomes the complex sinusoid
𝒆𝒋𝝎𝒕 and the transfer function of the system 𝑯(𝒔) becomes the frequency response of the system
𝑯( 𝒋𝝎).

• The function 𝒆𝒋𝝎𝒕 is called a complex sinusoid because, by Euler’s identity, 𝒆𝒋𝝎𝒕 = 𝒄𝒐𝒔(𝝎𝒕) +
𝒋 𝒔𝒊𝒏(𝝎𝒕), the sum of a real cosine and an imaginary sine, both of radian frequency 𝝎. From
𝒀𝒆𝒔𝒕 = 𝐇(𝐬)𝑿𝒆𝒔𝒕 , letting 𝒔 = 𝒋𝝎

• Dividing through by 𝒆𝒋𝝎𝒕

• Equating magnitudes, we get |𝒀 | = |𝑯( 𝒋𝝎)||𝑿 | and equating phases we get ∡𝒀 =

∡𝑯(𝒋𝝎) + ∡𝑿. The function 𝑯(𝒋𝝎) is called the frequency response of the system because,
at any radian frequency 𝝎, if we know the magnitude and phase of the excitation and the magnitude
and phase of the frequency response, we can find the magnitude and phase of the response.

• In Chapter 4 we showed, using principles of linearity and superposition, that if a complex excitation
𝒙(𝒕) is applied to a system and causes a response 𝒚(𝒕), that the real part of 𝒙(𝒕) causes the real
part of 𝒚(𝒕) and the imaginary part of 𝒙(𝒕) causes the imaginary part of 𝒚(𝒕). Therefore, if the
actual excitation of a system is 𝒙(𝒕) = 𝑨𝒙 𝒄𝒐𝒔(𝝎𝒕 + 𝜽𝒙 ), we can find the response of the
system to an excitation

𝒙𝒄 (𝒕) = 𝑨𝒙 𝒄𝒐𝒔(𝝎𝒕 + 𝜽𝒙 ) + 𝒋 𝑨𝒙 𝒔𝒊𝒏(𝝎𝒕 + 𝜽𝒙 ) = 𝑨𝒙 𝒆 𝒋(𝝎𝒕+𝜽𝒙)

in the form

𝒚𝒄 (𝒕) = 𝑨𝒚 𝒄𝒐𝒔(𝝎𝒕 + 𝜽𝒚 ) + 𝒋 𝑨𝒚 𝒔𝒊𝒏(𝝎𝒕 + 𝜽𝒚 ) = 𝑨𝒚 𝒆 𝒋(𝝎𝒕+𝜽𝒚)

CH05 - 12
• and we can take the real part 𝒚(𝒕) = 𝑨𝒚 𝒄𝒐𝒔(𝝎𝒕 + 𝜽𝒚 ) as the response to the real excitation
𝒙(𝒕) = 𝑨𝒙 𝒄𝒐𝒔(𝝎𝒕 + 𝜽𝒙 ). Using |𝒀| = |𝑯(𝒋𝝎)||𝑿 | and ∡𝒀 = ∡𝑯(𝒋𝝎) + ∡𝑿 we get:

𝑨𝒚 = |𝑯( 𝒋𝝎)| 𝑨𝒙 𝒂𝒏𝒅 𝜽𝒚 = ∡𝑯(𝒋𝝎) + 𝜽𝒙

Example: Transfer Function and Frequency Response

• An LTI system is described by the differential equation:

𝒚″ (𝒕) + 𝟑𝟎𝟎𝟎 𝒚′ (𝒕) + 𝟐 × 𝟏𝟎𝟔 𝒚(𝒕) = 𝟐 × 𝟏𝟎𝟔 𝒙(𝒕).

a) Find its transfer function.

For this differential equation of the form:

𝑵 𝑴

∑ 𝒂𝒌 𝒔𝒌 𝒀𝒆𝒔𝒕 = ∑ 𝒃𝒌 𝒔𝒌 𝑿𝒆𝒔𝒕
𝒌=𝟎 𝒌=𝟎

𝑵 = 𝟐, 𝑴 = 𝟎, 𝒂𝟎 = 𝟐 × 𝟏𝟎𝟔 , 𝒂𝟏 = 𝟑𝟎𝟎𝟎, 𝒂𝟐 = 𝟏 𝒂𝒏𝒅 𝒃𝟎 = 𝟐 × 𝟏𝟎𝟔 .

Therefore the transfer function is

b) If 𝒙(𝒕) = 𝑿 𝒆𝒋𝟒𝟎𝟎𝝅𝒕 and 𝒚(𝒕) = 𝒀 𝒆𝒋𝟒𝟎𝟎𝝅𝒕 and 𝑿 = 𝟑 𝒆𝒋𝝅/𝟐 , find the magnitude and
phase of 𝒀. The frequency response is:

The radian frequency is 𝝎 = 𝟒𝟎𝟎𝝅. Therefore,

CH05 - 13
 c) If 𝒙(𝒕) = 𝟖 𝒄𝒐𝒔(𝟐𝟎𝟎𝝅𝒕) and 𝒚(𝒕) = 𝑨𝒚 𝒄𝒐𝒔(𝟐𝟎𝟎𝝅𝒕 + 𝜽𝒚 ), find 𝑨𝒚 and 𝜽𝒚 Using:

5.3 DISCRETE TIME

IMPULSE RESPONSE
• Just as was true for continuous-time systems, there is a convolution method for discrete-time
systems and it works in an analogous way. It is based on knowing the impulse response of the system,
treating the input signal as a linear combination of impulses and then adding the responses to all the
individual impulses.
• If we have a mathematical description of the system, we may be able to find the impulse response analytically.

Consider first a system described by a difference equation of the form

𝒂𝟎 𝒚[𝒏] + 𝒂𝟏 𝒚[𝒏 − 𝟏] + ⋯ + 𝒂𝑵 𝒚[𝒏 − 𝑵] = 𝒙[𝒏]

• This system is causal and LTI. To find the impulse response, we let the excitation 𝒙[𝒏] be a unit
impulse at 𝒏 = 𝟎. Then

𝒂𝟎 𝒉[𝒏] + 𝒂𝟏 𝒉[𝒏 − 𝟏] + ⋯ + 𝒂𝑵 𝒉[𝒏 − 𝑵] = 𝜹[𝒏]

𝜹[𝒏] − 𝒂𝟏 𝒉[𝒏 − 𝟏] − ⋯ − 𝒂𝑵 𝒉[𝒏 − 𝑵]

𝒉[𝒏] =
𝒂𝟎
CH05 - 14
• Now consider a more general system described by a difference equation of the form

𝒂𝟎 𝒚[𝒏] + 𝒂𝟏 𝒚[𝒏 − 𝟏] + ⋯ + 𝒂𝑵 𝒚[𝒏 − 𝑵] = 𝒃𝟎 𝒙[𝒏] + 𝒃𝟏 𝒙[𝒏 − 𝟏] + ⋯ + 𝒃𝑴 𝒙[𝒏 − 𝑴]

Or
𝑵 𝑴

∑ 𝒂𝒌 𝒚[𝒏 − 𝒌] = ∑ 𝒃𝒌 𝒙[𝒏 − 𝒌]
𝒌=𝟎 𝒌=𝟎

• Since the system is LTI we can find the impulse response by first finding the impulse responses to
systems described by the difference equations:

• And then adding all those impulse responses. Since all the equations are the same except for the
strength and time of occurrence of the impulse, the overall impulse response is simply the sum of
a set of impulse responses of the systems:

𝒉[𝒏] = 𝒃𝟎 𝒉𝟏 [𝒏 ] + 𝒃𝟏 𝒉𝟏 [𝒏 − 𝟏] − ⋯ − 𝒃𝑴 𝒉𝟏 [𝒏 − 𝑴]
where 𝒉𝟏 [𝒏 ] is the impulse response found earlier.

• Example: Find the impulse response 𝒉[𝒏] of the system described by the difference equation

𝟖 𝒚[𝒏] + 𝟔 𝒚[𝒏 − 𝟏] = 𝒙[𝒏]

If the excitation is an impulse

𝟖 𝒉[𝒏] + 𝟔 𝒉[𝒏 − 𝟏] = 𝜹[𝒏]

This equation describes a causal system so 𝒉[𝒏] = 𝟎, 𝒏 < 𝟎. We can find the first response to a unit
impulse at 𝒏 = 𝟎

CH05 - 15
The general solution (homogeneous solution) is
𝟑
𝒉[𝒏] + 𝟔 𝒉[𝒏 − 𝟏] = 𝟎 → 𝒉[𝒏] = − 𝒉[𝒏 − 𝟏]
𝟒
This is a first-order recurrence relation, so the general solution is:

𝒉[𝒏] = 𝑲𝒉 (−𝟑/𝟒)𝒏

To include causality:

𝟑 𝒏
𝒉[𝒏] = 𝑲𝒉 (− ) 𝒖[𝒏]
𝟒
where 𝒖[𝒏] is the unit step function (0 for n<0, 1 for n≥0).
Then the impulse response of the system is

𝟏 𝟑 𝒏
𝒉[𝒏] = ( ) (− ) 𝒖[𝒏]
𝟖 𝟒

DISCRETE-TIME CONVOLUTION

Graphical and Analytical Examples of Convolution

CH05 - 16
CH05 - 17
• Example: Response of a moving-average digital filter
A moving-average digital filter has an impulse response of the form:

𝒉[𝒏] = (𝒖[𝒏] − 𝒖[𝒏 − 𝑵])/𝑵.

Find the response of a moving-average filter with 𝑵 = 𝟖 to 𝒙[𝒏] = 𝒄𝒐𝒔(𝟐𝝅𝒏/𝟏𝟔).

CH05 - 18
Using convolution, the response is

𝒚[𝒏] = 𝒙[𝒏] ∗ 𝒉[𝒏] = 𝒄𝒐𝒔(𝟐𝝅𝒏/𝟏𝟔) ∗ (𝒖[𝒏] − 𝒖[𝒏 − 𝟖])/𝟖

Applying the definition of the convolution sum,
∞
𝟏 𝟐𝝅𝒎
𝒚[𝒏] = ∑ 𝒄𝒐𝒔( )(𝒖[𝒏 − 𝒎] − 𝒖[𝒏 − 𝒎 − 𝟖])
𝟖 𝟏𝟔
𝒎=−∞

The effect of the two unit sequence functions is to limit the summation range.
𝟐
𝟏 𝟐𝝅𝒎
𝒚[𝒏] = ∑ 𝒄𝒐𝒔( )
𝟖 𝟏𝟔
𝒎=𝒏−𝟕
𝒆𝒋𝒙 + 𝒆−𝒋𝒙
Using the trigonometric identity 𝒄𝒐𝒔(𝒙) =
𝟐
𝟐
𝟏
𝒚[𝒏] = ∑ 𝒆𝒋𝟐𝝅𝒎/𝟏𝟔 + 𝒆−𝒋𝟐𝝅𝒎/𝟏𝟔
𝟏𝟔
𝒎=𝒏−𝟕

Let 𝒒 = 𝒎 − 𝒏 + 𝟕. Then
𝟕
𝟏
𝒚[𝒏] = ∑ 𝒆𝒋𝟐𝝅(𝒒+𝒏−𝟕)/𝟏𝟔 + 𝒆−𝒋𝟐𝝅(𝒒+𝒏−𝟕)/𝟏𝟔
𝟏𝟔
𝒒=𝟎
𝟕 𝟕
𝟏
𝒚[𝒏] = [ 𝒆𝒋𝟐𝝅(𝒏−𝟕)/𝟏𝟔 ∑ 𝒆𝒋𝟐𝝅𝒒/𝟏𝟔 + 𝒆−𝒋𝟐𝝅(𝒏−𝟕)/𝟏𝟔 ∑ 𝒆−𝒋𝟐𝝅𝒒/𝟏𝟔
𝟏𝟔
𝒒=𝟎 𝒒=𝟎

The summation of a geometric series with 𝑵 terms is

This formula works for any complex value of 𝒓. Therefore, summing these geometric series each of length 8,

CH05 - 19
Finding a common denominator and simplifying,

Then, using the periodicity of the cosine, 𝒄𝒐𝒔(𝝅(𝒏 − 𝟖)/𝟖) = 𝒄𝒐𝒔(𝝅𝒏/𝟖)

𝒚[𝒏] = 𝟏. 𝟔𝟒𝟐𝟏[𝒄𝒐𝒔(𝝅(𝒏 − 𝟕)/𝟖) + 𝒄𝒐𝒔(𝝅𝒏/𝟖)]

CH05 - 20