Signals and Systems Lecture 1

Dr. Ghandi Manasra and Ahmed Rabei Fall 2025

Lecture 1: Introduction to Signals

Reference
Oppenheim & Willsky, Signals and Systems, Chapter 1, Sections 1.0–1.2.1

Review of Prerequisites
ˆ Basic calculus and functions
ˆ Understanding of complex numbers
ˆ Familiarity with mathematical notation
ˆ Engineering problem-solving approach

1.0 Welcome & Introduction

1.1 What is a Signal?

Definition 1. A signal is information that changes over time (or space). Mathematically, it is a
function of one or more independent variables.

ˆ Continuous-Time (CT): x(t), t ∈ R.

ˆ Discrete-Time (DT): x[n], n ∈ Z.
ˆ Convention: use parentheses for CT, brackets for DT.

1.2 What is a System?

Definition 2. A system is any process or operation that acts on a signal to produce another signal.
It takes a signal, performs some operation, and outputs a new signal.

ˆ Systems can be physical (amplifier, filter) or abstract (algorithms in DSP).

ˆ Systems can be modeled by an operator T {·}:

Continuous-Time (CT): y(t) = T {x(t)}, Discrete-Time (DT): y[n] = T {x[n]}.

1.3 Two Basic Types of Signals and Systems
1.3.1 Continuous-Time (CT) Signals and Systems
ˆ x(t), t ∈ R.
ˆ Defined for every instant of continuous time.
ˆ Examples: analog audio, temperature variation, vibrations.
ˆ Systems act on CT signals: y(t) = T {x(t)}.

x(t) −→ CT S −→ y(t)

Example 1. Consider a Gaussian pulse as a CT signal:

Gaussian Pulse as a CT Signal

1
t2
g(t) = e− 2
0.8

0.6
g(t)

0.4

0.2

0
−5 −4 −3 −2 −1 0 1 2 3 4 5
t

Figure 1.1: Gaussian pulse as continuous-time signal example.

1.3.2 Discrete-Time (DT) Signals and Systems

ˆ x[n], n ∈ Z.
ˆ Defined only at specific, discrete points.
ˆ Systems: y[n] = T {x[n]}.
x[n] −→ DT S −→ y[n]
ˆ Examples: sampled audio, stock prices, digital images.
ˆ Often arise from sampling CT signals.

2
Example 2. Consider a sampled Gaussian pulse as a DT signal:

Sampled Gaussian Pulse as DT Signal

0.8

0.6
x[n]

0.4

0.2

0
−6 −4 −2 0 2 4 6
n

Figure 1.2: Sampled Gaussian pulse as discrete-time signal.

1.4 Deterministic vs Random Signals

ˆ Deterministic: Completely predictable, e.g. x(t) = cos(2πt).
ˆ Random: Unpredictable, described by statistics (mean, variance, spectrum). Example:
thermal noise.
ˆ Course focus: deterministic signals, but real-world signals often contain randomness.

1.5 Energy and Power of Signals

The energy and power of a signal provide fundamental measures of its ”strength” or ”intensity.”
These definitions have deep physical significance: for a voltage signal x(t) across a 1-ohm resistor,
the energy represents the total energy dissipated, and the power represents the average rate of
energy dissipation.

Formal Definitions (CT case)

Z ∞ Z T
2 1
E∞ = |x(t)| dt, P∞ = lim |x(t)|2 dt
−∞ T →∞ 2T −T

Discrete-Time Analogs
∞ N
X 1 X
E∞ = |x[n]|2 , P∞ = lim |x[n]|2
N →∞ 2N + 1
n=−∞ n=−N

Key Categories
ˆ Energy Signal: 0 < E∞ < ∞, P∞ = 0.
ˆ Power Signal: E∞ = ∞, 0 < P∞ < ∞.

3
Example 1: Rectangular Pulse (Energy)
(
1
−1 ≤ t ≤ 1
Z
1
x(t) = ⇒ E∞ = 1 dt = 2
0 otherwise −1

Example 2: Cosine Wave (Power)

x(t) = A cos(ω0 t)
A2
E∞ → ∞, P∞ = 2

Signals That Are Neither

ˆ Condition: E∞ = ∞, P∞ = ∞.
ˆ Example: Ramp signal (
t, t≥0
x(t) = tu(t) =
0, t<0

1.6 Transformations of Signals

ˆ Time Shift: x(t − t0 ) or x[n − n0 ]. Delay if t0 > 0, advance if t0 < 0.
ˆ Time Reversal: x(−t), x[−n]. Reflection about zero.
ˆ Time Scaling: x(at). If |a| > 1, compression; if 0 < |a| < 1, expansion.
ˆ Combined: x(αt + β). To avoid ambiguity, always factor as x(α(t + β/α)).
– Clear method: Factor to get x(α(t + β/α)), then apply transformations in order: first
shift by −β/α, then scale by α.

Summary and Next Lecture

ˆ Defined signals, systems, CT vs DT signals.
ˆ Deterministic vs random signals.
ˆ Energy and power classification.
ˆ Time-domain transformations.
ˆ Next time: Elementary signals, periodicity, and symmetry.

4
 Signals and Systems Lecture 2
Dr. Ghandi Manasra and Ahmed Rabei Fall 2025

Lecture 2: Elementary Signals and Signal Properties

Reference
Oppenheim & Willsky, Signals and Systems, Chapter 1, Sections 1.2.2 – 1.4

Review of Lecture 1
ˆ CT and DT signals defined
ˆ Time-domain transformations learned
ˆ Audio signal examples
ˆ Energy and power concepts

2.1 Periodic Signals

Continuous-Time (CT) Periodic Signals
Definition 1. A CT signal x(t) is periodic if there exists a constant T > 0 such that
x(t) = x(t + T ), ∀t ∈ R.
ˆ The fundamental period T0 is the smallest such positive value of T .
ˆ If T0 is the fundamental period, then mT0 is also a period for any integer m.
ˆ The fundamental frequency is given by:
1 2π
f0 = (in Hz), ω0 = 2πf0 = (in rad/s).
T0 T0

x(t)

−6 −4 −2 2 4 6 t

−1

T0 = 2

Figure 2.1: Periodic continuous-time triangular wave.

Example 1. Show that x(t) = A cos(ω0 t + θ) has a fundamental period of
2π
T0 = .
|ω0 |

Sum of Periodic CT Signals: A sum of two periodic signals, x(t) = x1 (t) + x2 (t) with
fundamental periods T1 and T2 , is periodic if and only if the ratio of their periods is a rational
number:
T1 m
= , where m, n ∈ Z+ .
T2 n
The fundamental period is then
T0 = nT1 = mT2 ,
where m, n are the smallest such integers. This means T0 is the smallest time when the first signal
has completed m full cycles and the second signal has completed n full cycles simultaneously,
marking the moment they ”align” perfectly and the combined signal repeats.

x1 (t) = cos(πt) with period T1 = 2

1

1 2 3 4 5 6 7 8 9 10 11 12 t
−1
x2 (t) is a triangular wave with period T2 = 3
1

1 2 3 4 5 6 7 8 9 10 11 12 t
−1
Sum: x(t) = x1 (t) + x2 (t), with period T0 = lcm(2, 3) = 6
2

1 2 3 4 5 6 7 8 9 10 11 12 t
−1

−2 T0 = 6

Figure 2.2: Sum of two periodic continuous-time signals.

2
 Signal 1: cos(2πt) with T1 = 1

−5 −4 −3 −2 −1 1 2 3 4 5

−1
√ √
Signal 2: cos(2π 2 t) with T2 = 1/ 2

−5 −4 −3 −2 −1 1 2 3 4 5

−1
√
Aperiodic Sum: cos(2πt) + cos(2π 2 t)
2

−5 −4 −3 −2 −1 1 2 3 4 5
−1

−2

Figure 2.3: Sum of periodic signals creating aperiodic result.

3
Discrete-Time (DT) Periodic Signals
Definition 2. A DT signal x[n] is periodic if there exists an integer N > 0 such that

x[n] = x[n + N ], ∀n ∈ Z.

ˆ The fundamental period N0 is the smallest such positive integer.

ˆ If N0 is a period, then mN0 is also a period for any integer m.
ˆ The fundamental angular frequency is ω0 = N 2π
0
(in rad/sample).

Discrete-Time Periodic Triangular Wave

0.5

8 16 24
−0.5

−1
N0 = 8

Figure 2.4: Periodic discrete-time triangular wave.

Sum of Periodic DT Signals: If x1 [n] has period N1 and x2 [n] has period N2 , their sum
x[n] = x1 [n] + x2 [n] is periodic with period N0 given by

N1 N2
N0 = lcm(N1 , N2 ) = .
gcd(N1 , N2 )

For x[n] to be periodic, both x1 [n] and x2 [n] must complete an integer number of their respective
periods simultaneously. The smallest such interval is the fundamental period N0 of the sum.

4
Example 2. ˆ N1 = 6, N2 = 9 =⇒ N0 = lcm(6, 9) = 18.
ˆ N1 = 8, N2 = 12 =⇒ N0 = lcm(8, 12) = 24.

2π


x1 [n] = cos 6 n with period N1 = 6

6 12 18 24 30 36 n
−1
2π


x2 [n] = cos 9 n with period N2 = 9

9 18 27 36 n
−1

Sum: x[n] = x1 [n] + x2 [n], with period N0 = lcm(6, 9) = 18

9 18 27 36 n

−2
N0 = 18

Figure 2.5: Sum of two periodic discrete-time signals.

Average Power of Periodic Signals

Periodic signals are classified as power signals. Their average power can be computed over one
fundamental period (T0 or N0 ) as follows:

ˆ CT average power: Z
1
Pave = |x(t)|2 dt.
T0 t0

ˆ DT average power:
1 X
Pave = |x[n]|2 .
N0
n=⟨N0 ⟩

5
2.2 Even and Odd Signals
A signal can be classified based on its symmetry with respect to the time origin.
Definition 3. Even signal: Symmetric about the vertical axis.

x(−t) = x(t) (CT), x[−n] = x[n] (DT).

Even: x(−t) = x(t)

Even: x(−t) = x(t)
1
1
0.8
0.5
0.6

−4 −2 2 4t 0.4

−0.5 0.2

−1
−4 −2 2 4t

Figure 2.6: Even signal symmetry examples.

Definition 4. Odd signal: Symmetric about the origin.

x(−t) = −x(t) (CT), x[−n] = −x[n] (DT).

Odd: x(−t) = −x(t) Odd: x(−t) = −x(t)

1 1

0.5 0.5

−4 −2 2 4t −4 −2 2 4t

−0.5 −0.5

−1 −1

Figure 2.7: Odd signal symmetry examples.

6
Proposition 1. Any signal can be decomposed into the sum of an even part and an odd part:

x(t) = xe (t) + xo (t),

where the even and odd components are given by:

1 1
xe (t) = (x(t) + x(−t)) , xo (t) = (x(t) − x(−t)) .
2 2
The same decomposition applies to DT signals by replacing t by n.

Original Signal: x(t) = e−t u(t)

0.5

−2 −1.5 −1 −0.5 0.5 1 1.5 2 2.5 3 3.5 4t

1
x(t) + x(−t) = 12 e−|t|


Even Part: xe (t) = 2

0.5

−4 −3.5 −3 −2.5 −2 −1.5 −1 −0.5 0.5 1 1.5 2 2.5 3 3.5 4t

1 1
sgn(t) e−|t|


Odd Part: xo (t) = 2 x(t) − x(−t) = 2

0.5

−4 −3.5 −3 −2.5 −2 −1.5 −1 −0.5 0.5 1 1.5 2 2.5 3 3.5 4t

−0.5

Figure 2.8: Signal decomposition into even and odd parts.

Energy of Even/Odd Parts: Because the even and odd components of a signal are orthogonal,
the total energy of a signal is the sum of the energies of its even and odd parts:
Z ∞ Z ∞ Z ∞
2 2
|x(t)| dt = |xe (t)| dt + |xo (t)|2 dt.
−∞ −∞ −∞

The same applies for DT signals by replacing integration with summation and t with n.

7
2.3 Exponential and Sinusoidal Signals
Continuous-Time (CT) Exponentials
The general form is:
x(t) = Ceat , where a, C ∈ C.

Complex Exponential ejω0 t

Real Exponential Signals
4 Im
x(t) a>0
1
a<0
3 ejω0 t

ω0 t Re
2
−1 1

−1
t
|ejω0 t | = 1
−1 1
(a) Real exponential signals. (b) Complex exponential signal.

Figure 2.9: Visualization of real and complex exponential signals.

ˆ Real Exponent (a, C ∈ R):

– If a > 0, the signal grows exponentially.
– If a < 0, the signal decays exponentially.
ˆ Complex Exponent (a = jω0 ): The signal x(t) = ejω0 t is periodic. Using Euler’s relation:

x(t) = ejω0 t = cos(ω0 t) + j sin(ω0 t).

It represents a vector of unit length rotating in the complex plane at angular frequency ω0 .
Its fundamental period is
2π
T0 = .
|ω0 |
ˆ The higher |ω0 |, the faster the vector spins, resulting in faster oscillations in the real and
imaginary parts.

Continuous-Time (CT) Sinusoids

n o
x(t) = A cos(ω0 t + θ) = ℜ Aej(ω0 t+θ) .

This shows that sinusoids are projections of rotating complex exponentials onto the real axis.
The fundamental frequency f0 in Hertz (cycles per second) and the fundamental period T0 are
related to ω0 as:
ω0 1 2π
f0 = , T0 = = .
2π f0 ω0

8
Discrete-Time (DT) Exponentials
The general form is:
x[n] = Cαn .

x[n] = 1.2n x[n] = 0.8n

6 3

4 2

2 1

−5 −2 1 4 7 10 n −5 −2 1 4 7 10 n

x[n] = (−1.2)n x[n] = (−0.8)n

5
2

−5 −2 1 4 7 10 n −5 −2 1 4 7 10 n

−2
−5

Figure 2.10: Discrete-time exponential signal examples.

ˆ Real Base (α ∈ R):

– If |α| > 1, the sequence grows.
– If |α| < 1, the sequence decays.
ˆ Complex Base α = ejω0 : The signal x[n] = ejω0 n has unique properties compared to its CT
counterpart.

9
Key Differences of DT Complex Exponentials from CT:

ˆ For x[n] = ejω0 n to be periodic with period N , it must satisfy:

ω0 N = 2πm

for some integers m and N . This implies that periodicity requires ω2π0 to be a rational number.
ˆ Frequency Aliasing: Frequencies separated by integer multiples of 2π produce identical
signals:
ej(ω0 +2πk)n = ejω0 n ej2πkn = ejω0 n , for any integer k.
In contrast, for CT signals, distinct values of ω0 always produce distinct signals.
Key insight: Because frequencies alias every 2π, all unique frequency content is captured in
an interval like [−π, π). This fundamental property of discrete-time signals means we only
need to consider a frequency interval of length 2π.
ˆ Rate of oscillation is not monotonic with ω0 :
– Low frequencies (slow oscillation) occur when ω0 is near an even multiple of π (e.g.,
0, ±2π, . . .).
– High frequencies (fast oscillation) occur when ω0 is near an odd multiple of π (e.g.,
±π, ±3π, . . .).
– The fastest possible oscillation corresponds to ω0 = π, which gives:

x[n] = ejπn = (−1)n .

10
2.4 Unit Impulse and Unit Step Functions
Discrete-Time (DT) Signals
Unit Impulse δ[n]: Represents a single instantaneous pulse at n = 0:
(
1, n = 0
δ[n] =
0, n ̸= 0

Unit Step u[n]: Represents a signal that switches on at n = 0 and stays on:
(
1, n ≥ 0
u[n] =
0, n < 0

Unit Impulse δ[n] Unit Step u[n]

1 1

−4 −3 −2 −1 1 2 3 4 n −4 −3 −2 −1 1 2 3 4 n
(a) Unit Impulse (b) Unit Step

Figure 2.11: Fundamental discrete-time signals

11
Properties:

ˆ The impulse is the first difference of the step:

δ[n] = u[n] − u[n − 1].

u[n]
1

−u[n − 1] n

−1
δ[n]
1

−3 −2 −1 0 1 2 3 4 5 n

Figure 2.12: Relation between discrete-time step and impulse.

ˆ The step is the running sum of the impulse:

n
X ∞
X
u[n] = δ[k] = δ[n − k].
k=−∞ k=0

ˆ For any x[n]: x[n]δ[n − n0 ] = x[n0 ]δ[n − n0 ].

ˆ The impulse is even:
δ[−n] = δ[n].

12
Continuous-Time (CT) Signals
Unit Step u(t): (
1, t≥0
u(t) =
0, t<0

−3 −2 −1 0 1 2 3

Figure 2.13: Continuous-time unit step function.

Unit Impulse δ(t) (Dirac Delta Function): The CT unit impulse is an idealized function
defined by its properties under integration rather than its value. Conceptually, it is an infinitely
tall, infinitely narrow pulse at t = 0 with a total area of 1.
Formally, it can be viewed as the limit of a pulse gT (t) shrinking in width but maintaining unit
area:
δ(t) = lim gT (t).
T →0

T ↓0
gT (t) → δ(t)
6 T =1
T = 0.5
5 T = 0.25
4

−1 −0.5 0 0.5 1

Figure 2.14: Continuous-time unit impulse approximation.

13
 1

−2 −1 0 1 2

Figure 2.15: Continuous-time unit impulse representation.

Properties:

ˆ Sifting property: The impulse ”sifts out” the value of a function at a single point:
Z ∞
x(t)δ(t − t0 ) dt = x(t0 ).
−∞

ˆ The total area is unity: Z ∞

δ(t) dt = 1.
−∞
ˆ The impulse is the derivative of the step:
d
δ(t) = u(t),
dt
and the step is the integral of the impulse:
Z t
u(t) = δ(τ ) dτ.
−∞

ˆ Scaling property:  
1 b
δ(at + b) = δ t+ .
|a| a
ˆ The impulse is even:
δ(−t) = δ(t).

Summary and Next Lecture

ˆ Defined periodicity for CT and DT signals.
ˆ Explored signal symmetries: even and odd decomposition.
ˆ Introduced fundamental exponential and sinusoidal signals.
ˆ Defined the unit impulse and unit step functions.
ˆ Next time: System properties and classifications.

14
 Signals and Systems Lecture 3
Dr. Ghandi Manasra and Ahmed Rabei Fall 2025

Lecture 3: Systems and Their Properties

Reference
Oppenheim & Willsky, Signals and Systems, Chapter 1, Sections 1.5–1.6; MIT OCW lecture notes
on LTI systems.

Review of Lecture 2
ˆ Elementary signals toolbox introduced
ˆ Unit impulse sifting property
ˆ Signal representation building blocks
ˆ Today: systems and properties

3.1 Systems and Interconnections

A system is any process that transforms an input signal into an output signal. It can be modeled
by an operator T {·} mapping an input signal to an output signal:

Continuous-Time (CT): y(t) = T {x(t)}, Discrete-Time (DT): y[n] = T {x[n]}.

Input Output Input Output

x(t) CTS y(t) x[n] DTS y[n]

(a) Continuous-Time System (b) Discrete-Time System

Figure 3.1: Continuous-time and discrete-time system blocks.

Systems are often interconnected to compose complex processing chains. Common interconnec-
tions include:

1. Series (Cascade): Output of T1 feeds input of T2 .

2. Parallel: Same input fed to multiple systems; outputs summed.
3. Feedback: Output fed back to the input and combined with the original input.

y1 (t)
x(t) System T1 System T2 y(t)

Figure 3.2: Series cascade interconnection example.

 System T1

x(t) + y(t)

System T2

Figure 3.3: Parallel interconnection example.

+ Forward System
x(t) y(t)
T1
−

Feedback System
T2

Figure 3.4: Feedback interconnection example.

3.2 Fundamental System Properties

Systems can be characterized by the following six properties:

1. Memory
Definition 1. A system is memoryless if the output at any time depends solely on the input at
that time; otherwise, it has memory (is dynamic).

Example 1.

ˆ y[n] = x[−n]

ˆ y(t) = d
dt x(t)

ˆ y[n] = max{x[n], x[n − 1]}

ˆ y(t) = (3 + sin t) x(t)

Rt
ˆ y(t) = −∞ x(τ ) dτ

ˆ y[n] = sgn(x[n])

ˆ y(t) = x(t − 1)

2
2. Invertibility
Definition 2. A system is invertible if distinct inputs produce distinct outputs, and an inverse
system T −1 exists such that T −1 {y} = x.
Example 2.

ˆ Invertible Systems: The original input x can be uniquely recovered from the output y.

– y(t) = 2x(t) (Amplifier): The inverse system is x(t) = 12 y(t).

Rt d
– y(t) = P−∞ x(τ ) dτ (Integrator): The inverse is x(t) = dt y(t).
n
– y[n] = k=−∞ x[k] (Accumulator): The inverse is x[n] = y[n] − y[n − 1].

ˆ Non-Invertible Systems: Different inputs can lead to the same output.

– y(t) = x2 (t) (Squarer): Loses the sign of the input; x(t) = 2 and x(t) = −2 both produce
y(t) = 4.
– y[n] = nx[n]: For n = 0, y[0] = 0 regardless of the value of x[0]. Information about x[0]
is lost.
d d
– y(t) = dt x(t) (Differentiator): Loses any DC offset in the input, since dt (x(t) + C) =
d
dt x(t).
– y[n] = x[2n] (Downsampler): Loses all odd-indexed samples of the input signal.

Invertible System

x(t) System T y(t) Inverse T −1 x(t)

3. Causality
Definition 3. A system is causal if its output at time t0 depends only on the input at the same or
previous time {x(t) : t ≤ t0 }. Otherwise, it is non-causal.
Remark 1. All real-time physical systems must be causal, as they cannot react to future events.
Example 3.
ˆ y(t) = x(t − t0 ) ( t0 ≥ 0)
ˆ y[n] = Rx[2n]
t+t
ˆ y(t) = t 0 x(τ ) dτ ( t0 > 0)
ˆ y(t) = x(−2t)
Rt
ˆ y(t) = P −∞ x(τ ) dτ
ˆ y[n] = nk=−∞ x[k]
ˆ y[n] = x[n − 1]
ˆ y(t) = x(t
( + 1)
x(3t) if t < 0
ˆ y(t) =
x(t − 1) if t ≥ 0

3
4. Stability
Definition 4 (BIBO Stability). A system is Bounded-Input, Bounded-Output (BIBO) stable
if every bounded input produces a bounded output.
Example 4.
ˆ Stable Systems: A bounded input |x| ≤ Bx produces a bounded output |y| ≤ By .
– Time Delay: y[n] = x[n − 1]. A simple shift does not alter the amplitude bounds of the
signal. RStable.
t+t
– y(t) = t 0 x(τ ) dτ . The output is bounded by |y(t)| ≤ |t0 | · Bx . Stable.
– Moving Average: y[n] = 2M1+1 M
P
k=−M x[n − k]. The output is bounded by |y[n]| ≤ Bx .
Stable.
– y(t) = ex(t) . If |x(t)| ≤ Bx , the output is bounded by |y(t)| ≤ eBx . Stable.

ˆ Unstable Systems: A bounded input can produce an unbounded output.

d
– Differentiator: y(t) = dt x(t). A bounded input like x(t) = sin(t2 ) produces an
unbounded output y(t) 2
R t = 2t cos(t ). Unstable.
– Integrator: y(t) = −∞ x(τ ) dτ . The bounded input x(t) = u(t) produces the unbounded
ramp output y(t) = t · u(t).
PnUnstable.
– Accumulator: y[n] = k=−∞ x[k]. The bounded input x[n] = u[n] produces the
unbounded output y[n] = (n + 1)u[n]. Unstable.

5. Time-Invariance
Definition 5. A system is time-invariant if a time shift in the input results in an identical time
shift in the output:
y(t) = T {x(t)} =⇒ y(t − t0 ) = T {x(t − t0 )} ∀t0 .

x(t) System y(t)

x(t − t0 ) System y ∗ (t) ?

= y(t − t0 )

Figure 3.5: Time-invariance principle: shifting input shifts output.

Remark 2. Physical systems whose properties do not change over time are time-invariant (e.g., an
RC circuit with fixed component values).

Example 5.
ˆ y(t) = x(t + f (t))
ˆ y(t) = dt
d
x(t)
ˆ y(t) = f (x(t))
ˆ y(t) = x(f (t))
ˆ y(t) = Ax(t + α)
ˆ y(t) = Ax(αt)

4
6. Linearity
Definition 6 (Linearity). A system is linear if it satisfies superposition principles. This combines
two properties:

1. Additivity: T {x1 + x2 } = T {x1 } + T {x2 }.

2. Homogeneity (Scaling): T {αx} = αT {x}, for any scalar α ∈ C.

The two properties can be combined into a single statement:

T {αx1 + βx2 } = αT {x1 } + βT {x2 }

for any inputs x1 , x2 and any scalars α, β ∈ C.

x1 (t) System y1 (t)

+ y1 (t) + y2 (t)

x2 (t) System y2 (t)

?
y ∗ (t) = y1 (t) + y2 (t)

x1 (t)
+ System y ∗ (t)
x2 (t)

Additivity principle for linear systems.

×α
x(t) System y(t) α y(t)

×α
x(t) System y ∗ (t) y ∗ (t) =
?
α y(t)
Homogeneity scaling principle for linearity.

Remark 3. A necessary (but not sufficient) condition for linearity is that a zero input must produce
a zero output: T {0} = 0.

5
Example 6. ˆ Linear Systems: Obey the superposition principle T {ax1 + bx2 } = aT {x1 } +
bT {x2 }.

– y(t) = Ax(αt + β): Time scaling, shifting, and amplitude scaling are all linear operations.
d
– y(t) = dt x(t): Differentiation is a linear operator.
Rt
– y(t) = P −∞ x(τ ) dτ : Integration is a linear operator.
– y[n] = nk=−∞ x[k]: Summation is a linear operator.
– y(t) = x(f (t)): (Linear) The transformation is on the time variable, not the signal’s
amplitude.
– y(t) = x(t + f (t)): (Linear) Similar to the above, this is a time-varying shift, which is a
linear operation.

ˆ Non-Linear Systems: Violate either additivity or homogeneity.

– y[n] = x[n − 1]x[n]: Fails homogeneity, as scaling the input by a scales the output by a2 .
– y[n] = x[n] + C: Fails the zero-input test; if x[n] = 0, y[n] = C ̸= 0.
– y(t) = |x(t)|: Fails homogeneity, since | − x(t)| = |x(t)| ̸= −|x(t)|.
– y(t) = ℜ{x(t)}: Fails homogeneity for complex scalars (e.g., ℜ{j ·1} = 0, but j ·ℜ{1} = j).
It is, however, additive.
– y[n] = ℑ{x[n]}: Also fails homogeneity for complex scalars.
– y(t) = x∗ (t): The complex conjugate fails homogeneity for complex scalars, since
(ax(t))∗ = a∗ x∗ (t) ̸= ax∗ (t).
– y(t) = x(t) + f (t): Fails the zero-input test if f (t) ̸= 0.

3.3 Summary and Next Lecture

ˆ Defined systems and six fundamental properties: memory, invertibility, causality, stability,
time-invariance, and linearity.
ˆ System interconnections: series, parallel, and feedback.
ˆ Linear Time-Invariant (LTI) systems are most important for analysis.
ˆ Next time: LTI system characterization and convolution.

6
 Signals and Systems Lecture 4
[Link] Manasra and Ahmed Rabei Fall 2025

Lecture 4: LTI Systems and the Discrete-Time Convolution Sum

Reference
Oppenheim & Willsky, Signals and Systems, Chapter 2, Sections 2.0–2.1

Review of Lecture 3
ˆ Six system properties defined
ˆ Linearity and time-invariance most important
ˆ LTI systems mathematically tractable
ˆ Model many real-world phenomena

Today we will see the first major payoff of focusing on LTI systems: if we know the response of
an LTI system to a single, simple input—the unit impulse—we can determine its response to any
input. This leads us to the convolution sum.

4.1 Representation of Signals in Terms of Impulses

The key to understanding LTI systems is recognizing that any discrete-time signal can be represented
as a linear combination of scaled and shifted unit impulses.
Example 1 (Concrete Signal Decomposition). Consider the discrete-time signal with values:

x[−1] = 1, x[0] = 2, x[1] = 1, and x[n] = 0 elsewhere

We can write this signal as:

x[n] = x[−1]δ[n − (−1)] + x[0]δ[n − 0] + x[1]δ[n − 1] (1)

= 1 · δ[n + 1] + 2 · δ[n] + 1 · δ[n − 1] (2)

Each term contributes only at its respective time:

ˆ x[−1]δ[n + 1] = 1 when n = −1, and 0 elsewhere

ˆ x[0]δ[n] = 2 when n = 0, and 0 elsewhere
ˆ x[1]δ[n − 1] = 1 when n = 1, and 0 elsewhere
 Original Signal x[n]
x[n]
2
1

x[−1]δ[n + 1]
2
x[−1] = 1
1

x[0]δ[n]
2
x[0] = 2
1

x[1]δ[n − 1]
2
x[1] = 1
1

Sum
2
1

−2 −1 1 2 3 n

P
Figure 4.1: Signal decomposition using sifting property: x[n] = k x[k]δ[n − k].

Proposition 1 (General Sifting Property). Any discrete-time signal x[n] can be written as:
∞
X
x[n] = x[k]δ[n − k]
k=−∞

Similarly, for continuous-time signals:

Z ∞
x(t) = x(τ )δ(t − τ )dτ
−∞

2
4.2 The Discrete-Time Convolution Sum
Now let’s see what happens when we put the signal representation into an LTI system T .

∞
( )
X
y[n] = T {x[n]} = T x[k]δ[n − k] (3)
k=−∞

Step 1: Use Linearity. The response to a weighted sum of inputs is the weighted sum of
individual responses:
X∞
y[n] = x[k]T {δ[n − k]}
k=−∞

Step 2: Use Time-Invariance. Define the impulse response h[n] = T {δ[n]}. For an LTI
system:
hk [n] = T {δ[n − k]} = h[n − k]
Step 3: Combine. Substituting back gives us the discrete-time convolution sum:
Definition 1 (Discrete-Time Convolution Sum).
∞
X
y[n] = x[k]h[n − k] = (x ∗ h)[n]
k=−∞

Remark 1. The impulse response h[n] completely characterizes the behavior of an LTI system.
Given h[n], it is possible to compute the output y[n] for any input x[n]. This is not true for
nonlinear systems.

Input: Unit Impulse δ[n] Output: Impulse Response h[n]

1 1

0.5

−2 −1 0 1 2 3 4 5 −2 −1 0 1 2 3 4 5
n n

Figure 4.2: Impulse response characterizes LTI system completely.

3
4.3 Examples
Convolution as Echo: Imagine h[n] as an echo pattern. For example, h[n] = {1, 0, 0, 0.5, 0, 0.2}
models original sound plus delayed echoes.

Convolution as Smoothing: The impulse response h[n] = 15 {1, 1, 1, 1, 1} represents a 5-point

moving average filter.

h[n] Echo Impulse Response h[n] Moving Average Filter

1 0.2

0.5

0.2

1 2 3 4 5 6 1 2 3 4 5 6
n n

Figure 4.3: Echo system and moving average filter examples.

4.4 Computing Convolution: Flip-and-Slide Method

To compute convolution at a point n:

1. Plot signals x[k] and h[k] as functions of dummy variable k.

2. Flip h[k] around zero to get h[−k].
3. Slide h[−k] by n to obtain h[n − k].
4. Multiply pointwise: x[k] × h[n − k].
5. Sum all products over k.

4
4.4.1 Numerical Examples
Example 2 (Exponential Input with Unit Step Response). Given:
 n−1
1
x[n] = u[n − 1], h[n] = u[n]
2

Find y[n] = (x ∗ h)[n].

Solution:
∞ ∞  k−1
X X 1
y[n] = x[k]h[n − k] = u[k − 1] · u[n − k] (4)
2
k=−∞ k=−∞

For n < 1: No overlap, so y[n] = 0.

For n ≥ 1: The sum runs from k = 1 to k = n:
n  k−1 n−1 r  n 
X 1 − ( 12 )n

X 1 1 1
y[n] = = = 1 =2 1− (5)
2 2 1− 2 2
k=1 r=0

1 n



Therefore: y[n] = 2 1 − 2 u[n − 1]

x[n] = ( 12 )n−1 u[n − 1]

x[n]

h[n] = u[n] (unit step)

h[n]

y[n] = x[n] ∗ h[n]

y[n]
2

−1 1 2 3 4 5 6 7 8 n

Figure 4.4: Exponential decay convolved with unit step.

5
Example 3 (Finite Impulse Response System). Consider an LTI system with impulse response:

2
 n=0
h[n] = −1 n = 1

0 otherwise


Determine the output for input:




1 n = −1

2 n=0
x[n] =


−1 n=1

0 otherwise

Solution using the representation:

x[n] = δ[n + 1] + 2δ[n] − δ[n − 1]

Since y[n] = x[n] ∗ h[n], by linearity:

y[n] = h[n + 1] + 2h[n] − h[n − 1]

Computing each term:

ˆ y[−1] = h[0] = 2
ˆ y[0] = h[1] + 2h[0] = −1 + 4 = 3
ˆ y[1] = h[2] + 2h[1] − h[0] = 0 + 2(−1) − 2 = −4
ˆ y[2] = h[3] + 2h[2] − h[1] = 0 + 2(0) − (−1) = 1
ˆ y[3] = h[4] + 2h[3] − h[2] = 0 + 0 − 0 = 0
ˆ y[n] = 0 for n < −1 or n > 3
Example 4 (Complex Interval Analysis). Let:

x[n] = 2{u[n + 2] − u[n − 12]}, h[n] = αn {u[n − 2] − u[n − 13]}

The convolution requires careful interval analysis. The non-zero intervals are:
ˆ x[n] ̸= 0 for −2 ≤ n ≤ 11
ˆ h[n] ≠ 0 for 2 ≤ n ≤ 12
This leads to five distinct intervals for the output, each requiring separate analysis of the
convolution sum limits.

Summary and Next Lecture

ˆ Any
P discrete-time signal can be represented as a sum of scaled and shifted impulses: x[n] =
k x[k]δ[n − k].
ˆ The output of a discrete-time LTI system is the convolution sum: y[n] = ∞
P
k=−∞ x[k]h[n − k].
ˆ The impulse response h[n] completely characterizes any LTI system.
ˆ Convolution can be computed using the flip-and-slide method or by careful interval analysis.
ˆ Next time: Continuous-time convolution integral and properties of convolution.

6
 Signals and Systems Lecture 5
Dr. Ghandi Manasra and Ahmed Rabei Fall 2025

Lecture 5: Continuous-Time LTI Systems and the Convolution

Integral

Reference
Oppenheim & Willsky, Signals and Systems, Chapter 2, Section 2.2

Review of Lecture 4
ˆ Discrete-time convolution sum derived
ˆ Signal decomposition using impulses
ˆ LTI systems characterized by impulse response
ˆ Flip-and-slide method for computation

Today we extend the convolution concept to continuous-time LTI systems. The derivation and
intuition are identical to the discrete-time case—the only difference is that summation becomes
integration.

5.1 Representing CT Signals with Impulses

Just as in discrete-time, any continuous-time signal can be represented as a linear combination of
impulses. Because time is continuous, this ”sum” becomes an integral.
Proposition 1 (Continuous-Time Sifting Property). Any continuous-time signal x(t) can be written
as: Z ∞
x(t) = x(τ )δ(t − τ )dτ
−∞

Intuition: We can approximate x(t) with narrow rectangular pulses of width ∆ and height x(k∆).
As ∆ → 0, these pulses become impulses, and the sum becomes an integral. The value x(τ )dτ is
the infinitesimal ”weight” of the impulse at time τ .
5.2 The Convolution Integral
Now let’s apply this impulse representation to a continuous-time LTI system T . The logic is identical
to the discrete-time case.

Z ∞ 
y(t) = T {x(t)} = T x(τ )δ(t − τ )dτ (1)
−∞

Step 1: Use Linearity. The response to a weighted integral of inputs is the weighted integral
of individual responses: Z ∞
y(t) = x(τ )T {δ(t − τ )}dτ
−∞

Step 2: Use Time-Invariance. Define the impulse response h(t) = T {δ(t)}. For an LTI
system:
T {δ(t − τ )} = h(t − τ )
Step 3: Combine. Substituting back gives us the convolution integral:
Definition 1 (Continuous-Time Convolution Integral).
Z ∞
y(t) = x(τ )h(t − τ )dτ = (x ∗ h)(t)
−∞

Remark 1. The impulse response h(t) completely characterizes any continuous-time LTI system.
Given h(t), we can compute the output y(t) for any input x(t). This is not true for nonlinear
systems.

5.3 Physical Examples

1 −t/RC
1. RC Low-Pass Filter: The impulse response of an RC circuit is h(t) = RC e u(t). The
output voltage is the convolution of input voltage with this impulse response. The capacitor
”remembers” recent inputs more strongly; contributions from the distant past fade exponentially.

2. Mass-Spring-Damper System: For a mechanical system with mass m, spring constant p k,

1
and damping coefficient c, the impulse response is h(t) = mωd e −ζωn t sin(ωd t)u(t) where ωn = k/m,
√ p
2
ζ = c/(2 km), and ωd = ωn 1 − ζ . The system’s response to any force input is the convolution
of the force with this impulse response.

3. Heat Diffusion in a Rod: In a one-dimensional heat conduction problem, if we apply a

unit heat pulse at t = 0 at one end of a rod, the temperature distribution evolves as h(x, t) =
2
√ 1 e−x /(4αt) u(t) where α is thermal diffusivity. The temperature at any point is the convolution
4παt
of the heat input with this impulse response.

4. Acoustic Echo System: In a room with hard walls, sound reflects and creates echoes. The
impulse response h(t) = δ(t) + αδ(t − T ) + α2 δ(t − 2T ) + . . . represents the direct sound plus
exponentially decaying echoes with delay T and reflection coefficient α. The room’s response to any
sound is the convolution of the input with this impulse response.

2
 Input: Unit Step
x(t) = u(t)
1

h(t) = 1 −t/RC RC Circuit Impulse Response

RC e u(t)

1
h(t) = e−t u(t) (RC=1)

Output: y(t) = (1 − e−t )u(t)

y(t)
1
RC circuit step response
0.5 limt→∞ y(t) = 1

−1 −0.5 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5t

Figure 5.1: RC circuit convolution example with step input.

5.4 Computing Convolution: Flip-and-Slide Method

To compute convolution at a point t:

1. Plot signals x(τ ) and h(τ ) as functions of dummy variable τ .

2. Flip h(τ ) around zero to get h(−τ ).
3. Slide h(−τ ) by t to obtain h(t − τ ).
4. Multiply pointwise: x(τ ) × h(t − τ ).
5. Integrate the area under the product curve.

The integral gives the output y(t) at time t.

3
5.5 Numerical Examples
Example 1 (Two Rectangular Pulses). Given:
( (
1 0≤t≤1 1 0≤t≤1
x(t) = , h(t) =
0 otherwise 0 otherwise

Find y(t) = (x ∗ h)(t).

Solution: We need to consider different regions of overlap:

ˆt < 0: No overlap ⇒ y(t) = 0

ˆ0 ≤ t < 1: Partial overlap from τ = 0 to τ = t ⇒ y(t) = t
ˆ1 ≤ t < 2: Partial overlap from τ = t − 1 to τ = 1 ⇒ y(t) = 2 − t
ˆt ≥ 2: No overlap ⇒ y(t) = 0

t
 0≤t<1
Therefore: y(t) = 2 − t 1 ≤ t < 2

0 otherwise


Input: x(t) (rectangular pulse)

x(t)

Impulse Response: h(t) (identical rectangular pulse)

h(t)

Output: y(t) = (x ∗ h)(t) (triangular shape)

y(t)

1
Triangular result from identical rectangular pulses
0.5

−1 −0.5 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6t

Figure 5.2: Convolution of two identical rectangular pulses showing triangular result.

4
Example 2 (Exponential with Delayed Step). Given:

x(t) = e−t u(t), h(t) = u(t − 1)

Find y(t) = (x ∗ h)(t).

Solution:
Z ∞
y(t) = e−τ u(τ ) · u(t − τ − 1)dτ (2)
−∞

Since u(τ ) = 0 for τ < 0 and u(t − τ − 1) = 0 for τ > t − 1:

ˆ If t < 1: No overlap ⇒ y(t) = 0

ˆ If t ≥ 1: Integrate from τ = 0 to τ = t − 1

For t ≥ 1:
Z t−1 t−1
e−τ dτ = −e−τ 0 = 1 − e−(t−1)

y(t) = (3)
0

Therefore: y(t) = (1 − e−(t−1) )u(t − 1)

Input: Exponential decay

x(t) = e−t u(t)
1

Impulse Response: Delayed step

h(t) = u(t − 1)
1

Output: y(t) = (1 − e−(t−1) )u(t − 1)

y(t)
1
Delayed exponential growth
0.5

−1 −0.5 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5t

Figure 5.3: Convolution of exponential with delayed step function.

5
Example 3 (Two Exponentials with Different Decay Rates). Given:

x(t) = e−2t u(t), h(t) = e−3t u(t)

Find y(t) = (x ∗ h)(t).

Solution:
Z ∞
y(t) = e−2τ u(τ ) · e−3(t−τ ) u(t − τ )dτ (4)
−∞

For t ≥ 0:
Z t Z t
−2τ −3(t−τ ) −3t
y(t) = e ·e dτ = e eτ dτ (5)
0 0
=e −3t
[eτ ]t0 = e−3t (et − 1) = e −2t
− e−3t (6)

Therefore: y(t) = (e−2t − e−3t )u(t)

Input: e−2t u(t)

x(t) = e−2t u(t)
1

Impulse Response: e−3t u(t)

h(t) = e−3t u(t)
1

Output: y(t) = (e−2t − e−3t )u(t)

y(t)
1
Difference of exponentials
0.5

−0.5 0.5 1 1.5 2 2.5 3 3.5 4t

Figure 5.4: Convolution of two exponential functions with different decay rates.

6
Example 4 (Impulse Train with Exponential). Given:
2
X
x(t) = δ(t − n), h(t) = e−t u(t)
n=0

Find y(t) = (x ∗ h)(t).

Solution: Using the fact that convolution with impulses gives shifted impulse responses:
2
X 2
X
y(t) = h(t − n) = e−(t−n) u(t − n) (7)
n=0 n=0

This gives:
ˆ t < 0: y(t) = 0
ˆ 0 ≤ t < 1: y(t) = e−t
ˆ 1 ≤ t < 2: y(t) = e−t + e−(t−1) = e−t (1 + e)
ˆ t ≥ 2: y(t) = e−t + e−(t−1) + e−(t−2) = e−t (1 + e + e2 )

P2
Input: Impulse train n=0 δ(t − n)
x(t)
1

Impulse Response: Exponential decay

h(t) = e−t u(t)
1

Output: Sum of shifted exponentials

y(t)
3
Staircase of exponentials
2

−0.5 0.5 1 1.5 2 2.5 3 3.5 4t

Figure 5.5: Convolution of impulse train with exponential function.

Summary and Next Lecture

ˆ Continuous-time signals decompose into impulse integrals.
ˆ
R
CT LTI system output is convolution integral: y(t) = x(τ )h(t − τ )dτ .
ˆ Impulse response h(t) completely characterizes CT LTI systems.
ˆ Convolution computed using flip-and-slide method with integration.
ˆ Next time: Properties of convolution and LTI system characteristics.

7
 Signals and Systems Lecture 6
Dr. Ghandi Manasra and Ahmed Rabei Fall 2025

Lecture 6: LTI Properties and Equation-Described Systems

Reference
Oppenheim & Willsky, Signals and Systems, Chapter 2, Sections 2.3–2.4

Review of Lecture 5
ˆ Convolution integral for CT systems
ˆ Flip-and-slide method for computing convolution
ˆ Physical examples of convolution

6.1 Introduction
The core result from last time is the convolution representation:
Z ∞
y(t) = x(t) ∗ h(t) = x(τ )h(t − τ )dτ
−∞
∞
X
y[n] = x[n] ∗ h[n] = x[k]h[n − k]
k=−∞
Today, those formulas become a tool to analyze memory, invertibility, causality, and stability
purely from h, and to connect to systems given by constant-coefficient differential and difference
equations.

6.2 Convolution Properties

The convolution operation has several crucial algebraic properties that define the structure of LTI
systems.

6.2.1 Commutative Property

Convolution is commutative. The roles of the input and the impulse response are interchangeable.
x(t) ∗ h(t) = h(t) ∗ x(t), x[n] ∗ h[n] = h[n] ∗ x[n]
Intuition: You can flip and slide either x(t) or h(t) and get the same result.

6.2.2 Distributive Property

Convolution distributes over addition.
x(t) ∗ (h1 (t) + h2 (t)) = x(t) ∗ h1 (t) + x(t) ∗ h2 (t)
Intuition: The response to two LTI systems in parallel is the same as the response to a single
system whose impulse response is the sum of the two individual impulse responses.
6.2.3 Associative Property
Convolution is associative.

(x(t) ∗ h1 (t)) ∗ h2 (t) = x(t) ∗ (h1 (t) ∗ h2 (t))

Intuition: The overall impulse response of a cascade of LTI systems is the convolution of the
individual impulse responses. For LTI systems, the order of the cascade does not matter.
Remark 1. The commutativity and associativity of LTI systems is a very special and powerful
property. This is NOT true for nonlinear systems. A classic example: A ”square” block followed by
an ”add 3” block gives (x2 ) + 3. Reversing the order gives (x + 3)2 , which is completely different.

6.2.4 Demonstration: Commutative Property

Input Signal: x[n] Impulse Response: h[n]

x[n] h[n]
1 2

1 2 3 4 1 2 3 4
y1 [n] = x[n] ∗ h[n] y2 [n] = h[n] ∗ x[n]

3 3
2 2
1 1

1 2 3 4 n 1 2 3 4 n

Figure 6.1: Commutative property of convolution: x[n] ∗ h[n] = h[n] ∗ x[n].

6.2.5 Convolution with Sparse Impulse Responses

P
If h[n] = k ck δ[n − nk ], then: X
y[n] = ck x[n − nk ]
k

Convolution becomes a weighted sum of delayed copies, modeling echoes and multipath channels.

2
6.3 Correlation Functions
Correlation provides a measure of similarity or dissimilarity between two signals. It is a fundamental
tool in signal processing for pattern recognition, synchronization, and noise analysis.

6.3.1 Cross-Correlation Function

The cross-correlation function measures the similarity between two different signals as one is shifted
relative to the other.
Continuous-Time:
Z ∞ Z ∞
Rxy (t) = x(τ )y(τ + t)dτ = x(τ − t)y(τ )dτ
−∞ −∞

Discrete-Time:
∞
X ∞
X
Rxy [n] = x[k]y[k + n] = x[k − n]y[k]
k=−∞ k=−∞

Key Properties:

ˆ Rxy (t) = Rpyx (−t) (order matters)

ˆ |Rxy (t)| ≤ Rxx (0)Ryy (0) (Cauchy-Schwarz inequality)
ˆ If signals are uncorrelated: Rxy (t) ≈ 0 for all t

6.3.2 Autocorrelation Function

The autocorrelation function measures how much a signal matches a delayed version of itself.
Continuous-Time:
Z ∞ Z ∞
Rxx (t) = x(τ )x(τ + t)dτ = x(τ − t)x(τ )dτ
−∞ −∞

Discrete-Time:
∞
X ∞
X
Rxx [n] = x[k]x[k + n] = x[k − n]x[k]
k=−∞ k=−∞

Key Properties:

ˆ Rxx (t) = Rxx (−t) (even function)

ˆ Rxx (0) ≥ |R
R ∞xx (t)| for2 all t (maximum at zero lag)
ˆ Rxx (0) = −∞ |x(τ )| dτ (total energy)

6.3.3 Normalized Correlation Function

The normalized correlation function ρxy (t) provides a measure of similarity that is independent of
signal amplitude, ranging from -1 to +1. This technique is crucial for detecting known signals or
patterns within larger signals, especially when signal energy or power is unknown or varies.
The foundation is the cross-correlation function. However, the magnitude of Rxy (t) depends
on signal amplitudes. If you double the amplitude of both signals, the cross-correlation value
quadruples, even though their shape similarity remains [Link] overcome this amplitude

3
dependency, we use normalized correlation. The denominator is the geometric mean of the signal
energies.
Continuous-Time:
R∞
Rxy (t) x(τ )y(τ + t)dτ
ρxy (t) = p = qR −∞
Rxx (0)Ryy (0) ∞ R∞
|x(τ )|2 dτ |y(τ )|2 dτ
−∞ −∞

Discrete-Time:
P∞
Rxy [n] x[k]y[k + n]
ρxy [n] = p = qP k=−∞
∞
Rxx [0]Ryy [0] |x[k]|2 ∞ |y[k]|2
P
k=−∞ k=−∞

Key Properties:

ˆ −1 ≤ ρxy (t) ≤ +1 for all t (bounded range)

ˆ ρxy (t) = +1 when y(t) = kx(t) for k > 0 (perfect positive correlation)
ˆ ρxy (t) = −1 when y(t) = kx(t) for k < 0 (perfect negative correlation)
ˆ ρxy (t) = 0 when signals are uncorrelated (orthogonal)
ˆ ρxx (t) = 1 at t = 0 and |ρxx (t)| ≤ 1 for all t

Key takeaways:
The normalized correlation coefficient ρxy (t) always lies in the range of -1 to +1:

ˆ ρxy (t) = +1: Perfect positive correlation at time lag t. The two signals are identical in shape,
and one may be a positive scaled version of the other.
ˆ ρxy (t) = −1: Perfect negative correlation at time lag t. The signals have the exact same shape,
but one is inverted (a negative scaled version of the other).
ˆ ρxy (t) = 0: No correlation between the signals at that specific time lag. They are considered
orthogonal.
ˆ Values between 0 and 1: Indicate a degree of positive similarity. The closer to 1, the more
similar the signals are.
ˆ Values between -1 and 0: Suggest a degree of negative similarity. The closer to -1, the
more they are alike in shape but opposite in sign.

Example 1 (Normalized Cross-Correlation). For x(t) = e−t u(t) and y(t) = 2e−t u(t), find ρxy (t).
Solution: First, compute the autocorrelations:
Z ∞ Z ∞
−2τ 1
Rxx (0) = e dτ = , Ryy (0) = 4e−2τ dτ = 2
0 2 0

The cross-correlation is:

Z ∞ Z ∞
−τ −(τ +t) −t
Rxy (t) = e · 2e dτ = 2e e−2τ dτ = e−t
0 0

Therefore:
e−t
ρxy (t) = q = e−t
1
2 ·2

This shows perfect positive correlation at t = 0 (ρxy (0) = 1) that decays exponentially.

4
 Key Applications:

ˆ Pattern Recognition and Template Matching: A known ”template” signal is correlated

with a larger signal to find occurrences of the template. For example, in image processing, it
can find specific objects regardless of lighting conditions (which affect amplitude).
ˆ Radar and Sonar Systems: Used to detect returned signals (echoes) in the presence of
noise. By correlating the received signal with the transmitted pulse, the system can determine
distance to objects.
ˆ Communications: Helps synchronize receivers with incoming signals and detect specific bit
sequences in digital communications.
ˆ Biomedical Signal Processing: Identifies specific patterns in physiological signals like
ECGs or EEGs that might indicate medical conditions.
ˆ Feature Detection: Provides robust similarity measures unaffected by signal amplitude
variations.

6.3.4 Relationship Between Correlation and Convolution

Correlation and convolution are closely related. For cross-correlation:

Rxy (t) = x(t) ∗ y(−t)

This means correlation can be computed using convolution by time-reversing one of the signals.
Proof:
Z ∞
x(t) ∗ y(−t) = x(τ )y(−(t − τ ))dτ (1)
Z−∞
∞
= x(τ )y(τ − t)dτ (2)
−∞
= Rxy (t) (3)

6.3.5 Examples and Applications

Example 2 (Autocorrelation of Exponential Signal). For x(t) = 2e−3t u(t), find Rxx (t).
Solution: For t > 0:
Z ∞ Z ∞
−3τ −3(τ +t) −3t 2
Rxx (t) = (2e )(2e )dτ = 4e e−6τ dτ = e−3t
0 0 3

For t < 0: Z ∞
2
Rxx (t) = (2e−3τ )(2e−3(τ +t) )dτ = e3t
−t 3
2 −3|t|
Therefore: Rxx (t) = 3e for all t.
Example 3 (Cross-Correlation of Rectangular Pulses). For x(t) = u(t) − u(t − 1) and y(t) =
u(t − 32 ) − u(t − 52 ), find Rxy (t).
The cross-correlation shows the overlap between the two rectangular pulses as one slides past
the other, with maximum correlation when they align.

5
6.3.6 Periodic Signals (Power Signals)
For periodic signals with period T (CT) or N (DT), correlation functions are also periodic:
Continuous-Time:
1 T
Z
Rxy (t) = x(τ )y(τ + t)dτ
T 0
Discrete-Time:
N −1
1 X
Rxy [n] = x[k]y[k + n]
N
k=0

For periodic signals, Rxx (0) represents the average power rather than total energy.

6.4 System Properties via the Impulse Response

We can now connect the system properties from Lecture 3 to specific conditions on the impulse
response h(t) or h[n].

6.4.1 Memoryless Systems

An LTI system is memoryless if and only if its impulse response is a scaled impulse.

h(t) = Kδ(t) or h[n] = Kδ[n]

The system’s output can only depend on the current input if the impulse response itself is zero
everywhere except at time zero.

6.4.2 Invertible Systems

An LTI system is invertible if we can find an inverse system, hinv (t), that recovers the original
input. This means:
h(t) ∗ hinv (t) = δ(t) or h[n] ∗ hinv [n] = δ[n]
Practically, noninvertibility arises from nonminimum-phase zeros or spectral nulls that annihilate
components of inputs.

6.4.3 Causal Systems

An LTI system is causal if and only if its impulse response is zero for all negative time.

h(t) = 0 for t < 0 and h[n] = 0 for n < 0

The system cannot respond to an impulse before the impulse is applied. The convolution integral
for a causal system simplifies to:
Z t ∞
X
y(t) = x(τ )h(t − τ )dτ, y[n] = h[k]x[n − k]
−∞ k=0

6
Demonstration: Causality Analysis

h1 [n] = u[n]

h1 [n] = 0 for n < 0

Causal
1
0.5

h2 [n] = u[n + 1]

h2 [n] ̸= 0 for n < 0

Noncausal
1
0.5

−3 −2.5 −2 −1.5 −1 −0.5 0.5 1 1.5 2 2.5 3 3.5 4n

Figure 6.2: Causal vs noncausal systems based on impulse response.

6.4.4 BIBO Stable Systems

An LTI system is stable if and only if its impulse response is absolutely integrable (CT) or
absolutely summable (DT).
Z ∞ ∞
X
|h(t)|dt < ∞ and |h[n]| < ∞
−∞ n=−∞

Remark 2. Once the system is described by h(t), properties like causality, memorylessness,
invertibility, and stability can be checked directly from h(t) without further ambiguity; for stability
in particular, the tail of h(t) must decay so that the total ”effect” of past inputs remains finite.

7
6.4.5 Demonstration: Stability Analysis

Stable System
h1 [n] = (0.8)n u[n]

0.5

Unstable System
h2 [n] = u[n]

0.5

−1 1 2 3 4 5 6 7 8n

Figure 6.3: Stable vs unstable systems based on impulse response properties.

Stable System: h[n] = (0.8)n u[n]

∞
X 1
|0.8n | = =5<∞ (Stable)
1 − 0.8
n=0

Unstable System: Accumulator h[n] = u[n]

∞
X
|1| = ∞ (Unstable)
n=0

6.5 Step Response and Recovering the Impulse Response

6.5.1 Discrete-Time Case
With step input u[n] and step response s[n] = h[n] ∗ u[n], we can recover h via first difference:

h[n] = s[n] − s[n − 1] for all n

6.5.2 Continuous-Time Case

With step input u(t) and step response s(t) = h(t) ∗ u(t), we can recover h via differentiation:

d
h(t) = s(t)
dt
Rt
For causal LTI systems, s(t) = 0 h(τ )dτ and differentiability almost everywhere yields h.

8
6.6 Systems Described by Differential and Difference Equations
This is the most important and practical class of LTI systems in engineering applications.

6.6.1 Linear Constant-Coefficient Differential Equations (CT)

The general form is:
N M
X dk y(t) X dk x(t)
ak = bk
dtk dtk
k=0 k=0

where ak and bk are constants, and N is the order of the system.

Solution Structure and Auxiliary Conditions

The solution y(t) consists of two parts:

y(t) = yh (t) + yp (t)

where:

ˆ yh (t) is the homogeneous solution (natural response)

ˆ yp (t) is the particular solution (forced response)

Critical Requirements:

1. The differential equation alone does not completely specify the output in terms of the input.
2. We need auxiliary conditions corresponding to the values of y(t) and its first (N − 1)
derivatives at some point in time.
3. For the system to be linear, all auxiliary conditions must be zero.
4. For the system to be causal and LTI, we must assume initial rest.

Initial Rest Condition

Definition: If x(t) = 0 for t ≤ t0 , then y(t) = 0 for t ≤ t0 .
Under initial rest, the response for t > t0 can be calculated from the differential equation with
initial conditions:
dy(t0 ) dN −1 y(t0 )
y(t0 ) = = ··· = =0
dt dtN −1
This ensures the system is causal, linear, and time-invariant.

Normalized Form
When M = N , we can write:
"N N
#
1 X dk x(t) X dk y(t)
y(t) = bk − ak
a0 dtk dtk
k=0 k=1

This form is particularly useful for implementation and analysis.

9
Example 4 (RC Circuit). Consider the first-order RC circuit:
dy
RC +y =x
dt
With initial rest, this gives:
1 −t/(RC)
h(t) = e u(t)
RC
This system is causal, stable, and has step response 1 − e−t/(RC) .

6.6.2 Linear Constant-Coefficient Difference Equations (DT)

The general form is:
N
X M
X
ak y[n − k] = bk x[n − k]
k=0 k=0

Initial Rest Condition (DT)

For the system to be causal and LTI, we must assume initial rest:
ˆ If x[n] = 0 for n ≤ n0 , then y[n] = 0 for n ≤ n0 .
ˆ This ensures the system has no ”memory” of past inputs before the input begins.
ˆ Under initial rest, the system is causal, linear, and time-invariant.

Solution Structure and Properties

The solution approach is analogous to differential equations, but with an alternative recursive form:
"M N
#
1 X X
y[n] = bk x[n − k] − ak y[n − k]
a0
k=0 k=1

Key Properties:
1. The difference equation alone does not completely specify the output.
2. For linearity, auxiliary conditions must be zero.
3. If N ≥ 1: Infinite Impulse Response (IIR) system.
4. If N = 0: Finite Impulse Response (FIR) system.

FIR vs IIR Systems

FIR System (N = 0):
M
X bk
y[n] = x[n − k]
a0
k=0
Impulse response: (
bk
a0 0≤n≤M
h[n] =
0 otherwise
IIR System (N ≥ 1):
ˆ Has infinite-length impulse response
ˆ Output depends on past output values
ˆ More complex but more powerful

10
Cascade Interpretation
For M = N , the difference equation can be interpreted as a cascade of two systems:
Nonrecursive (FIR) part:
XN
w[n] = bk x[n − k]
k=0

Recursive (IIR) part:

N
" #
1 X
y[n] = w[n] − ak y[n − k]
a0
k=1

Example 5 (Moving Average Filter). Consider the 4-point moving average:

1
y[n] = (x[n] + x[n − 1] + x[n − 2] + x[n − 3])
4
This is an FIR system with impulse response:
1
h[n] = (δ[n] + δ[n − 1] + δ[n − 2] + δ[n − 3])
4
Example 6 (First-Order IIR Filter). Consider the recursive filter:

y[n] = αy[n − 1] + (1 − α)x[n]

This is an IIR system with impulse response:

h[n] = (1 − α)αn u[n]

For |α| < 1, the system is stable.

6.7 Implementation Structures

The normalized forms of LCCDEs can be implemented using block diagrams. Two common
structures are:

6.7.1 Direct Form I

This structure directly implements the cascade interpretation:

ˆ First: Nonrecursive (feedforward) section

ˆ Second: Recursive (feedback) section
ˆ Uses separate delay chains for input and output
ˆ Maps closely to the equation coefficients
ˆ Intuitive for understanding

11
 x[n] b0
w[n] 1/a0
+ + y[n]

b1 −a1
D + + D

b2 −a2
D + + D

.. .. .. ..
. . . .
bN −1 −aN −1
D + + D

bN −aN
D D

Figure 6.4: Direct Form I implementation structure for LCCDEs.

6.7.2 Direct Form II

This structure reverses the order and shares the delay chain:

ˆ First: Recursive section with shared delays

ˆ Second: Nonrecursive section
ˆ More efficient in terms of memory usage
ˆ Practical for embedded DSP designs
ˆ Requires careful numerical conditioning

12
 w[n]
x[n] b0 y[n]
+ +

−a1 b1
+ D +

−a2 b2
+ D +

..
.

−aN −1 bN −1
+ D +

−aN bN
+ D +

Figure 6.5: Direct Form II (Canonic) implementation structure for LCCDEs.

13
6.7.3 Continuous-Time Implementation Structures
The same implementation structures can be adapted for continuous-time systems by replacing delay
elements with integrators.

x(t) b0
w(t) 1/a0
+ + y(t)

R b1 −a1 R
+ +

R b2 −a2 R
+ +

.. .. .. ..
. . . .
R bN −1 −aN −1 R
+ +

R bN −aN R

Figure 6.6: Direct Form I implementation structure for continuous-time LCCDEs.

w(t)
x(t) b0 y(t)
+ +

−a1 R b1
+ +

−a2 R b2
+ +

..
.

−aN −1 R bN −1
+ +

−aN R bN
+ +

Figure 6.7: Direct Form II (Canonic) implementation structure for continuous-time LCCDEs.

14
6.6.3 Key Implementation Considerations
Direct Form I Advantages:
ˆ Clear separation between feedforward and feedback
ˆ Less sensitive to coefficient quantization
ˆ Easier to understand and debug
Direct Form II Advantages:
ˆ Minimal memory usage (half the delays)
ˆ More efficient for high-order systems
ˆ Standard in many DSP libraries
Implementation Challenges:
ˆ Numerical precision issues with high-order systems
ˆ Potential for limit cycles in fixed-point implementations
ˆ Need for careful coefficient scaling

6.8 Practical Examples and Applications

6.8.1 Continuous-Time Examples
Example 7 (RLC Circuit - Second Order). Consider an RLC circuit with input voltage x(t) and
output voltage y(t):
d2 y dy 1 dx
L 2 +R + y=
dt dt C dt
This is a second-order system with:
ˆ Natural frequency: ωn = √LC
1
q
ˆ Damping ratio: ζ = R2 CL
ˆ Can be underdamped, critically damped, or overdamped
Example 8 (Mechanical System - Mass-Spring-Damper). Consider a mass-spring-damper system:

d2 y dy
m 2
+c + ky = x(t)
dt dt
where m is mass, c is damping coefficient, and k is spring constant.

6.8.2 Discrete-Time Examples

Example 9 (Digital Filter Design). A common digital low-pass filter:

y[n] = 0.1x[n] + 0.9y[n − 1]

This is a first-order IIR filter with cutoff frequency determined by the coefficient 0.9.
Example 10 (Echo Cancellation). An echo cancellation system:

y[n] = x[n] + αx[n − D] + βy[n − D]

where D is the delay, and α, β are echo parameters.

15
6.8.3 System Analysis from LCCDEs
Stability Analysis:

ˆ CT: All poles must be in the left half-plane

ˆ DT: All poles must be inside the unit circle
ˆ For rational transfer functions, this is equivalent to BIBO stability

Remark 3. We are only introducing these today. We will spend a significant part of the course
developing powerful techniques (Fourier, Laplace, and z-transforms) to solve these equations and
analyze the systems they represent. For now, the goal is to recognize their form and their connection
to LTI systems under the initial rest condition.

6.9 Common Pitfalls and Clarifications

ˆ Initial rest vs linearity: A system is truly linear if its output is solely the result of the input
signal being processed. This relationship, described by pure convolution, only holds if the
system starts at rest (with zero stored energy). If a system has nonzero initial conditions, like
a charged capacitor, its output is a mix of the response to the input and a response from that
initial energy. Because this initial energy response doesn’t scale with the input, the overall
system no longer behaves linearly.
ˆ BIBO vs internal stability: There are two primary ways to judge if a system is stable.
The first, called BIBO stability, which is an external “black-box” view: if you feed it a
well-behaved signal that doesn’t go to infinity, you get a well-behaved signal out. The second,
internal stability, is an internal “glass-box” view that checks if any hidden component
within the system will blow up on its own. While it’s possible for a system to appear stable
on the outside while being internally unstable, this isn’t a concern for most standard systems
we encounter (specifically, rational and causal ones). For these systems, the two types
of stability are unified. We only need to check one thing: the location of the system’s
poles. If all the poles lie in the stable region of the complex plane (the left-half), the system
is guaranteed to be stable both inside and out.
ˆ Memoryless ̸= instantaneous hardware: Differentiators are not memoryless though they
”look” instantaneous.

6.10 Quick-Reference Formulas

R∞
ˆ Convolution: CT y(t) = −∞ x(τ )h(t − τ )dτ , DT y[n] = ∞
P
k=−∞ x[k]h[n − k]
ˆ Causality: h(t) =R0 (t < 0), h[n] = 0 (n < 0)
ˆ
P
BIBO stability: |h(t)|dt < ∞, |h[n]| < ∞
ˆ d
Step–impulse link: h(t) =Rdt s(t), h[n] = s[n] − s[n − 1] P
ˆ Correlation: CT Rxy (t) = x(τ )y(τ + t)dτ , DT Rxy [n] = x[k]y[k + n]
R (t) R [n]
ˆ Normalized correlation: CT ρxy (t) = √ xy , DT ρxy [n] = √ xy
Rxx (0)Ryy (0) Rxx [0]Ryy [0]
ˆ LCCDEs: CT ak y (k) = bk x(k) , DT
P P P P
ak y[n − k] = bk x[n − k]

16
6.11 Summary and Next Lecture
Today, we have connected the abstract properties of LTI systems to concrete conditions on their
impulse responses. This is a critical link between theory and practice. The key takeaways are:

ˆ Causality ⇔ h(t) = 0 for t < 0

ˆ Stability ⇔ the impulse response is absolutely integrable/summable
ˆ Convolution properties enable system analysis and design
ˆ LCCDEs describe the most important class of practical LTI systems

Next lecture: We will begin our journey into Fourier analysis. We will discover that complex
exponentials are ”eigenfunctions” of LTI systems, a property that turns the difficult operation of
convolution into simple multiplication. This will be our first step toward a new, and often much
easier, way to analyze signals and systems.

17