Discrete‑Time Signals & Systems —

Comprehensive Notes
Purpose: concise, exam‑oriented notes with clear definitions, formulas, worked examples and practice
problems (with solutions). Use this as a study sheet for discrete‑time signals, LTI systems, DTFT and
Z‑transform.

Quick table of contents

1. Discrete‑time signals — definitions & representations
2. Unit impulse and unit step
3. Basic signal operations (shifting, reversal, scaling, etc.)
4. LTI systems and impulse response; convolution
5. Discrete‑Time Fourier Transform (DTFT)
6. Z‑transform, ROC, poles & zeros
7. Causality and stability tests (examples)
8. Correlation (auto & cross) and spectral relations
9. Practical notes, computational methods, and sampling reminder
10. Practice problems with solutions

1. Discrete‑time signals — definition &

representations
A discrete‑time signal is a sequence of numbers indexed by an integer: x[n], n ∈ Z . It is a mapping
Z → C (or R ).

Common ways to represent a discrete signal:

• Graphical: a stem plot showing values at integer indices (most intuitive).

• Functional (analytical): e.g. x[n] = 0.5n u[n] or x[n] = sin(0.4πn) .
= 1, x[1] = 2, x[2] = 0 .
• Tabular: list values for a finite range, e.g. x[0]
• Sequence notation (vector-like): x[n] = {1, 2, 3} often means x[0] = 1, x[1] = 2, x[2] = 3 unless
an index origin is given.

x[n + N ] = x[n] for all n ; fundamental period if

Types of signals - Finite‑length vs infinite. - Periodic:
N exists. - Even/odd: even if x[n] = x[−n] , odd if x[n] = −x[−n] . - Real vs
smallest positive integer
∞
complex. - Energy signal: E = ∑n=−∞ ∣x[n]∣2 < ∞ . - Power signal: average power P =
limN →∞ 2N1+1 ∑n=−N ∣x[n]∣2 < ∞ .
N

1
2. Unit impulse δ[n] and unit step u[n]
Unit impulse (Kronecker delta)

δ[n] = {
1, n=0
0, n =0

Key properties - Sifting (sampling) property: x[n] δ[n − n0 ] = x[n0 ] δ[n − n0 ] . Intuitively, multiplying
by an impulse picks out a single sample. - Identity for convolution: x[n] ∗ δ[n] = x[n] . - Any sequence
can be decomposed as a weighted sum of shifted impulses:

∞
x[n] = ∑ x[k] δ[n − k].
k=−∞

So an LTI system is completely specified by its response to the impulse.

Unit step

u[n] = {
1, n≥0
0, n<0
∞
Relationship: u[n] = ∑k=0 δ[n − k] (formal); and δ[n] = u[n] − u[n − 1] .

Use in DSP: impulse response h[n] = output when input is δ[n] ; via superposition we obtain system output
for any input using convolution.

3. Operations on discrete‑time signals (with

examples)
1. Time‑shifting - Delay: x[n − n0 ] shifts the sequence right by n0 samples (output appears later). -
Advance: x[n + n0 ] shifts left by n0 samples.

Example: if x[n] = {1, 2, 3} with origin at n = 0 : - x[n − 1] = {… , 0, 1, 2, 3} (values move right).

2. Time‑reversal - x[−n] flips the sequence around n =0.

3. Time‑scaling (discrete) - For integer scaling x[an] with integer a : - If a = 2 , x[2n] is a compressed
version: only every other sample (original index jumps faster). This is a form of decimation (downsampling)
if interpreted as changing the sampling rate. - If a = 1/2 not meaningful directly unless one defines
interpolation; in discrete time, non‑integer scaling requires resampling/interpolation. - Two important

2
operations: decimation (downsample by integer M: xd [n] = x[M n] ) and upsampling (insert zeros: xu [n] =

{
x[n/M ], n = multiple of M
).
0, otherwise

4. Amplitude scaling y[n] = A x[n] multiplies every sample by constant A .

5. Addition / subtraction y[n] = x1 [n] ± x2 [n] (pointwise).

6. Multiplication (pointwise) y[n] = x1 [n]x2 [n] . Important: multiplication in time corresponds to

convolution in frequency (DTFT domain).

4. LTI systems, impulse response and convolution

Linear Time‑Invariant (LTI) system properties: - Linearity: output for a linear combination of inputs is the
same combination of outputs. - Time‑invariance: shifting input by n0 shifts the output by n0 .

Impulse response h[n] : output when input is δ[n] .

Convolution sum (discrete time):

∞ ∞
y[n] = x[n] ∗ h[n] = ∑ x[k] h[n − k] = ∑ h[k] x[n − k].
k=−∞ k=−∞

If x[n] and h[n] are finite length, the sum reduces to a finite sum.

Interpretation: express x[n] as weighted shifted impulses; for each impulse, system produces a shifted
impulse response; sum all contributions.

Worked example (finite sequences) Let x[n] = {1, 2} (meaning x[0] = 1, x[1] = 2 ) and h[n] = {3, 4}
(h[0]= 3, h[1] = 4 ). Compute y[n] = x[n] ∗ h[n] : - y[0] = x[0]h[0] = 1 ⋅ 3 = 3 - y[1] = x[0]h[1] +
x[1]h[0] = 1 ⋅ 4 + 2 ⋅ 3 = 4 + 6 = 10 - y[2] = x[1]h[1] = 2 ⋅ 4 = 8 So y[n] = {3, 10, 8} .

Why convolution matters: any LTI system output can be computed by convolution. For long sequences
convolution is computationally heavy; efficient approaches (FFT/overlap‑save, overlap‑add) exist.

5. Discrete‑Time Fourier Transform (DTFT)

Definition (DTFT) For sequence x[n] :

∞
X(ejω ) = ∑ x[n]e−jωn , −π ≤ ω ≤ π
n=−∞

3
The DTFT is periodic with period 2π : X(ejω ) = X(ej(ω+2π) ) .

Frequency variable: ω is radians per sample. To convert to cycles/sec (Hz), use f = ω/(2π) and account
for sampling frequency fs if converting to physical frequency.

Why move to frequency domain? - Many systems are easier to analyze because convolution in time
becomes multiplication in frequency:

Y (ejω ) = X(ejω ) ⋅ H(ejω ).

- A signal can be decomposed into sinusoidal components (superposition), and frequency response
describes how each sinusoid is scaled and phase shifted by the system.

1 N −1
Common DTFT pairs - δ[n] ↔ 1 - an u[n] ↔ 1−ae−jω , ∣a∣ < 1 - Finite sequence x[n] = ∑k=0 x[k] δ[n − k]
↔ polynomial in e−jω .

Properties (useful) Linearity, time‑shift (introduces phase factor e−jωn0 ), conjugation, convolution ↔
multiplication, modulation, Parseval (energy) relation.

6. Z‑transform and Region of Convergence (ROC)

Definition (one‑sided)

∞
X(z) = ∑ x[n]z −n , z ∈ C.
n=−∞

We usually write z = rejω . When r = 1 we are on the unit circle and X(ejω ) is DTFT if ROC includes unit
circle.

Why Z‑transform? - Good for analyzing stability and causality (convergence depends on r ). - Converts
convolution in time to multiplication in the z -domain: Y (z) = X(z)H(z) .

(z)
Poles and zeros - Express X(z) = N D(z) where zeros are roots of numerator, poles are roots of
denominator. - ROC: the set of z values for which the series converges; it is a region (ring) in the complex
plane.

ROC rules (summary) - For right‑sided sequences (causal or starting at some n0 and 0 for n<n0): ROC is
∣z∣ > rp where rp is radius of outermost pole. - For left‑sided sequences (anti‑causal): ROC is ∣z∣ < rp
(inside innermost pole). - For two‑sided sequences: ROC is a ring between poles (does not include poles). -
ROC cannot include any pole, and is always a connected region.

Physical significance - Causality: for an LTI system described by H(z) (rational), causal ⇔ ROC is exterior
of outermost pole (includes ∞ ). - Stability: BIBO stable ⇔ ROC includes the unit circle (every pole strictly
inside unit circle for causal systems).

4
 1+0.5z −1
Example: H(z) = 1−0.8z −1 . Pole at z = 0.8 (inside unit circle). If system is causal, ROC: ∣z∣ > 0.8 which
includes unit circle ⇒ stable and causal.

7. Tests & examples for causality and stability

Example system (difference equation):

y[n] = 1.8y[n − 1] − 0.72y[n − 2] + x[n] + 0.5x[n − 1].

Find stability.

Method: form homogeneous characteristic equation (set input to 0) and find poles: assume solution
yh [n] = rn gives

r2 − 1.8r + 0.72 = 0.

Compute discriminant: Δ = 1.82 − 4 ⋅ 0.72 = 3.24 − 2.88 = 0.36 ⇒ Δ = 0.6. Roots:

1.8 ± 0.6
r= = {1.2, 0.6}.
2
Poles at 1.2 and 0.6. Because one pole (1.2) lies outside unit circle, a causal realization (ROC |z|>1.2) will
not include unit circle ⇒ system is unstable.

Rule of thumb: For causal systems, check whether all poles are strictly inside the unit circle; if any pole has
magnitude ≥1 → not BIBO stable.

8. Correlation: autocorrelation & cross‑correlation

Cross‑correlation between sequences x[n], y[n] :

Rxy [τ ] = ∑ x∗ [n]y[n + τ ].
n

Autocorrelation: Rxx [τ ] = ∑n x∗ [n]x[n + τ ] .

∗
Properties: - Rxx [τ ] = Rxx [−τ ] (conjugate symmetry). - Rxx [0] gives signal energy (for finite energy
signals). - Wiener–Khinchin: Fourier transform of autocorrelation gives energy (or power) spectral density:
Sxx (ω) = F{Rxx [τ ]} = ∣X(ejω )∣2 (for energy signals / deterministic case).

5
9. Practical notes & computational methods
Convolution cost: direct convolution of length‑N and length‑M sequences costs O(NM) multiplies. For long
sequences use FFT‑based convolution (O((N+M) log(N+M))) via overlap‑save or overlap‑add.

Why use Z‑transform for convolution: multiply polynomials (or rational functions) instead of computing
sums; inverse Z‑transform yields time domain.

Sampling reminder: when converting analog x(t) to discrete (samples x[n] = x(nTs ) ), Nyquist states
sampling rate must satisfy fs > 2fmax to avoid aliasing; otherwise high frequency energy folds into
baseband irreversibly.

10. Practice problems (with solutions)

Problem 1 (convolution) Given x[n] = {1, 2} (nonzero at n=0,1) and h[n] = {3, 4} (nonzero at n=0,1).
Compute y[n] = x[n] ∗ h[n] .

Solution: (worked earlier) y[0] = 3, y[1] = 10, y[2] = 8 ⇒ y[n] = {3, 10, 8} .

Problem 2 (DTFT pair) Find DTFT of x[n] = δ[n] + 0.5 δ[n − 1] .

Solution:

X(ejω ) = 1 + 0.5e−jω .

Magnitude and phase follow from complex algebra; periodic in ω with period 2π .

Problem 3 (Z‑transform & ROC) Let x[n] = 0.9n u[n] . Find X(z) and ROC.

Solution:

∞
1
X(z) = ∑(0.9)n z −n = , ROC: ∣z∣ > 0.9.
n=0
1 − 0.9z −1

(Converges when ∣0.9z −1 ∣ < 1 i.e. ∣z∣ > 0.9 ).

Problem 4 (stability check) System given by y[n] = 1.8y[n − 1] − 0.72y[n − 2] + x[n] + 0.5x[n − 1] .
Is it stable?

Solution: characteristic roots are 1.2, 0.6 (see worked section). One root outside unit circle ⇒ unstable.

6
Problem 5 (inverse Z‑transform via polynomial division) Given X(z) = 3 + 10z −1 + 8z −2 , find x[n]
(assume causal, finite length).

Solution: read coefficients as x[0] = 3, x[1] = 10, x[2] = 8 so x[n] = {3, 10, 8} .

Problem 6 (autocorrelation property) Show that autocorrelation Rxx [τ ] has maximum at τ =0.

Sketch of solution: Rxx [0] = ∑ ∣x[n]∣2 = E (energy). For any τ , by Cauchy–Schwarz inequality,
∣Rxx [τ ]∣ ≤ Rxx [0] . So energy concentrated at zero lag.

11. Handy cheat‑sheet (properties)

Convolution: x[n] ∗ h[n] — commutative, associative, distributive.

DTFT: linearity, time shift: x[n − n0 ] ↔ e−jωn0 X(ejω ) .

Z‑transform: time shift: x[n − n0 ] ↔ z −n0 X(z) . Time reversal: x[−n] ↔ X(1/z) (watch ROC).

Stability/Causality summary: - Causal LTI ⇒ ROC is ∣z∣ > rp (outside outermost pole) and contains ∞ . -
Stable LTI ⇒ ROC includes unit circle. - Causal + stable ⇒ all poles strictly inside unit circle.

12. Recommended next steps / study tips

• Practice convolution and inverse Z‑transform problems by hand (polynomial multiplication and
partial fraction decomposition).
• Plot pole‑zero diagrams and identify ROC; practice reading stability and causality from diagrams.
• Learn use of FFT for fast convolution and spectral estimation.
• Work a few DTFT examples to get intuition about periodicity in frequency (2π periodicity) and how
time shifts affect phase.

If you want, I can: - convert these notes to a printable PDF, or - create a one‑page concise cheat sheet or
flashcards from these notes, or - add more worked examples (inverse Z via partial fractions, pole‑zero
sketches, filter design examples).