Chapter 1: Introduction to Communication Systems

1 Communication System Model

• A communication system is designed to transmit information.
• The process involves several key components that work in sequence:
1. Information Source: This is where the message originates, which could be human speech, a picture,
or text.
2. Input Transducer: This device converts the initial message into an electrical signal, known as the
baseband signal. It processes this signal for efficient transmission.
3. Transmitter: The transmitter modifies the baseband signal for transmission over a specific channel.
4. Channel: This is the physical medium through which the signal travels. Examples include coaxial
cables, optical fibers, or wireless links. The channel can introduce distortion and noise to the signal.
5. Receiver: The receiver’s function is to capture the signal from the channel and process it to reverse
the modifications made by the transmitter.
6. Output Transducer: This final component converts the electrical signal back into its original form,
such as sound or an image, so the recipient can understand it.
• The overall model can be visualized as a flow:
Input Message → Input Transducer → Transmitter → Channel → Receiver → Output Transducer →
Output Message

2 Digital vs. Analog Messages

Messages can be categorized into two main forms:
• Digital Messages: These are constructed from a finite set of symbols. A classic example is Morse code
telegraphy. Digital signals are generally more robust against noise compared to analog signals.
• Analog Messages: These are characterized by continuously varying data. An example would be the
temperature at a specific location, which can take any value within a range.

3 Digital Communications & Analog-to-Digital Conversion (ADC)

• Analog signals can be converted into a digital format using an Analog-to-Digital Converter (ADC).
• This process involves three main steps:
1. Filtering: An initial filter is used to process the analog signal v(t).
2. Sampling: The Sampler takes discrete samples of the analog signal m(t) at regular time intervals.
3. Quantizing: The Quantiser maps the amplitude of each sample to one of a finite number of levels.
• The key parameters of an Analog-to-Digital Converter (ADC) are:
– Resolution: The number of discrete values the ADC can produce.
– Conversion Time: The time it takes for the ADC to convert an analog sample to a digital value
(typically in microseconds to milliseconds).
– Sampling Frequency: The number of samples taken per second.
– Quantization Error: The difference between the actual analog value and the quantized digital
value.

4 Sampling of Signals
• The process of converting a continuous signal into a discrete signal is known as sampling.
• It involves two primary steps:
1. The signal m(t) is sampled in the time domain.
2. The amplitude of these signal samples is then quantized into a finite number of levels.

1
5 Sampling Theorem
• This is a fundamental principle in digital signal processing.

• The Sampling Theorem states that if the highest frequency in a signal’s spectrum is B, the original
signal can be perfectly reconstructed from its samples, provided that the sampling frequency is at least
2B.
• This minimum sampling rate (2B) is often referred to as the Nyquist rate.

• A signal can be perfectly reconstructed if the sampling frequency (fs ) is at least twice the highest
frequency (B) present in the signal’s spectrum.

fs ≥ 2B

6 Information and Channel Capacity

The ultimate goal of a communication system is to transmit information. Information theory, pioneered by
Claude Shannon, provides the mathematical framework to quantify this.

• Entropy (H(X)): Measures the average uncertainty or information content of a source. For a source
producing symbols {x1 , x2 , . . . , xm } with probabilities P (xi ), the entropy is:
m
X
H(X) = − P (xi ) log2 (P (xi )) (in bits/symbol)
i=1

• The Shannon-Hartley Theorem: This cornerstone theorem defines the theoretical maximum rate
at which information can be transmitted over a noisy channel with a specific bandwidth, known as the
channel capacity (C).  
S
C = B · log2 1 +
N
where:

– C: Channel capacity in bits per second (bps).

– B: Bandwidth of the channel in Hertz (Hz).
– S/N : The signal-to-noise ratio (a dimensionless quantity).
This theorem establishes that for any rate R < C, there exist coding techniques that allow for error-free
communication.

• Deeper Dive into Noise: While the document focuses on Additive White Gaussian Noise (AWGN),
other types of noise exist:
– Thermal Noise: From the random motion of electrons in conductors.
– Man-made Noise: Interference from motors, ignition systems, etc.
– Atmospheric Noise: From natural events like lightning.

7 Problem-Solving Technique: Calculating Entropy

1. Identify the Source Model: List all symbols and their probabilities P (xi ).
P
2. Verify Probabilities: Ensure P (xi ) = 1.
3. Apply the Entropy Formula: Calculate −P (xi ) · log2 (P (xi )) for each symbol.
4. Sum the Terms: The sum is the entropy H(X) in bits/symbol.
5. Calculate Information Rate: Rate (bps) = H(X)× Symbol Rate (symbols/sec).

2
Example: Source Entropy Calculation
Problem: A source produces four symbols with probabilities P(A)=0.5, P(B)=0.25, P(C)=0.125, P(D)=0.125.
The symbol rate is 1,000 symbols/sec. Calculate the source entropy and information rate.
Solution:
X
H(X) = − P (xi ) log2 (P (xi ))
i
= −[0.5 log2 (0.5) + 0.25 log2 (0.25) + 0.125 log2 (0.125) + 0.125 log2 (0.125)]
= −[0.5(−1) + 0.25(−2) + 0.125(−3) + 0.125(−3)]
= −[−0.5 − 0.5 − 0.375 − 0.375]
= −[−1.75] = 1.75 bits/symbol

Information Rate R:

R = H(X) × Symbol Rate = 1.75 bits/symbol × 1000 symbols/sec = 1750 bps

Chapter 2: Signals
1 Introduction to Signals
• A signal is fundamentally a collection of information or data, which can be represented as a function of
time.
• Examples range from television and telephone signals to economic data like monthly sales or stock market
prices.

2 Signal Analysis Techniques

• Cross-correlation and Autocorrelation are fundamental techniques for measuring the similarity be-
tween two signals or within a single signal.
– Cross-correlation measures the similarity between two different signals as a function of the time
lag applied to one of them.
∗ For continuous signals, f (t) and g(t):
The cross-correlation is given by the formula:
Z ∞ Z ∞
(f ⋆ g)(τ ) = f (t)g ∗ (t − τ )dt = f (t + τ )g ∗ (t)dt
−∞ −∞

∗
where g (t) is the complex conjugate of g(t).
∗ For discrete signals, x[n] and y[n]:
The cross-correlation is given by:
∞
X
(x ⋆ y)[m] = x[n]y ∗ [n − m]
n=−∞

– Autocorrelation is a special case of cross-correlation where a signal is compared with a time-shifted

version of itself. It is useful for finding repeating patterns in a signal.

3 Classification of Signals
Signals can be categorized based on several key characteristics:

3.1 Continuous-Time vs. Discrete-Time Signals

• Continuous-Time Signal: A signal that has a defined value for every moment in time.

3
 • Discrete-Time Signal: A signal that is defined only at specific, discrete points in time. It can be
obtained by sampling a continuous-time signal.

The Sampling Theorem: This crucial theorem, also known as the Nyquist-Shannon theorem, states that if the
highest frequency in a signal’s spectrum is B, the original continuous-time signal can be perfectly reconstructed
from its samples, provided the sampling rate is at least 2B samples per second.

3.2 Analog vs. Digital Signals

• A digital signal can be created from an analog signal through a process called quantization, where the
continuous amplitude range is divided into a finite number of intervals (L), and each sample’s amplitude
is mapped to the nearest quantization level.
• It’s important to note that quantization is a lossy process, meaning some information is lost and cannot
be recovered.
• A discrete-time digital signal is obtained by both sampling and quantizing a continuous-time analog
signal.
– Analog Signal: A signal whose amplitude can take on any value within a continuous range.
– Digital Signal: A signal whose amplitude can only take on a finite number of discrete values.

3.3 Periodic vs. Aperiodic Signals

• Periodic Signal: A signal g(t) is periodic if it repeats itself over a fixed time interval, known as the
period T0 . This can be expressed mathematically as:

g(t) = g(t + T0 ) for all t.

Common examples include sine and cosine waves (sin(ω0 t), cos(ω0 t)), where the angular frequency ω0 =
2π/T0 .
• Aperiodic Signal: A signal that does not repeat itself at any regular interval.

Euler’s Formula provides a fundamental relationship between complex exponentials and trigonometric func-
tions:
ejω0 t = cos(ω0 t) + j sin(ω0 t)

3.4 Energy vs. Power Signals

This classification is based on the signal’s energy and power content.
• Signal Energy (Eg ) is the integral of the squared magnitude of the signal over all time.
– For a real signal g(t): Z ∞
Eg = g 2 (t)dt
−∞

– For a complex signal g(t):

Z ∞ Z ∞
Eg = |g(t)|2 dt = g(t)g ∗ (t)dt
−∞ −∞

• Energy Signal:
– A signal is classified as an energy signal if its total energy is finite and non-zero (0 < Eg < ∞).
– These signals typically exist for a limited duration or decay to zero over time.
• Signal Power (Pg ) is the average energy of the signal per unit of time. It is used for signals with infinite
energy.

• Average Power Formula:

Z T /2
1
Pg = lim |g(t)|2 dt
T →∞ T −T /2

4
 • Power Signal: A signal is a power signal if its average power is finite and non-zero (0 < Pg < ∞).
Periodic signals are a common example of power signals.
• Periodic Signal Power: For a periodic signal with period T0 , the average power can be calculated over
a single period:
Z T0 /2
1
Pg = |g(t)|2 dt
T0 −T0 /2

A signal can be either an energy signal or a power signal, but it cannot be both.

3.5 Deterministic vs. Random (Stochastic) Signals

• Deterministic Signal: A signal whose physical description is completely known, meaning its value can
be predicted for any point in time.
• Random (Stochastic) Signal: A signal whose behavior is not predictable and can only be described in
terms of probabilities and statistical properties.

4 Examples of Energy and Power Signal Calculations

Example 1: Energy Signal
Consider a signal g(t) defined as: 
2
 for − 1 < t < 0
g(t) = 2e−t/2 for t ≥ 0

0 otherwise


To calculate its energy:

Z 0 Z ∞
Eg = 2
(2) dt + (2e−t/2 )2 dt
−1 0
Z 0 Z ∞
= 4dt + 4e−t dt
−1 0
= [4t]0−1 + [−4e−t ]∞
0
= (0 − (−4)) + (0 − (−4))
=4+4=8

Since the energy is a finite value (8), this is an energy signal.

Example 2: Power Signal
Consider a sinusoidal signal g(t) = A cos(ω0 t + θ).
To calculate its average power:
Z T /2
1
Pg = lim [A cos(ω0 t + θ)]2 dt
T →∞ T −T /2
T /2
A2
Z  
1 2 1 + cos(2x)
= lim [1 + cos(2(ω0 t + θ))]dt using cos (x) =
T →∞ T −T /2 2 2
"Z #
T /2 T /2
A2
Z
= lim 1dt + cos(2ω0 t + 2θ)dt
T →∞ 2T −T /2 −T /2

The integral of a cosine function over a very long period approaches zero. Therefore:

A2 T /2
Pg = lim · [t]−T /2
T →∞ 2T
A2
= lim ·T
T →∞ 2T
A2
=
2
Since the power is a finite, non-zero value, this is a power signal.

5
5 Operations on Signals
These operations manipulate the time variable t of a signal x(t).
• Time Shifting: y(t) = x(t − T )
– If T > 0, the signal is delayed (shifted to the right).
– If T < 0, the signal is advanced (shifted to the left).
• Time Scaling: y(t) = x(kt)
– If k > 1, the signal is compressed (sped up).
– If 0 < k < 1, the signal is expanded (slowed down).

• Time Inversion (Reflection): y(t) = x(−t)

This is a special case of time scaling where k = −1. The signal is reflected about the vertical axis.

6 Problem-Solving Steps
Problem 1: Determine the measure (energy or power) for a signal.
1. Step 1: Visually inspect the signal. If the signal is periodic or has a constant non-zero amplitude as time
approaches infinity, it is likely a power signal. If it is confined to a finite time interval or decays to zero,
it is likely an energy signal.

2. Step 2: Formulate the mathematical expression for the signal g(t).

3. Step 3: Calculate the signal energy Eg .
4. Step 4: If Eg is finite and non-zero, the signal is an energy signal. If Eg is infinite, proceed to calculate
the average power Pg .

5. Step 5: If Pg is finite and non-zero, the signal is a power signal.

Problem 2: Draw transformed versions of a signal g(t).
• a) g(t − 4): Shift the entire graph of g(t) to the right by 4 units.
• b) g(t + 6): Shift the entire graph of g(t) to the left by 6 units.

• c) g(3t): Compress the graph of g(t) horizontally by a factor of 3. Divide all time-axis values by 3.
• d) g(6 − t): This is a two-step transformation:
1. Step 1: Time Inversion. First, find g(−t) by reflecting g(t) about the vertical axis.
2. Step 2: Time Shifting. Then, find g(−(t − 6)). This means shifting the reflected signal g(−t) to
the right by 6 units.

7 Advanced Theory: Analysis of Random Signals

Real-world signals, especially noise, are random (stochastic). We analyze them using their statistical properties.

• The Wiener-Khinchin Theorem: A fundamental link between the time and frequency domains for
random signals. It states that the Power Spectral Density (PSD) of a random process is the Fourier
Transform of its autocorrelation function (R(τ )).
– Autocorrelation R(τ ): Measures the similarity of a signal with a time-shifted version of itself.
– Power Spectral Density (PSD): Describes how the power of a signal is distributed over frequency.
– Formula: Sx (f ) = F{Rx (τ )}
– Why it matters: This theorem allows us to find the spectrum of random signals like modulated
data streams and noise, which is essential for filter design and bandwidth calculations.

6
8 Problem-Solving Technique: Using the Wiener-Khinchin Theorem
1. Obtain the Autocorrelation Function R(τ ): This will typically be given in the problem statement.
2. Compute the Fourier Transform: Use standard Fourier transform tables or integration to find
F{R(τ )}. The result is the Power Spectral Density, Sx (f ) or PSD.
3. Interpret the PSD: Analyze the resulting spectrum to determine the signal’s frequency characteristics
(e.g., low-pass, band-pass).

Example: PSD from Autocorrelation

Problem: A noise signal has a triangular autocorrelation function given by R(τ ) = A(1 − |τ |/T ) for |τ | ≤ T ,
and R(τ ) = 0 otherwise. Find its PSD.
Solution:
• The autocorrelation function is a triangle function, which can be written as R(τ ) = A · Λ(τ /T ).
• The Wiener-Khinchin theorem states that P SD(f ) = F {R(τ )}.
• The Fourier transform of a triangle function is a squared-sinc function. This is a standard transform pair:

F{A · Λ(t/T )} = AT · sinc2 (f T )

sin(πx)
where sinc(x) = πx .

• Therefore, the Power Spectral Density is PSD(f ) = AT · sinc2 (fT).

• Analysis: This is a low-pass spectrum, with its maximum power concentrated at DC (f=0) and the
first nulls at f = ±1/T .

Chapter 3: Digital Communication Systems

1 Basic Concepts of Digital Communication Systems
• Digital communication systems operate by transmitting sequences of symbols from a discrete ”alphabet.”
• While historical examples include human writing and Morse code, modern systems predominantly use a
binary alphabet {0, 1}.
– Core Focus: The document centers on systems that transmit binary data sequences over wired
or wireless channels.
– Analog to Digital Conversion: To transmit analog information like voice or video, it must first be
converted into a binary data sequence through sampling and quantization, a process also known
as source coding.
– Examples of Digital Systems: Common examples include GSM/UMTS mobile networks, tele-
phone modems, optical fiber communications, LANs (wired and wireless), and GPS/Galileo satellite
systems.

Analog-to-Digital Conversion (ADC)

The process of converting analog signals into a digital format involves three main steps:
1. Filtering: An initial filter is used to process the analog signal v(t).
2. Sampling: The Sampler takes discrete samples of the analog signal at regular time intervals. The
Sampling Theorem states the sampling frequency fs must be at least twice the highest frequency B
present in the signal (fs ≥ 2B). This minimum sampling rate, 2B, is often referred to as the Nyquist
rate.
3. Quantizing: The Quantizer maps the amplitude of each sample to one of a finite number of levels. This
is a lossy process that introduces quantization error.

7
2 Key Parameters of Digital Communication Systems
The design and performance of a digital communication system are evaluated based on several key parameters:

• Bit Rate (Rb ): This is the number of bits transmitted per second, measured in bps.
• Bandwidth (B): Binary sequences must be converted into an analog waveform, s(t), for transmission.
The bandwidth, measured in Hertz (Hz), is the frequency range occupied by this waveform, characterized
by its Power Spectral Density (PSD), Gs (f ).

• Power (S): The received signal power, measured in Watts (W) or dBm, is crucial for reliable commu-
nication. Its effectiveness is determined by the Signal-to-Noise Ratio (SNR) at the receiver.
• Error Probability: In real-world channels, the received waveform r(t) is not identical to the transmitted
waveform s(t). This leads to errors in the received binary sequence uR . The Bit Error Probability
(BER), or P (uR [i] ̸= uT [i]), quantifies the likelihood of such errors.

• Complexity: This refers to the difficulty and cost of implementing the system.
• Delay (D): The time difference between the transmission and reception of information, measured in
seconds (s).
A typical system design involves balancing these parameters to meet specific requirements, such as achieving a
target bit rate and BER within given bandwidth, power, and complexity constraints.

3 Transmitting Binary Data

The process of converting a binary data sequence into a transmittable waveform involves several key concepts
and techniques.

3.1 Binary Data Sequences

• A binary data sequence, uT = (uT [0], uT [1], . . . , uT [i], . . . ), is a series of bits from the binary alphabet
Z2 = {0, 1}.
• Bit Duration (Tb ): Each bit in the sequence has a duration of Tb = 1/Rb seconds.

• Transmission Clock: A clock pulse with a frequency equal to the bit rate Rb synchronizes the trans-
mission.
• Statistical Properties: For analysis, random binary sequences are often assumed to be statistically
independent, with an equal probability of 0s and 1s.

3.2 From Bits to Waveforms: Bipolar NRZ Example

To be transmitted, the binary sequence is converted into a real-valued signal s(t). A common method is Bipolar
Non-Return-to-Zero (NRZ) encoding:
• A ’1’ bit is represented by a positive voltage signal (+V ).

• A ’0’ bit is represented by a negative voltage signal (−V ).

This can be expressed mathematically using a rectangular pulse PTb (t) of duration Tb :
• uT [i] = 1 is mapped to the signal +V · PTb (t − iTb ).

• uT [i] = 0 is mapped to the signal −V · PTb (t − iTb ).

4 Signal Sets, Labeling, and Waveform Construction

More advanced systems group bits together to transmit more complex signals, increasing efficiency.

8
4.1 Signal Sets
• A signal set M is a collection of m distinct waveforms, {s1 (t), s2 (t), . . . , sm (t)}.
• The size of the set is typically a power of two, m = 2k , where k is the number of bits each waveform
represents.
• Each signal si (t) has a finite duration, T = k · Tb .

4.2 Hamming Space and Binary Labeling

• A k-bit binary vector v is a sequence of k bits.
• The Hamming space Hk is the set of all 2k possible k-bit binary vectors.
• Binary Labeling is a one-to-one mapping e that assigns a unique waveform from the signal set M to
each unique vector in the Hamming space Hk .

e : Hk 7→ M

4.3 Constructing the Transmitted Waveform: A Step-by-Step Process

1. Split the data: The incoming binary data sequence uT is divided into a sequence of k-bit vectors,
vT = (vT [0], vT [1], . . . , vT [n], . . . ).
2. Map vectors to signals: Each vector vT [n] is mapped to its corresponding signal e(vT [n]) from the
signal set M. Each of these signals has a duration of T seconds.
3. Align signals in time: To create the final continuous waveform s(t), the individual signals are placed
sequentially in time. The signal corresponding to vector vT [n] is transmitted in the time interval from nT
to (n + 1)T .
• This time alignment is achieved by applying a time-shifting operator Tn such that the n-th signal
segment is s[n](t) = Tn (e(vT [n])) = e(vT [n])(t − nT ).

5. Practical Examples of Signal Sets (Modulation Schemes)

Various signal sets are used in practice, each with different characteristics.

• Bipolar Non-Return-to-Zero (NRZ): M = {+V PT (t), −V PT (t)}. Represents 1 bit (k = 1) with

positive or negative voltage for the full symbol duration T .
• Unipolar Non-Return-to-Zero (NRZ): M = {+V PT (t), 0}. Uses positive voltage for ’1’ and zero
voltage for ’0’.

• Bipolar Return-to-Zero (RZ): M = {+V PT /2 (t), −V PT /2 (t)}. Similar to Bipolar NRZ, but the pulse
only lasts for half the symbol duration, returning to zero for the second half.
• Unipolar Return-to-Zero (RZ): M = {+V PT /2 (t), 0}. A combination of Unipolar NRZ and the RZ
principle.

• m-PAM (Pulse Amplitude Modulation): Uses multiple amplitude levels to represent multiple bits.
For example, 4-PAM uses four voltage levels (e.g., +3V, +V, -V, -3V) to represent k = 2 bits per symbol.
• m-ASK (Amplitude Shift Keying): Varies the amplitude of a carrier wave. 4-ASK uses four different
amplitudes of a cosine wave to encode k = 2 bits.
• m-PSK (Phase Shift Keying): Varies the phase of a carrier wave. 2-PSK (BPSK) uses two phases (0◦
and 180◦ ), while 4-PSK (QPSK) uses four phases (e.g., 0◦ , 90◦ , 180◦ , 270◦ ) to encode k = 2 bits.
• m-FSK (Frequency Shift Keying): Uses different carrier frequencies to represent different symbols.
2-FSK uses two distinct frequencies, f1 and f2 , to encode the two binary values.

9
5 Solving the Exercise
To solve the exercise on page 44:

• Given:
– Binary sequence uT = (10011100 . . . )
– Bit rate Rb = 1 Mbps, which means Tb = 1/Rb = 1µs.
– Signal set M for a 4-PSK (or QPSK) system, with m = 4 signals. This means each signal represents
k = 2 bits, and the symbol duration is T = k · Tb = 2µs.
– Carrier frequency f0 = 1 MHz.
• Steps:
1. Group bits: Split the binary sequence into 2-bit vectors: (10), (01), (11), (00), . . .
2. Map vectors to signals: Use a binary labeling scheme to map each 2-bit vector to one of the four
signals in the set M. A common mapping (Gray coding) is:

(00) → s1 (t) = V PT (t) cos(2πf0 t)

(01) → s2 (t) = V PT (t) sin(2πf0 t) = V PT (t) cos(2πf0 t − π/2)
(11) → s4 (t) = −V PT (t) sin(2πf0 t) = V PT (t) cos(2πf0 t − 3π/2)
(10) → s3 (t) = −V PT (t) cos(2πf0 t) = V PT (t) cos(2πf0 t − π)

3. Draw the waveform:

– For t = 0 to 2 µs: The vector is (10). Draw the corresponding signal s3 (t). This is an inverted
cosine wave.
– For t = 2 to 4 µs: The vector is (01). Draw s2 (t). This is a sine wave.
– For t = 4 to 6 µs: The vector is (11). Draw s4 (t). This is an inverted sine wave.
– For t = 6 to 8 µs: The vector is (00). Draw s1 (t). This is a standard cosine wave.
– Continue this process for the rest of the sequence, creating a continuous waveform composed of
2 µs segments, each with a different phase determined by the corresponding 2-bit vector.

Chapter 4: Decision Theory & Signal Space Representation

1 The Communication Channel and Noise
• In a digital communication system, a binary data sequence (uT ) is first converted into a waveform (s(t))
and then transmitted over a channel.
• The channel inevitably introduces noise, corrupting the signal.

The AWGN Channel Model

A widely used and fundamental model for this process is the Additive White Gaussian Noise (AWGN)
channel.
• The received signal r(t) is the sum of the transmitted signal s(t) and a noise component n(t).

r(t) = s(t) + n(t)

• This model assumes the channel has a linear, time-invariant, and ideal frequency response (H(f ) = 1),
meaning it doesn’t distort the signal’s shape, only adds noise.

Characteristics of AWGN n(t):

1. Additive: The noise is simply added to the signal.
2. White: The noise has a constant Power Spectral Density (PSD) across all frequencies, given by
Gn (f ) = N0 /2. This implies that the noise power is uniformly distributed over the frequency spectrum.

10
 3. Gaussian: At any given time, the noise amplitude follows a Gaussian (or normal) distribution with
a zero mean.
Why is Noise Gaussian?
• The Central Limit Theorem provides the theoretical foundation for this assumption.
• It states that the sum of a large number of independent and identically distributed (i.i.d.) random variables
will tend to have a normal distribution, regardless of the original distribution of the individual variables.
• In a communication system, the total noise is the aggregation of noise from many different independent
sources (thermal noise, atmospheric interference, etc.), making the Gaussian model highly appropriate.
Ergodic Random Process:
• Noise is also modeled as an ergodic random process.
• This means that its statistical properties, such as its mean and variance, can be determined by observing
a single, sufficiently long sample of the process.

2 The Receiver’s Challenge and Signal Space Solution

• The primary challenge at the receiver is to accurately recover the original binary data sequence uT from
the noisy received waveform r(t).
• This problem can be broken down into two main steps:
1. Recover s(t) from r(t): This is the difficult part, as it involves distinguishing the signal from the
random noise.
2. Recover uT from s(t): This is relatively easy, as it’s a simple matter of reversing the one-to-one
mapping (labeling) used at the transmitter.
• To tackle the difficult first step, instead of processing the complex, continuous-time waveforms directly, it
is far more efficient to work with vectors.
• This is the core idea behind signal space representation.

3 Orthonormal Basis and Signal Space

Given a signal set M containing m possible transmitted waveforms, we can represent all these signals in a
signal space S that is spanned by an orthonormal basis B.

3.1 Properties of an Orthonormal Basis B

An orthonormal basis B = {b1 (t), b2 (t), . . . , bd (t)} is a set of d signals (where d ≤ m) that satisfies three key
conditions:
1. Mutually Orthogonal: The integral of the product of any two different basis signals over the symbol
duration T is zero. This means they are mathematically ”perpendicular.”
Z T
bj (t)bi (t)dt = 0 for j ̸= i
0

2. Unit Energy: The energy of each basis signal is equal to one.

Z T
b2i (t)dt = 1
0

3. Minimum Dimension (d): The basis contains the smallest number of signals necessary to represent
every signal in the original set M as a linear combination of the basis signals.
d
X
si (t) = sij bj (t) where sij are real-valued coefficients.
j=1

11
3.2 Constructing the Basis: The Gram-Schmidt Algorithm
The Gram-Schmidt algorithm provides a systematic method for constructing an orthonormal basis B from any
given signal set M.
Steps of the Algorithm:

1. Step 1 (First Basis Vector):

• Start with the first signal s1 (t).
• Calculate its energy E(s1 ).
• The first basis vector is b1 (t) = s1 (t)/ E(s1 ).
p

2. Step 2 (Second Basis Vector):

• Take the second signal s2 (t).
RT
• Calculate its projection onto the first basis vector: s21 = 0
s2 (t)b1 (t)dt.
• Subtract this projection from s2 (t) to get an intermediate vector b′2 (t) = s2 (t) − s21 b1 (t). This new
vector is orthogonal to b1 (t).
• If b′2 (t) is not zero, normalize it by its energy to get the second basis vector: b2 (t) = b′2 (t)/ E(b′2 ).
p

If b′2 (t) is zero, it means s2 (t) is linearly dependent on s1 (t), and no new basis vector is added.
3. Step i (General Step):
• For the i-th signal si (t), calculate its projections onto all previously found basis vectors bj (t).
• Subtract all these projections from si (t) to create an intermediate vector b′i (t) that is orthogonal to
all previous basis vectors.
• Normalize b′i (t) to get bi (t). If b′i (t) is zero, si (t) is a linear combination of the existing basis vectors,
and no new vector is added.
4. Final Step: Collect all the non-zero basis vectors bi (t) to form the complete orthonormal basis B.

4 Vector Representation and Euclidean Space

Once the orthonormal basis is established, any signal si (t) in the signal space can be uniquely represented by
a d-dimensional vector si .

• Signal to Vector: The components of the vector si = (si1 , si2 , . . . , sid ) are found by projecting the signal
si (t) onto each basis vector bj (t).
Z T
sij = si (t)bj (t)dt
0

• Vector to Signal: The original signal can be reconstructed from its vector representation using the linear
combination:
si (t) = si1 b1 (t) + si2 b2 (t) + · · · + sid bd (t)

This establishes an isomorphism between the signal space S and the d-dimensional Euclidean space Rd . The
signal set M can now be viewed as a constellation of m points in this d-dimensional space.

5 Signal Energy in Vector Space

Parseval’s Identity provides a simple way to calculate the energy of a signal using its vector representation.
The energy of a signal a(t) is equal to the sum of the squares of its vector components.
RT
• Waveform Energy: E(a) = 0 a2 (t)dt
Pd
• Vector Energy: E(a) = j=1 a2j

12
Average Energy of the Signal Set (Es ): This is the average energy of all signals in the set, assuming each
signal is transmitted with equal probability (P (si ) = 1/m).
m
1 X
Es = E(si )
m i=1

Energy per Bit (Eb ): This is the average energy required to transmit a single bit. If each signal represents k
bits, then:
Es
Eb =
k

6 Step-by-Step Solutions for Exercises

The document concludes with exercises applying these concepts to common modulation schemes. The general
procedure to solve them is as follows:

1. Construct the Orthonormal Basis (B):

• For simple signal sets, the basis can often be found by inspection. Look for a fundamental pulse
shape or carrier wave. Normalize it to have unit energy. For a 2D-constellation (like 4-PSK), you’ll
need two orthogonal basis signals (e.g., a cosine and a sine wave).
• For more complex cases, apply the Gram-Schmidt algorithm.

2. Represent the Signal Set in Vector Form:

• For each signal si (t) in the set M, find the coefficients sij that express it as a linear combination of
the basis vectors bj (t). These coefficients form the vector si .
• This can be done by calculating the projection integral or, more simply, by direct algebraic manipu-
lation.
3. Plot in Euclidean Space:
• Treat the vectors si as coordinates and plot them as points in a d-dimensional space (usually 1D or
2D for these examples). This plot is called the signal constellation.

4. Determine the Signal Space S:

• The signal space is the set of all possible linear combinations of the basis vectors you found.
5. Calculate Es and Eb :
• For each signal vector si , calculate its energy E(si ) using the sum of the squares of its components.
• Calculate the average signal energy Es .
• Determine k, the number of bits per symbol (m = 2k ).
• Calculate the energy per bit Eb = Es /k.

7 The AWGN Channel and the Receiver’s Problem

The foundation for decision theory in this context is the Additive White Gaussian Noise (AWGN) channel
model.
Key Properties of AWGN:
• It is an ergodic random process, meaning its statistical properties can be determined from a single,
long-term observation.
• The noise n(t) at any time instant is a Gaussian random variable with a zero mean.
• The Power Spectral Density (PSD), Gn (f ), is constant at N0 /2 for all frequencies. This ”white” prop-
erty implies that noise samples at different time instants, n(t1 ) and n(t2 ), are statistically independent.
This is a crucial property, as it means the noise affecting one symbol does not affect others.

13
The Problem: The receiver observes a waveform r(t) = s(t) + n(t) and must recover the original transmitted
signal s(t).
Intersymbol Interference (ISI):
• Because the noise n(t) is statistically independent from one time interval to the next, and each transmitted
signal segment s[n](t) is also independent, the received signal segment r[n](t) in a given time interval T
depends only on the signal and noise within that same interval.
• This ideal condition is known as having NO Intersymbol Interference (ISI), and it allows the receiver
to analyze and make a decision for each symbol independently.

8 Transforming the Problem into Signal Space

• The core strategy is to simplify the decision-making process by moving from continuous waveforms to
discrete vectors using the signal space representation.
1. Projecting Noise onto the Basis: The AWGN n(t) generally does not belong to the signal space
S. However, we can project it onto the orthonormal basis B = {bj (t)} of the signal space. The j-th
component of this projection is:
Z T
nj = n(t)bj (t)dt
0

2. Properties of Noise Components (nj ): It can be proven that these projection components nj
are Gaussian random variables with the following properties:
– Mean: E[nj ] = 0
– Variance: σ 2 = N0 /2
– They are statistically independent of each other.
3. Decomposing the Received Signal: The total noise n(t) can be split into two parts:
P
– ns (t) = j nj bj (t): The part of the noise that lies within the signal space S.
– e(t): The part of the noise that is outside (orthogonal to) the signal space S.
• The received signal is then r(t) = s(t) + ns (t) + e(t).

• The component e(t) contains no information about the transmitted signal and is statistically independent
of the components within the signal space.
• Therefore, it can be disregarded without losing information relevant to the decision.
• This leads to the concept of a sufficient statistic: the projection of the received signal onto the signal
space, rs (t) = s(t) + ns (t), contains all the necessary information to make an optimal decision.

9 The Decision Problem in Vector Form

By focusing only on the components within the signal space, the problem is transformed from waveforms to
vectors:
• The problem ”given r(t), recover s(t)” is simplified to ”given rs (t), recover s(t)”.

• This is equivalent to the vector problem: ”given r = sT + n, recover sT ”.

Here:
• r is the received vector.
• sT is the transmitted signal vector from the signal set M.

• n is the noise vector, with components nj .

Each component of the received vector is rj = sj + nj , which is a Gaussian random variable with a mean sj
and a variance of N0 /2.

14
10 Decision Criteria for Optimal Detection
The goal is to choose a signal sR from the set M that minimizes the probability of a symbol error, P (sR ̸= sT ).

10.1 Maximum a Posteriori (MAP) Criterion

• The MAP criterion aims to maximize the a posteriori probability, which is the probability of a par-
ticular transmitted signal si given that a specific vector r was received.
• The decision rule is:
Choose sR = si that maximizes P (sT = si |r)

• This criterion is optimal as it directly minimizes the error probability.

10.2 Maximum Likelihood (ML) Criterion

• Using Bayes’ Theorem, the posterior probability can be expressed as:

P (r|sT = si )P (sT = si )
P (sT = si |r) =
P (r)

• P (r|sT = si ) is the likelihood: the probability of receiving r given that si was sent. For continuous
variables, this is represented by the probability density function (PDF) fr (ρ|sT = si ).

• P (sT = si ) is the a priori probability: the probability of si being transmitted, before any observation
is made.
• If we assume that all transmitted signals are equally likely (a very common assumption in digital
communications), then P (sT = si ) is a constant 1/m.

• In this case, maximizing the MAP criterion is equivalent to maximizing the likelihood. This leads to the
Maximum Likelihood (ML) criterion:

Choose sR = si that maximizes the likelihood fr (ρ|sT = si )

10.3 From ML to Minimum Distance

• For an AWGN channel, the received vector r, given that si was sent, has a d-dimensional Gaussian PDF.

• Maximizing this PDF is equivalent to minimizing the exponent term, which is the sum of squared differ-
ences between the received components ρj and the signal components sij .
• ML Expression: " Pd !#
1 j=1 (ρj − sij )2
sR = arg max √ exp −
si ( πN0 )d N0

• This simplifies to:  

Xd
sR = arg min  (ρj − sij )2 
si
j=1

• This final expression is the formula for the squared Euclidean distance between the received vector ρ
and the signal vector si .

• Therefore, for an AWGN channel with equally likely signals, the optimal ML decision rule is the Minimum
Distance Criterion:
Choose the signal point si in the constellation that is closest in Euclidean distance to
the received point r.

15
11 Voronoi Regions: A Geometric Interpretation
The minimum distance criterion partitions the d-dimensional signal space into Voronoi regions (or decision
regions).
• For each signal point si in the constellation, its Voronoi region V (si ) is the set of all points in the space
that are closer to si than to any other signal point sj .
• The decision rule is then simply: if the received vector r falls within the Voronoi region V (si ), the receiver
decides that si was the transmitted signal.

12 The Signal Space Receiver: A Three-Step Process

Assuming a signal ρ(t) is received over a time interval [0, T ], an optimal receiver based on signal space concepts
performs the following three steps:
1. Compute Projections:
• The received waveform ρ(t) is projected onto each of the d orthonormal basis functions bj (t) of the
signal space.
• This generates a d-dimensional received vector ρ = (ρ1 , . . . , ρd ), where each component is calculated
by the integral:
Z T
ρj = ρ(t)bj (t)dt
0

2. Make a Decision:
• The received vector ρ is fed into a decision device that applies the ML criterion.
• This is equivalent to finding which of the possible transmitted signal vectors si in the set M is closest
to ρ in terms of Euclidean distance (the minimum distance or Voronoi region criterion).
• The chosen vector is denoted as sR .
3. Recover the Bits: The chosen signal vector sR is then mapped back to its corresponding binary infor-
mation vector uR using the inverse of the mapping e that was used at the transmitter: uR = e−1 (sR ).

13 Implementation with Matched Filters

• The projection integral in Step 1 can be implemented in a more practical and elegant way using a Matched
Filter (MF).
– Concept: A matched filter is a linear filter whose impulse response h(t) is ”matched” to a specific
signal, in this case, one of the basis functions bj (t).
– Impulse Response: For a basis signal bj (t) defined over [0, T ], the corresponding matched filter
has an impulse response that is a time-reversed and shifted version of the basis signal:

h(t) = bj (T − t)

• Output:
– When the received signal ρ(t) is passed through this filter, the output y(t) is the convolution of ρ(t)
and h(t).
– The key insight is that if this output y(t) is sampled at the precise time t = T , the resulting value is
exactly equal to the projection ρj :
Z T
y(T ) = ρ(τ )bj (τ )dτ = ρj
0

• Therefore, a bank of d matched filters, one for each basis function, can be used to compute the d compo-
nents of the received vector ρ.
• This approach is often simpler to implement than using correlators (multiplier-integrator blocks).

16
14 The Complete Receiver and Symbol Synchronization
A receiver must process a continuous stream of symbols, not just a single one in the interval [0, T ].
• Handling Continuous Data:
– For any symbol interval [nT, (n + 1)T ], the receiver uses the same set of matched filters.
– The output of the j-th filter is sampled at t = (n + 1)T to get the projection ρj [n] for the n-th
symbol.
– This means the bank of matched filters is followed by a sampler that operates at the symbol rate,
R = 1/T .
• The Challenge of Symbol Synchronization:
– This process relies on a critical assumption: the receiver knows the exact symbol rate R and the
precise moments t = (n + 1)T to sample the filter outputs (the timing phase).
– In practice, the clocks at the transmitter and receiver are never perfectly identical.
– Symbol Synchronization is the essential process of estimating and recovering the correct symbol
rate and timing phase from the received signal itself.
– Accurate synchronization is crucial for correctly computing the projections and minimizing detection
errors.

15 The Correlation Receiver: An Alternative Implementation

The ML decision rule can be mathematically reformulated to create an alternative receiver structure.
• Derivation: Starting with the minimum Euclidean distance criterion, arg min ||ρ − si ||2 , the expression
can be expanded: X X X
||ρ − si ||2 = ρ2j + s2ij − 2 ρj sij
P 2
– Since ρj is the energy of the received signal and is the same for all si , it can be ignored in the
minimization.
P 2
– The term sij is simply the energy of the signal E(si ).
– Minimizing the original expression is therefore equivalent to maximizing the remaining term:
X 
1
sR = arg max ρj sij − E(si )
si 2

• The Correlation Term: The term

P
ρj sij can be shown to be equal to the correlation between the
received waveform ρ(t) and the signal waveform si (t):
Z T
ρ(t)si (t)dt
0

• Correlation Receiver Structure: This leads to the Correlation Receiver, which implements the
ML criterion as follows:
1. Correlate the received signal ρ(t) with every possible transmitted signal si (t) in the signal set M.
2. For each correlation result, subtract half the energy of the corresponding signal, 21 E(si ).
3. Choose the signal si that results in the maximum value.

16 Comparison of Receiver Types

• The Matched Filter receiver and the Correlation receiver are mathematically equivalent ways of im-
plementing the optimal ML detector.
• They will always yield the same decision.
• The choice between them is often based on implementation complexity.

17
 – Matched Filter Receiver:
∗ Requires d matched filters (where d is the dimension of the signal space).
∗ Has one decision unit that calculates d-dimensional Euclidean distances.
∗ Generally more efficient when the number of signals m is much larger than the dimension d (i.e.,
m ≫ d).
– Correlation Receiver:
∗ Requires m correlators (one for each possible signal).
∗ Has one decision unit that finds the maximum among m values.
∗ Can be simpler when m is small.

17 Signal Space Representation

Instead of dealing with complex waveforms, we can represent a set of m signals in a d-dimensional space (where
d ≤ m) using a set of orthonormal basis functions. This transforms the problem from continuous-time signals
to simple vectors.

• Orthonormal Basis: A set of basis signals {bi (t)} that are mutually orthogonal and have unit energy.
This means: Z T (
1 j=i
bj (t)bi (t)dt =
0 0 j ̸= i

• Gram-Schmidt Algorithm: A systematic procedure for constructing an orthonormal basis from any
given set of signals.
• Signal Constellation: A plot of the resulting signal vectors in this d-dimensional Euclidean space. The
geometry of this constellation (e.g., the distance between points) is critical for system performance.

18 Solving the Exercise: A Practical Example

For the exercise involving a bipolar signal set M = {PT (t), −PT (t)}:
1. Transmitted Waveform: For the bit sequence 1010..., the waveform would be a square wave alternating
between +1 and -1 in intervals of T .
2. Matched Filter:
• The basis function is b1 (t) = √1
T
PT (t).
• The matched filter impulse
√ response is h(t) = b1 (T − t), which is a rectangular pulse from t = 0 to
t = T with amplitude 1/ T .
3. Matched Filter Output (Noiseless):
• When the transmitted square wave√ is passed
√ through this filter, the output will be a sequence of
triangular pulses, peaking at + T or − T at the end of each symbol interval t = (n + 1)T .
√ √ √
4. Verification: Sampling the triangular output at t = T, 2T, 3T, . . . will yield the values + T , − T , + T , . . . .
These values directly correspond to the transmitted symbols, confirming the filter’s operation.

Chapter 5: Receiver Performance – Probability of Error

1 The Receiver’s Task and Performance Metrics
• The fundamental problem for a receiver operating over an AWGN channel is to recover the original binary
data sequence uT from a noisy received waveform r(t).
• The optimal strategy, as previously established, is to use the minimum distance criterion (or the
equivalent Voronoi region criterion) on the vector representation of the received signal.
• The quality of this process is measured by the probability of making an incorrect decision.

18
 • There are two primary types of error probabilities:
1. Symbol Error Rate (SER) or PS (e): The probability that the receiver chooses a symbol sR that
is different from the transmitted symbol sT .

SER = P (sR [n] ̸= sT [n])

2. Bit Error Rate (BER) or Pb (e): The probability that a decoded bit uR [i] is different from the
transmitted bit uT [i].
BER = P (uR [i] ̸= uT [i])

2 Key System Parameters for Performance Analysis

To calculate SER and BER, several key parameters are used:
• Rates:
– Bit Rate (Rb ): The number of bits transmitted per second.
– Symbol Rate (R): The number of symbols transmitted per second. If each symbol represents k
bits, then R = Rb /k.
– Durations: Bit duration Tb = 1/Rb , Symbol duration T = 1/R = k · Tb .
• Energy and Power:
– Energy per Bit (Eb ): The average energy required to transmit one bit.
– Energy per Symbol (Es ): The average energy required to transmit one symbol.
– Signal Power (S): The average energy per unit time, S = Eb · Rb = Es · R.
• Noise:
– Noise Power Spectral Density (N0 ): The noise power per unit of bandwidth (W/Hz).
– Noise Power (N ): The total noise power over a given signal bandwidth B, N = N0 · B.
• Key Performance Ratios:
– Signal-to-Noise Ratio (S/N): The ratio of signal power to noise power.
– Eb /N0 : The ratio of energy per bit to the noise power spectral density. This is the most fundamental
figure of merit for a digital communication system. The system’s performance (BER) is almost always
expressed as a function of Eb /N0 . The relationship between these ratios is:

S/N = (Eb /N0 ) · (Rb /B) = (Eb /N0 ) · η

where η is the spectral efficiency.

3 Computation of Symbol Error Rate (SER)

• The SER is calculated by averaging the probability of error over all possible transmitted symbols.
• Assuming each of the m symbols is equally likely to be transmitted (with probability 1/m):
m
1 X
PS (e) = PS (e|sT = si )
m i=1

where PS (e|sT = si ) is the conditional probability of a symbol error, given that symbol si was transmitted.
• An error occurs if si is sent, but the received vector ρ falls outside of si ’s Voronoi region, V (si ).
• This can be expressed in two ways:
1. Complementary Probability:

PS (e|sT = si ) = 1 − P (ρ ∈ V (si )|sT = si )

19
 2. Union of Probabilities:
X
PS (e|sT = si ) = P (ρ ∈ V (sj )|sT = si )
j̸=i

This means we calculate the probability that the noise was large enough to push the received vector into the
decision region of a different symbol.

4 Computation of Bit Error Rate (BER)

• The BER is related to the SER, but it also depends on the specific binary labeling (mapping from bit
vectors to symbols).
• When a symbol error occurs (e.g., sj is received instead of si ), the number of resulting bit errors is given
by the Hamming distance, dH (vj , vi ), between the corresponding bit vectors.

• The average BER is then:

m
1 X
Pb (e) = Pb (e|sT = si )
m i=1
where the conditional bit error probability is:
X dH (vj , vi )
Pb (e|sT = si ) = · P (ρ ∈ V (sj )|sT = si )
k
j̸=i

• Each term in the sum represents the probability of confusing symbol si with sj , weighted by the number
of bit errors that this specific confusion causes.

5 The Complementary Error Function (erfc)

• The probability of a Gaussian random variable exceeding a certain value is calculated using the comple-
mentary error function, ‘erfc‘. The BER for many modulation schemes follows a characteristic ”waterfall”
curve.
• To calculate the probability of a Gaussian random variable falling into a certain region (which is what’s
needed to evaluate P (ρ ∈ V (sj )|sT = si )), the complementary error function (erfc) is used.
• For a Gaussian random variable n with mean µ and variance σ 2 , the probability that n exceeds a value x
is:  
1 x−µ
P (n > x) = · erfc √
2 2σ

• In our case, the noise components have a mean of 0 and a variance of σ 2 = N0 /2. This simplifies the
formula to:  
1 x
P (n > x) = · erfc √
2 N0

• For the binary antipodal signaling case (like BPSK), this simplifies to a classic formula relating BER to
the energy-per-bit to noise-power-spectral-density ratio (Eb /N0 ).
r !
1 Eb
BER for BPSK = erfc
2 N0

6 Performance of Binary Antipodal Signals

• A fundamental case is binary antipodal signaling, a 1-dimensional scheme with two signals symmetric
about the origin, such as s1 = (+A) and s2 = (−A).

• This is the vector representation for BPSK.

20
 – Decision Boundary: The Voronoi regions are divided by the origin. If the received value ρ1 > 0,
the receiver decides s1 ; if ρ1 < 0, it decides s2 .
– Calculating Conditional SER:
∗ Assume s1 is sent. The received value is ρ1 = A + n1 . An error occurs if ρ1 < 0, which means
A + n1 < 0, or n1 < −A.
∗ Using the erfc function and noting that the Gaussian distribution of n1 is symmetric (P (n1 <
−A) = P (n1 > A)), we get:
 
1 A
PS (e|sT = s1 ) = P (n1 > A) = · erfc √
2 N0

∗ By symmetry, the probability of error given s2 was sent is the same.

• Final SER: Since the conditional error probabilities are identical, the total SER is:
 
1 A
PS (e) = · erfc √
2 N0

• Expressing in terms of Eb /N0 :

– The energy of each signal is E(s1 ) = E(s2 ) = A2 .
– The average energy per symbol is Es = A2 .
– Since k = 1 for binary signaling, the energy per bit is Eb = Es = A2 .
√
– Substituting A = Eb , the final formula is:
r !
1 Eb
SER = · erfc
2 N0

• BER for Antipodal Signals: For binary signaling, a symbol error always results in one bit error.
Therefore, BER = SER.
r !
1 Eb
BER = · erfc
2 N0

• This classic ”waterfall” curve shows that the BER decreases exponentially as the Eb /N0 ratio increases.
• Importantly, this performance depends only on the distance between the signal points in the vector space
(A) and the noise level (N0 ), not on the specific waveform used to create the basis function (e.g., a simple
pulse or a modulated carrier).

7 Optimal Detection
• In an Additive White Gaussian Noise (AWGN) channel, the optimal decision rule is to choose the signal
point in the constellation that is closest in Euclidean distance to the received signal vector.

• This is known as the Minimum Distance Criterion. Geometrically, this partitions the signal space
into Voronoi regions around each constellation point.

Chapter 6: Calculating and Evaluating Signal Spectra

1 Receiver Performance: BER Comparison & Asymptotic Analysis
• The Bit Error Rate (BER) is a key metric for the performance of a digital communication system.
• For an Additive White Gaussian Noise (AWGN) channel, the BER is fundamentally a function of the
ratio of energy per bit to noise power spectral density, Eb /N0 .

21
1.1 BER for Binary Antipodal Signals
• As a benchmark, the BER for binary antipodal signals (like BPSK) is given by:
r !
1 Eb
Pb (e) = · erfc
2 N0

• A crucial insight is that different signal waveforms (e.g., a baseband pulse vs. a modulated carrier) that
have the same underlying vector space representation will achieve the exact same BER performance.
• The performance is determined by the geometry of the signal constellation in vector space, not the specific
shape of the basis signals.

1.2 Comparison: Antipodal vs. Orthogonal Signaling

• The choice of signal space geometry has a significant impact on performance.
• A comparison between antipodal and orthogonal signal spaces reveals:
q 
Eb
– Antipodal BER: Pb (e)antipodal = 21 · erfc N0
q 
Eb
– Orthogonal BER: Pb (e)orthogonal = 12 · erfc 2N0

Conclusion: Antipodal signaling has demonstrably better performance.

• For a fixed Eb /N0 , an antipodal system will achieve a significantly lower BER. For example, at Eb /N0 = 12
dB, an antipodal system might achieve a BER of 10−8 , while an orthogonal system only reaches 5 × 10−5 .
• To achieve a target BER (e.g., 10−6 ), an antipodal system requires a lower Eb /N0 . This difference is 3
dB, meaning an orthogonal system requires twice the signal power to achieve the same performance.
• This 3 dB advantage has practical implications.
• Based on the Friis transmission equation,
√ which relates received power to transmission distance, an
antipodal system can either transmit 2 times farther with the same power or reduce its transmit power
by half for the same distance.

1.3 Asymptotic Performance (High Eb /N0 )

• At high Eb /N0 ratios (low noise), errors are most likely to occur between adjacent signal points in the
constellation.
• This allows for an approximation of the error probability based on the geometry of the signal set.
– Minimum Distance (dmin ): The smallest Euclidean distance between any two signal points in the
constellation. A larger dmin generally leads to better performance.
– Multiplicity (Amin ): The number of signal pairs separated by the minimum distance.
– Input Multiplicity (wmin ): The average number of bit errors that occur when a symbol is mistaken
for one of its nearest neighbors.
• The asymptotic error probabilities are approximately:
 
– Symbol Error Rate (SER): PS (e) ≈ Amin · erfc √d4Nmin
0
 
wmin d
– Bit Error Rate (BER): Pb (e) ≈ 2k · erfc √4Nmin
0

1.4 Gray Labeling

• To optimize BER, we use Gray labeling.
• This is a technique for assigning bit patterns to signal points such that adjacent points (those separated
by dmin ) differ by only a single bit.
• This minimizes the number of bit errors that occur during the most likely symbol errors, effectively
minimizing the wmin term and, consequently, the overall BER.

22
2 Calculating and Evaluating Signal Spectra
• The Power Spectral Density (PSD), Gs (f ), describes how the power of a transmitted signal s(t) is
distributed over different frequencies.
• Understanding the PSD is essential for determining the bandwidth a signal will occupy and ensuring it is
suitable for the given communication channel.

2.1 General Theory of PSD for Linearly Modulated Signals

A common way to represent a transmitted signal is as a sequence of pulses:
X
s(t) = a[n] · p(t − nT )
n

where a[n] is a sequence of random variables representing the symbol amplitudes, and p(t) is the pulse shape.
The PSD of such a signal is given by the formula:

|P (f )|2
Gs (f ) = Sa (f ) ·
T
where:
• T is the symbol period.
• P (f ) is the Fourier transform of the pulse p(t).
• Sa (f ) is derived from the autocorrelation of the symbol sequence a[n].
Case 1: Statistically Independent Symbols with Zero Mean
This is a common and important case, corresponding to antipodal signal spaces (like BPSK or bipolar NRZ)
centered at the origin.
• The symbols a[n] are statistically independent.
• The mean of the symbols is zero (µa = 0).
In this scenario, Sa (f ) simplifies to the variance of the symbols, σa2 . The PSD formula becomes:

σa2
Gs (f ) = · |P (f )|2
T
This shows that the PSD of the transmitted signal is directly proportional to the squared magnitude of the
pulse’s Fourier transform.
Example: Rectangular Pulse (Bipolar NRZ)
• Pulse (p(t)): A rectangular pulse of duration T , p(t) = √1
T
PT (t).
√
• Fourier Transform (P (f )): The Fourier transform is a sinc function: P (f ) = T · sinc(f T ).
• PSD: The resulting PSD is proportional to sinc (f T ):
2

Gs (f ) = A2 T · sinc2 (f T )

• Spectral Characteristics:
– This is a baseband spectrum, with most of its energy concentrated around DC (f = 0).
– The main lobe has a width of 2/T .
– There are spectral nulls at integer multiples of 1/T .
– This type of signal is suitable for low-pass channels.
Case 2: Signal Space with Non-Zero Mean
This case applies to signal spaces that are not centered at the origin, such as On-Off Keying (OOK) or unipolar
NRZ.
• The symbols a[n] are still statistically independent.

23
 • The mean of the symbols is non-zero (µa ̸= 0).
The presence of a non-zero mean (a DC component in the symbol sequence) introduces discrete spectral lines
(Dirac impulses) into the PSD. The general formula for the PSD in this case is:
σ2 µ2 X
Gs (f ) = a |P (f )|2 + a2 |P (n/T )|2 · δ(f − n/T )
T T n

Example: Unipolar NRZ (a[n] ∈ {0, +a})

• This sequence has µa = a/2 and σa2 = a2 /4.
• Using a rectangular pulse, P (n/T ) = 0 for n ̸= 0.
• PSD: The formula simplifies, resulting in the same continuous sinc2 shape as the bipolar case, but with
an added Dirac impulse at f = 0:
A2 T A2
Gs (f ) = sinc2 (f T ) + δ(f )
4 4
• This impulse at DC represents a portion of the transmitted power that is constant and carries no infor-
mation. It is a direct consequence of the non-zero mean of the signaling scheme.

2.2 Linear Modulation and Bandpass Spectra

If a baseband pulse p(t) is modulated by a carrier cos(2πf0 t) to create a new pulse p′ (t), the resulting spectrum
is shifted to be centered around the carrier frequency ±f0 .
1
Gs′ (f ) = [Gs (f − f0 ) + Gs (f + f0 )]
4
Example: BPSK
q
• The pulse is p(t) = 2
T · PT (t) · cos(2πf0 t).

• The baseband PSD has a sinc2 shape.

• The modulated PSD consists of two sinc2 lobes centered at +f0 and −f0 .
• Spectral Characteristics:
– This is a bandpass spectrum.
– The main lobe width is 2/T , centered at f0 .
– This signal is suitable for bandpass channels.

Chapter 7: Intersymbol Interference (ISI)

1 The Origin of ISI: The Time-Frequency Trade-off
• In digital communications, a sequence of symbols is transmitted using a specific pulse shape, p(t).
• A fundamental principle of signal processing states a trade-off between a signal’s duration in the time
domain and its bandwidth in the frequency domain:
– Time-Limited Signals have Infinite Bandwidth: A signal that is strictly limited in time (e.g.,
a rectangular pulse that exists only for a duration T ) will have a spectrum that extends infinitely in
frequency.
– Band-Limited Signals have Infinite Duration: Conversely, a signal whose spectrum is strictly
limited to a finite bandwidth will have an infinite duration in the time domain.
• Real-world communication channels are always band-limited; they act as filters that only allow a certain
range of frequencies to pass.
• When a signal with an infinite spectrum (like one using rectangular pulses) is sent through such a channel,
the filtering action causes the signal to spread out in the time domain.
• This spreading is the root cause of ISI, as the energy from one symbol ”leaks” into the time slots of its
neighbors.

24
2 The Mechanism of Intersymbol Interference
To understand ISI, we analyze the signal at the receiver’s sampling instant, assuming an ideal (noiseless) channel
for clarity.
P
1. Transmitted Signal: An infinite sequence of symbols a[n] is transmitted as s(t) = n a[n]p(t − nT ).
2. Receiver Operation: The receiver uses a matched filter, q(t), and samples the output y(t) at regular
intervals to recover the symbols. The output of the matched filter is the convolution of the received signal
p(t) and the filter’s impulse response q(t): y(t) = p(t) ∗ q(t).
3. The Combined Response x(t): Since p(t) = s(t) in an ideal channel, the output can be expressed as:
!
X X
y(t) = a[n]p(t − nT ) ∗ q(t) = a[n]x(t − nT )
n n

where x(t) = p(t) ∗ q(t) is the convolution of the transmitted pulse shape with the matched filter’s impulse
response. This function x(t) represents the complete, end-to-end pulse shape as seen at the output of the
matched filter.
4. Sampling: The receiver samples y(t) at time t = t0 + nT to get the value for the n-th symbol, ρ[n].
Substituting this into the equation for y(t) gives:
+∞
X
ρ[n] = a[n − i]x(t0 + iT )
i=−∞

3 The Anatomy of a Received Symbol with ISI

• The equation for ρ[n] reveals how interference occurs.
• By expanding the summation, we can see that the value of the received symbol ρ[n] is composed of three
parts:
∞
X −1
X
ρ[n] = x(t0 )a[n] + x(t0 + iT )a[n − i] + x(t0 + iT )a[n − i]
i=1 i=−∞

1. Desired Component: x(t0 )a[n] is the contribution from the desired transmitted symbol a[n], scaled
by the value of the combined response x(t) at the sampling instant.
P∞
2. Pre-cursor ISI: i=1 x(t0 + iT )a[n − i] is the interference caused by all previous transmitted
symbols (a[n − 1], a[n − 2], etc.).
P−1
3. Post-cursor ISI: i=−∞ x(t0 +iT )a[n−i] is the interference caused by all subsequent transmitted
symbols (a[n + 1], a[n + 2], etc.).
• Intersymbol Interference (ISI) is the phenomenon where the received symbol ρ[n] is corrupted by
contributions from symbols other than the intended symbol a[n].

4 The Condition for Zero ISI (Nyquist Criterion)

• For the receiver to correctly detect the transmitted symbol in an ideal channel (i.e., for ρ[n] to be directly
proportional to a[n]), all the interference terms must be zero.
• This leads to the Nyquist criterion for zero ISI:
• The system is free of ISI if and only if the samples of the combined response x(t), taken at intervals of T ,
satisfy:
– x(t0 + iT ) = 1 for i = 0 (or some non-zero constant)
– x(t0 + iT ) = 0 for all integers i ̸= 0
• This means the overall pulse shape x(t) must have its peak at the sampling instant for the desired symbol
and pass through zero at the sampling instants for all other symbols.

25
5 Analysis of Finite Time Domain Signals
• Consider a signal space where the basis function b1 (t) (and thus the pulse p(t)) has a finite duration,
existing only in the interval [0, T ].
• An example is the rectangular pulse used in basic NRZ signaling.

– The Combined Response x(t): The matched filter q(t) is a time-reversed version of p(t). The
convolution of two identical rectangular pulses of duration T results in a triangular pulse, x(t), with
a duration of 2T .
– Verifying the Zero-ISI Condition: This triangular pulse x(t) has its peak value of T at t = T ,
and is zero for t ≤ 0 and t ≥ 2T . If the receiver samples at t0 = T , then the samples x(T + iT ) are:
∗ x(T ) = T for i = 0.
∗ x(T + iT ) = 0 for all other integers i, since T + iT will fall outside the (0, 2T ) range where the
triangle is non-zero.

• Conclusion:
– For an ideal channel, using signals with a strictly finite time domain of one symbol period T auto-
matically satisfies the condition for zero ISI.
– The spreading caused by the matched filter is predictable and controlled, ensuring no interference at
the sampling points.
– The challenge of ISI arises when trying to use more spectrally efficient, band-limited pulses, which
necessarily have an infinite duration.

Chapter 8: Nyquist Criterion for No ISI

1 The Nyquist Criterion for Zero ISI
• As established previously, for a receiver to correctly interpret a sequence of transmitted symbols without
interference from adjacent symbols, the overall system response x(t) must meet specific conditions at the
sampling instants.
– System Response: x(t) = p(t) ∗ q(t), where p(t) is the transmit pulse shape and q(t) is the impulse
response of the matched filter at the receiver.
• The Nyquist Criterion in the Time Domain: The system is free of ISI if the samples of x(t), taken
at intervals of the symbol period T (and aligned with a timing offset t0 ), satisfy:
– x(t0 + iT ) = 1 for i = 0
– x(t0 + iT ) = 0 for all integers i ̸= 0

• For simplicity, assuming a timing offset t0 = 0, this means the pulse x(t) must have its peak at t = 0 and
pass through zero at all other integer multiples of the symbol period T .
• The Nyquist Criterion in the Frequency Domain:

– Using Fourier transform properties, the time-domain criterion can be shown to be equivalent to a
condition in the frequency domain.
– This is known as Nyquist’s Second Theorem.
• A function x(t) satisfies the no-ISI criterion if and only if the sum of its frequency spectrum X(f ) shifted
by all integer multiples of the symbol rate 1/T is a constant:
+∞
X  n
X f− =T
n=−∞
T

• This means that if you take the spectrum of the pulse, replicate it at every interval of 1/T , and sum all
the overlapping versions, the result must be a flat, constant value.

26
2 Band-Limited Pulses and the Nyquist Criterion
• The primary motivation for using pulses that satisfy the Nyquist criterion is to achieve spectral efficiency
by using a finite frequency domain.
• The challenge is to find a pulse shape x(t) that is band-limited (its spectrum X(f ) is zero outside a certain
range [−fmax , fmax ]) and also satisfies the criterion.
• There are three key cases based on the relationship between the maximum frequency fmax and the symbol
period T :

– Case 1: fmax < 1/(2T )

∗ When the bandwidth is this narrow, the shifted spectral replicas X(f − n/T ) will have gaps
between them.
∗ It is impossible to sum them to form a constant value.
∗ Therefore, no ISI-free solution exists if the bandwidth is less than half the symbol rate.
– Case 2: fmax = 1/(2T ) (The Nyquist Bandwidth)
∗ This is the absolute minimum theoretical bandwidth required for ISI-free transmission at a
symbol rate of 1/T . A solution exists: the ideal low-pass filter.
∗ Frequency Domain: The spectrum X(f ) is a rectangular pulse (a ”brick-wall” filter) with a
bandwidth of 1/T . When shifted by 1/T , the replicas line up perfectly edge-to-edge to sum to a
constant.
∗ Time Domain: The corresponding time-domain pulse is the sinc function:
 
t sin(πt/T )
x(t) = sinc =
T πt/T

∗ The sinc pulse perfectly satisfies the time-domain criterion, with a peak at t = 0 and zero-
crossings at all other integer multiples of T .
∗ However, it is impractical to implement due to its infinite duration and slow decay rate, which
makes it highly sensitive to timing errors.
– Case 3: fmax > 1/(2T ) When the bandwidth is greater than the Nyquist minimum, there is ”excess
bandwidth.” This allows for smoother, more practical filter designs. The overlapping spectral replicas
can be shaped to sum to a constant.

3 The Raised Cosine Filter: A Practical Solution

The most important and widely used family of Nyquist pulses is the Raised Cosine filter.
• Time Domain:
sin(πt/T ) cos(απt/T )
x(t) = ·
πt/T 1 − (2αt/T )2

• Roll-off Factor (α): The parameter α is the roll-off factor, which ranges from 0 to 1. It controls the
”excess bandwidth” of the filter.
– α = 0: This reduces to the ideal low-pass (sinc) filter.
– α = 1: The filter has an excess bandwidth of 100%, meaning it occupies twice the Nyquist bandwidth.

• Properties:
– It always satisfies the Nyquist criterion in the time domain, regardless of the value of α.
– As α increases from 0, the pulse decays more rapidly in the time domain, making it less sensitive to
timing jitter.
• Frequency Domain: The frequency response X(f ) has a flat central portion (the ”passband”) and
smooth ”roll-off” regions shaped like a cosine function.
– Bandwidth: The total occupied bandwidth is (1 + α)/(2T ).

27
4 Transmit (TX) and Receive (RX) Filters
• The overall system response x(t) that must satisfy the Nyquist criterion is the result of the convolution
of the transmit filter p(t) and the receive filter q(t).
• To optimize performance (maximize SNR), q(t) should be a filter matched to p(t).
• This implies a ”square-root” relationship in the frequency domain:
– X(f ) = P (f ) · Q(f ). If q(t) is matched to p(t), then X(f ) = |P (f )|2 .
Therefore, the transmit and receive filters should be designed such that |P (f )|2 = X(f ), or P (f ) =
– p
X(f ).
• Root Raised Cosine (RRC) Filter:
– If the desired overall response is a Raised Cosine filter, then both the transmit filter p(t) and the
receive filter q(t) should be Root Raised Cosine (RRC) filters.
– The RRC filter’s frequency response is the square root of the Raised Cosine response.
• An RRC filter by itself does not satisfy the Nyquist criterion.
• It is only when an RRC-shaped signal is passed through an RRC matched filter that the combined response
becomes a Raised Cosine pulse, thus ensuring zero ISI at the receiver’s sampling instants.
• This is the standard approach used in modern digital communication systems.

Chapter 9: Digital Modulations, Pulse Amplitude Modulation (PAM)

1 Introduction to Digital Modulation
• Modulation is about efficiently converting bits into waveforms. Channel Coding is about adding struc-
tured redundancy to the bit stream before modulation to allow the receiver to detect and correct errors
introduced by the channel.
• Digital modulation is the process of encoding a digital information stream onto a carrier signal suitable
for transmission over a physical channel. This lecture series systematically examines various modulation
techniques by considering the following key aspects:
– The Channel Coding Theorem (Shannon’s Second Theorem): This theorem states that for
any noisy channel with capacity C, and for any information rate R < C, there exist codes that can
achieve arbitrarily low probability of error.
– Key Parameters:
∗ Code Rate (r): The ratio of information bits (k) to total coded bits (n), r = k/n. It measures
the overhead of the code.
∗ Hamming Distance (d): The number of bit positions in which two codewords differ. The
minimum Hamming distance (dmin ) of a code determines its error-handling capability.
· Max errors guaranteed to be detected: tdetect = dmin − 1.
· Max errors guaranteed to be corrected: tcorrect = floor dmin −1


2 .

– Signal Space and Constellation: The geometric representation of the signals used for modulation.
– Binary Labeling: Assigning bit patterns to signal points, often using Gray coding to minimize bit
errors.
– Transmitted Waveform and Spectrum: The mathematical and frequency-domain representation
of the signal.
– Bandwidth and Spectral Efficiency: How efficiently the modulation scheme uses the available
frequency spectrum.
– System Architecture: The structure of the modulator (transmitter) and receiver.
– Error Probability: The theoretical performance of the system in the presence of noise.
– Practical Applications: Where the technique is used in real-world systems.

28
 • Modulation techniques are broadly categorized into:
– Baseband Modulation: The signal’s power is concentrated around DC (zero frequency). PAM is
a primary example.
– Bandpass Modulation: The signal’s power is centered around a carrier frequency f0 ̸= 0. Examples
include PSK, QAM, and FSK.

– Pulse Amplitude Modulation (PAM): Information is encoded in the amplitude of the pulse.
∗ 2-PAM: A 1-dimensional, antipodal baseband scheme. It is the baseband equivalent of BPSK.
∗ m-PAM: Uses m = 2k levels to transmit k bits per symbol. This increases spectral efficiency
(more bits/sec/Hz) at the cost of power efficiency (requires a higher SNR for the same BER).
– Phase Shift Keying (PSK): Information is encoded in the phase of a carrier signal. This is a
bandpass modulation scheme.
∗ 2-PSK (BPSK - Binary PSK): Uses two phases (e.g., 0° and 180°). Highly power-efficient
but spectrally inefficient (1 bit/symbol).
∗ 4-PSK (QPSK - Quadrature PSK): Uses four phases (e.g., 45°, 135°, 225°, 315°). It trans-
mits two bits per symbol and can be viewed as two orthogonal BPSK modulators. It achieves
twice the spectral efficiency of BPSK with the exact same BER performance, making
it extremely popular.
– Advanced PSK Schemes:
∗ Offset QPSK (OQPSK): Delays one of the two bit streams (the quadrature stream) by half
a symbol period. This prevents the signal from ever making an instantaneous 180° phase tran-
sition, which reduces amplitude fluctuations and makes the signal more robust to non-linear
amplification.
∗ π/4-QPSK: Rotates the constellation by 45° (π/4 radians) for every other symbol. This also
prevents the signal’s trajectory from passing through the origin, offering similar benefits to
OQPSK.

• The shape of the transmitted signal is determined by a low-pass filter or pulse p(t), with common choices
being the ideal low-pass filter, the Root Raised Cosine (RRC) filter, and the simple square pulse.

2 2-PAM Signal Space

• 2-PAM is the simplest form of Pulse Amplitude Modulation, encoding one bit per symbol.
• Characteristics:
– It is a baseband modulation technique.
– It uses a 1-dimensional signal space.
– The constellation is antipodal, meaning the two signal points are symmetric opposites (e.g., +α
and −α).
– The information is encoded in the amplitude of the transmitted pulse.
• Signal Space Representation:

– Signal Set: M = {s1 (t) = −ap(t), s2 (t) = +ap(t)}

– Basis Vector: The signal space is spanned by a single basis function, b1 (t) = p(t).
– Vector Space (Constellation): The two signals are represented by scalar values on a 1D line:
M = {s1 = (−α), s2 = (+α)}.
– Binary Labeling: Typically, 0 is mapped to s1 and 1 is mapped to s2 .
• Since one symbol represents one bit (k = 1), the symbol duration T is equal to the bit duration Tb , and
the symbol rate R equals the bit rate Rb .

29
Problem-Solving Technique: Determining Code Capabilities
1. List All Codewords in the code’s dictionary.
2. Calculate All Pairwise Hamming Distances between every distinct pair of codewords.
3. Find dmin : The smallest non-zero distance found is dmin .
4. Determine Capabilities: Use the formulas for tdetect and tcorrect .

Example: A Simple Repetition Code

Problem: A simple code takes 1 data bit and repeats it three times. The codewords are {000, 111}. What are
its capabilities?
Solution:
1. Codewords: C = {000, 111}.
2. Pairwise Hamming Distance: There is only one pair to check. The distance between 000 and 111 is 3
(they differ in all 3 positions).
3. Minimum Distance: dmin = 3.
4. Capabilities:
• tdetect = dmin − 1 = 3 − 1 = 2. The code can detect up to 2 bit errors.
• tcorrect = floor((dmin − 1)/2) = floor((3 − 1)/2) = floor(1) = 1. The code can correct up to 1 bit
error.

3 Bandwidth and Spectral Efficiency of 2-PAM

• The spectrum of a 2-PAM signal is determined by the Power Spectral Density (PSD), which for a zero-
mean antipodal signal is:
σ2
Gs (f ) = a |P (f )|2
T
where σa2 is the variance of the symbol amplitudes and P (f ) is the Fourier transform of the pulse p(t).
• Bandwidth Definitions: There is no single definition of bandwidth. Common metrics include:
– Total Bandwidth: The full frequency range where the spectrum is non-zero.
– Half-Power (-3dB) Bandwidth: The range where the spectrum is above 50% of its peak value.
– Null-to-Null Bandwidth: The width of the main spectral lobe.
– 99% Power Bandwidth: The band that contains 99% of the signal’s total power.
• Spectral Efficiency (η): This metric measures how many bits per second can be transmitted per Hertz
of bandwidth (η = Rb /B).
• Analysis Based on Pulse Shape:
– Case 1: Ideal Low-Pass Filter Pulse p(t):
∗ Bandwidth: The spectrum is a perfect rectangle with a total bandwidth of B = R/2 = Rb /2.
∗ Spectral Efficiency: η = Rb /B = 2 bps/Hz. This is the theoretical maximum for a baseband
signal.
– Case 2: Root Raised Cosine (RRC) Filter Pulse p(t):
∗ Bandwidth: The bandwidth depends on the roll-off factor α: B = (R/2) · (1 + α).
∗ Spectral Efficiency: η = 2/(1 + α) bps/Hz. As α increases (allowing for a more practical
filter), the spectral efficiency decreases.
– Case 3: Square Pulse p(t):
∗ Bandwidth: The spectrum is a sinc2 function, which has an infinite total bandwidth. The
null-to-null bandwidth is B = 1/Tb = Rb .
∗ This is the least spectrally efficient but simplest to generate.

30
4 2-PAM System Architecture
• Modulator (Transmitter):
1. The input binary stream is mapped to a sequence of amplitude levels a[n] (e.g., {−α, +α}).
2. This sequence of impulses is then passed through a pulse-shaping filter p(t) to generate the continuous
transmitted waveform s(t).
• Demodulator (Receiver):
1. The received signal r(t) is passed through a matched filter q(t).
2. A symbol synchronization circuit determines the optimal sampling instants t0 + nT .
3. The output of the matched filter y(t) is sampled at these instants to produce the received symbol
values ρ[n].
4. These values are fed to a decision device that applies the ML criterion (a simple threshold detector
at zero for antipodal PAM) to determine the most likely transmitted symbol sR [n].
5. Finally, the decided symbol is mapped back to a binary value vR [n].

5 The Eye Diagram: A Powerful Analysis Tool

• The Eye Diagram is a practical tool used to assess the quality of a received digital signal and diagnose
issues like Intersymbol Interference (ISI).
– Generation: It is created by overlaying segments of the received signal at the output of the matched
filter. Each segment typically has a duration of two symbol periods (2T ).
– Interpretation: The resulting pattern resembles a human eye.
∗ Eye Opening: The height of the central opening indicates the noise margin. A wide, open
eye means the signal is clean and resistant to noise.
∗ Optimal Sampling Time: The point where the eye is most open is the best time to sample
the signal to maximize the signal-to-noise ratio.
∗ Timing Jitter Sensitivity: The slope of the eye’s sides indicates sensitivity to timing errors.
Steeper slopes mean the system is more robust against jitter.
∗ Zero-Crossing Distortion: The thickness of the signal trace as it crosses the zero line indicates
the amount of timing jitter.
∗ Peak Distortion: The vertical thickness of the top and bottom of the eye indicates the amount
of ISI.

• In an ideal, noiseless system with no ISI, the eye diagram for 2-PAM consists of clean, overlapping triangles.
• In practice, noise and filtering imperfections cause the traces to blur, but the key features of the eye
diagram remain invaluable for system performance evaluation.

6 Error Probability
As established in previous lectures, the Bit Error Rate (BER) for 2-PAM (an antipodal scheme) is given by the
classic formula: r !
1 Eb
BER = · erfc
2 N0

7 Baseband Line Codes as 2-PAM Schemes

Several common baseband line codes can be viewed as specific implementations of 2-PAM, where the primary
difference lies in the pulse shape p(t) used to represent the bits.

31
7.1 Bipolar NRZ (Non-Return-to-Zero)
• Signal Set: Uses a positive rectangular pulse +APT (t) for one binary value and a negative one −APT (t)
for the other. This is identical to a standard 2-PAM using a rectangular pulse.

• Transmitted Waveform: A sequence of high and low voltage levels, where the voltage remains constant
for the entire symbol duration T .
• Signal Spectrum: The Power Spectral Density (PSD) has the characteristic sinc2 (f T ) shape, which is
a baseband spectrum with a main lobe width of 2/T and infinite total bandwidth.

7.2 Bipolar RZ (Return-to-Zero)

• Signal Set: Uses pulses that last for only half the symbol duration, +APT /2 (t) and −APT /2 (t), returning
to zero for the second half of the symbol period.
• Transmitted Waveform: A sequence of positive and negative pulses separated by zero-voltage intervals.

• Signal Spectrum:
– The PSD is proportional to sinc2 (f T /2).
– Because the pulse is shorter in the time domain, its spectrum is wider in the frequency domain.
– The main lobe of the RZ spectrum is twice as wide as that of the NRZ spectrum.

7.3 Other Bipolar Schemes

• Bipolar Triangular:
– Uses triangular pulses instead of rectangular ones.
– The resulting PSD is proportional to sinc4 (f T /2), which has the same null-to-null bandwidth as the
RZ signal but with side lobes that decay much faster, leading to better spectral containment.
• Manchester (Biphase): This scheme uses a pulse shape x(t) = PT /2 (t) − PT /2 (t − T /2), which involves
a mandatory transition in the middle of each symbol period.

– Signal Spectrum:
∗ The PSD has a null at DC (f = 0), which is advantageous for systems that cannot pass DC
components.
∗ The spectrum is proportional to sin4 (πf T /2)/(πf T /2)2 , with its maximum power concentrated
around f ≈ 0.74/T , not at DC.

8 Unipolar Line Codes

Unipolar schemes use one amplitude level (e.g., +A) and a zero level. This results in a signal that is not centered
at the origin (it has a non-zero mean).
• Unipolar NRZ: Uses a full-duration rectangular pulse APT (t) for a ’1’ and 0 for a ’0’.
• Unipolar RZ: Uses a half-duration pulse APT /2 (t) for a ’1’ and 0 for a ’0’.

Spectrum of Unipolar Signals: Because the symbol sequence has a non-zero mean, the PSD consists of two
components:
1. A continuous part similar to the corresponding bipolar scheme (e.g., sinc2 (f T ) for NRZ).
2. Discrete spectral lines (Dirac delta functions) at DC (f = 0) and sometimes at other multiples of
the symbol rate. The presence of the DC component is a key characteristic and potential drawback of
unipolar signaling, as it represents power that does not carry information.

32
9 m-PAM Modulation
To transmit more than one bit per symbol, m-PAM is used, where m is the number of distinct amplitude
levels.
• Characteristics:
– Baseband, 1-dimensional modulation.
– The constellation consists of m points spaced evenly along a line, symmetric with respect to the
origin. For example, 4-PAM uses levels {−3α, −α, +α, +3α}.
• Parameters: m levels can represent k = log2 (m) bits per symbol. This means the symbol duration T is
k times the bit duration Tb , and the symbol rate R is Rb /k.
• Binary Labeling:

– Gray coding is used to assign bit patterns to the amplitude levels.

– With Gray coding, adjacent levels differ by only one bit, which minimizes the bit error rate since the
most likely symbol errors (mistaking a level for an adjacent one) will only cause a single bit error.
• System Architecture:

– The modulator and demodulator are similar to the 2-PAM case, but the mapping/decision blocks
must handle m levels instead of just two.
– The demodulator’s decision device partitions the 1D space into m decision regions.
• Eye Diagram: The eye diagram for m-PAM shows m − 1 open ”eyes” stacked vertically.

10 m-PAM Performance and Trade-offs

• Bandwidth and Spectral Efficiency: The bandwidth of an m-PAM signal depends only on the pulse
shape and the symbol rate R.
– For an ideal low-pass filter pulse, B = R/2. The spectral efficiency is η = Rb /B = 2k bps/Hz.
– For an RRC filter pulse, B = (R/2)(1 + α). The spectral efficiency is η = 2k/(1 + α) bps/Hz.
By increasing m (and thus k), we can increase the spectral efficiency and transmit a higher bit rate within
the same bandwidth.
• Error Probability: The probability of a bit error in m-PAM can be approximated by:
r !
m−1 3k Eb
Pb (e) ≈ · erfc
mk m2 − 1 N0

This formula shows that as m increases, a higher Eb /N0 is required to achieve the same BER. For example,
to get the same BER, 4-PAM requires about 4 dB more power than 2-PAM.
• The Fundamental Trade-off: There is a direct trade-off between spectral efficiency and power efficiency.

– Increasing m: Increases spectral efficiency (higher data rate in the same bandwidth).
– Increasing m: Decreases power efficiency (requires more signal power for the same BER).
This means that for a fixed transmit power, increasing m will significantly reduce the achievable commu-
nication distance.

33
11 Linear Modulation and m-ASK
Linear Modulation is a simple operation that transforms a baseband signal into a bandpass signal by multi-
plying the pulse p(t) with a carrier wave, cos(2πf0 t).
• Effect on Spectrum: This modulation translates the baseband spectrum G(f ) to be centered around
the carrier frequency ±f0 .
• Effect on Performance: Since this is a linear operation and the underlying vector space (constellation)
remains unchanged, the BER performance is identical to the original baseband modulation.

When linear modulation is applied to an m-PAM scheme, the resulting modulation is called m-ASK (Ampli-
tude Shift Keying).
• m-ASK Properties:

– It has the same 1D constellation and the same BER performance as m-PAM.
– Its spectral efficiency is halved compared to m-PAM because the bandpass spectrum occupies twice
the bandwidth of the baseband spectrum (due to the positive and negative frequency components).
– Because of this inefficiency, m-ASK (for m > 2) has few practical applications. The important
exception is 2-ASK, which is identical to 2-PSK (BPSK) and is widely used.

Chapter 10: Digital Modulations - Phase Shift Keying (PSK)

1 2-PSK: Characteristics and Constellation
2-PSK is a digital modulation technique that conveys data by changing or modulating the phase of a carrier
signal.

1.1 Key Characteristics:

• Bandpass Modulation: The signal’s power is concentrated around a carrier frequency f0 , making it
suitable for transmission over radio channels.
• One-Dimensional, Antipodal Constellation: The signal space of 2-PSK is one-dimensional and its
constellation is identical to that of 2-PAM. The two signal points are symmetric opposites.
• Information in Carrier Phase: The binary data (0s and 1s) is encoded by shifting the phase of the
carrier wave.

1.2 Signal Set and Constellation:

The two signals in the set can be represented in two equivalent ways:
1. Amplitude View: s1 (t) = +ap(t) cos(2πf0 t) and s2 (t) = −ap(t) cos(2πf0 t)
• This view treats 2-PSK as a form of 2-PAM (or 2-ASK) where the baseband signal ±ap(t) is linearly
modulated onto a carrier.
2. Phase View: s1 (t) = +ap(t) cos(2πf0 t) and s2 (t) = +ap(t) cos(2πf0 t − π)
• This view highlights the core principle of PSK.
• Since cos(θ − π) = − cos(θ), the two forms are identical.
• A binary ’1’ is represented by one phase (0 degrees), and a binary ’0’ is represented by a 180-degree
phase shift.

• Basis Vector: The signal space is spanned by a single basis function: b1 (t) = p(t) cos(2πf0 t).
• Vector Set: The constellation is identical to 2-PAM: M = {s1 = (+α), s2 = (−α)}.

34
2 2-PSK Waveform, Bandwidth, and Spectral Efficiency
• Transmitted Waveform:

– Since it is a binary scheme (m = 2, k = 1), the symbol rate equals the bit rate (R = Rb ).
– The waveform is a sequence of pulses p(t) modulated by a carrier cos(2πf0 t), where the sign of each
pulse corresponds to the transmitted bit.
• Bandwidth and Spectral Efficiency: The bandwidth of 2-PSK is determined by the bandwidth of the
baseband pulse p(t), shifted to be centered at the carrier frequency f0 .
– Case 1: Ideal Low-Pass Filter Pulse: The total bandwidth is B = R = Rb . The spectral
efficiency is η = Rb /B = 1 bps/Hz.
– Case 2: Root Raised Cosine (RRC) Filter Pulse: The total bandwidth is B = R(1 + α) =
Rb (1 + α). The spectral efficiency is η = 1/(1 + α) bps/Hz.

3 2-PSK Modulator and Demodulator

3.1 Modulator (Transmitter)
A practical 2-PSK modulator can be implemented by first generating a 2-PAM baseband signal and then
up-converting it to the carrier frequency.
1. Mapping: Input bits uT are mapped to antipodal amplitude levels a[n] (e.g., +α and −α).

2. Pulse Shaping:
P The sequence of amplitudes is passed through a transmit filter p(t) to create the baseband
signal v(t) = n a[n]p(t − nT ).
3. Up-conversion: The baseband signal v(t) is multiplied by the carrier wave cos(2πf0 t) to produce the
final transmitted signal s(t).

3.2 Demodulator (Receiver)

• The process is reversed at the receiver.
• This is a coherent demodulator, which means it requires a local carrier signal that is perfectly syn-
chronized in phase and frequency with the incoming carrier.

1. Down-conversion: The received signal r(t) is multiplied by a locally generated carrier cos(2πf0 t).
This produces a signal containing both a baseband component and a high-frequency component
around 2f0 .  
2 1 1
r(t) cos(2πf0 t) = v(t) cos (2πf0 t) = v(t) + cos(2(2f0 )t)
2 2
2. Matched Filtering: The resulting signal is passed through a filter matched to the baseband pulse,
q(t) = p(T − t). This filter is a low-pass filter, which eliminates the high-frequency component at
2f0 .
3. Sampling and Decision: The output of the matched filter, which is now just the recovered baseband
signal, is sampled at the correct instants (determined by symbol synchronization) and fed into a
decision device to recover the bits.
• Synchronization: A 2-PSK demodulator requires both carrier synchronization (to generate the local
cos(2πf0 t)) and symbol synchronization (to determine the optimal sampling instants).

4 2-PSK Performance
• Eye Diagram:
– The eye diagram for 2-PSK is generated from the baseband signal at the output of the matched filter.
– It is identical to the eye diagram for 2-PAM, showing the quality of the signal in terms of noise
margin and timing jitter.

35
 • Error Probability: Since the constellation of 2-PSK is identical to that of 2-PAM (antipodal), their
error performance is also identical. The Bit Error Rate (BER) is given by:
r !
1 Eb
BER = · erfc
2 N0

This confirms that 2-PSK is a very power-efficient modulation scheme, just like its baseband counterpart.

5 Quadrature Modulation
• Quadrature modulation is a technique that enables the transmission of two independent data streams over
the same frequency band, effectively doubling the spectral efficiency.
• It forms the basis for 2D constellations like PSK and QAM.
– Basis Signals: It uses two orthogonal basis signals, a cosine carrier (in-phase component) and a
sine carrier (quadrature component), which are created from the same baseband pulse p(t):
∗ b1 (t) = p(t) cos(2πf0 t)
∗ b2 (t) = p(t) sin(2πf0 t)
– Transmitted Signal: A symbol si = (αi , βi ) is transmitted by modulating the cosine carrier with
the α component and the sine carrier with the β component, and then summing them:
X X
s(t) = α[n]b1 (t − nT ) + β[n]b2 (t − nT )
n n

– I/Q Components:
∗ The part of the signal modulated onto the cosine carrier is called the In-phase (I) component,
α(t).
∗ The part of the signal modulated onto the sine carrier is the Quadrature (Q) component,
β(t).
• Spectrum:
– The spectra of the I and Q components have the same shape and occupy the same frequency band.
– Since they are orthogonal, the power of the total signal s(t) is the sum of the powers of α(t) and
β(t).
– The resulting spectrum Gs (f ) depends only on the shape of the baseband pulse p(t), not on the
specific values of the α and β sequences.

6 Complex Envelope Representation

Quadrature modulation can be elegantly represented using complex numbers.

• Complex Envelope (s̃(t)): The pair of I and Q baseband signals can be combined into a single complex
signal called the complex envelope: s̃(t) = i(t) − jq(t).
• Complex Symbol (γ[n]): Similarly, the pair of symbol amplitudes (α[n], β[n]) can be represented as a
complex symbol: γ[n] = α[n] − jβ[n].
• Analytic Signal (s+ (t)): The actual transmitted bandpass signal s(t) is the real part of the analytic
signal, which is the complex envelope modulated onto a complex carrier: s(t) = Re[s̃(t) · ej2πf0 t ].
This complex representation allows the 2D constellation to be viewed as a set of complex numbers in the complex
plane, which simplifies analysis and system design.

36
3. 4-PSK (QPSK) Modulation
4-PSK, or Quadrature Phase Shift Keying, is a widely used modulation scheme that encodes two bits per symbol
(m = 4, k = 2).
• Characteristics:
– Bandpass modulation with a 2D signal set.
– Uses orthogonal cosine and sine basis signals.
– The constellation consists of four signals placed equidistantly on a circle.
– Information is encoded in the carrier phase, with four possible phases (e.g., 0◦ , 90◦ , 180◦ , 270◦ ).
• Constellation:
– Vector Set: A common representation is M = {(A, 0), (0, A), (−A, 0), (0, −A)}. This can be rotated
by an arbitrary starting phase Φ.
– For example, a 45◦ rotation gives the points {(α, α), (−α, α), (−α, −α), (α, −α)}.
– Binary Labeling: Gray labeling is crucial. It ensures that adjacent points on the circle (the most
likely error destinations) differ by only one bit.
• Bandwidth and Spectral Efficiency:

– Each symbol represents k = 2 bits, so the symbol rate R is half the bit rate (R = Rb /2).
– Ideal Case (Ideal LPF pulse): The bandwidth is B = R = Rb /2. The spectral efficiency is
η = Rb /B = 2 bps/Hz.
– Practical Case (RRC filter): The bandwidth is B = R(1 + α) = (Rb /2)(1 + α). The spectral
efficiency is η = 2/(1 + α) bps/Hz.
• Modulator and Demodulator:
– The 4-PSK modulator/demodulator can be implemented as two parallel 2-PSK systems.
– The incoming bitstream is split into two half-rate streams, one for the I-channel (modulating the
cosine carrier) and one for the Q-channel (modulating the sine carrier).
– The demodulator separates the I and Q components, processes them independently, and then recom-
bines the resulting bits.
• Performance Interpretation:
– Because 4-PSK can be perfectly decomposed into two independent 2-PSK channels, it achieves the
same BER performance as 2-PSK.
– However, since it transmits two bits in the same bandwidth as one bit in 2-PSK, it has double the
spectral efficiency.
This combination of excellent power efficiency (same BER as 2-PSK) and good spectral efficiency (twice that of
2-PSK) makes 4-PSK one of the most widely used digital modulation schemes, common in satellite links, GPS,
and various wireless standards.

7 m-ary PSK (m-PSK) Modulation

The concept can be generalized to m phases, where m is a power of 2.
• Characteristics:
– A 2D bandpass modulation where m signals are placed equidistantly on a circle.
– Each symbol represents k = log2 (m) bits.
• Constellation: The signal points are given by si = (A cos(ϕi ), A sin(ϕi )), where the phases ϕi are spaced
2π/m radians apart.
• Bandwidth and Spectral Efficiency:

37
 – Ideal Case: Spectral efficiency η = k bps/Hz.
– Practical Case (RRC): Spectral efficiency η = k/(1 + α) bps/Hz.
Spectral efficiency increases with m.

• Modulator/Demodulator: For m > 4, the system can no longer be seen as a simple Cartesian product
of two PAM schemes. The modulator must directly generate the α and β components for each symbol,
and the demodulator’s decision regions become wedge-shaped sectors in the 2D plane.
• Error Probability: The BER performance of m-PSK can be approximated (for high SNR) as:
r !
1 Eb 2
π
Pb (e) ≈ · erfc k· · sin
k N0 m

• Performance Degradation: As m increases, the distance between adjacent signal points on the circle
decreases rapidly. This makes the system much more susceptible to noise.
– 8-PSK requires about 3.6 dB more power than 4-PSK for the same BER.
– 16-PSK requires about 4.6 dB more power than 8-PSK.
Because of this severe performance degradation, m-PSK is rarely used for m > 8. Higher-order modulations
typically use QAM, which spaces the points more efficiently.

Chapter 11: Intersymbol Interference (ISI)

1 Origin and Mitigation of ISI
• The Origin of ISI: ISI occurs when the energy from one symbol ”leaks” or spreads into the time slots
of adjacent symbols. This is primarily caused by the filtering effect of the transmission channel, which
causes the transmitted pulses to spread out in time.
• The Nyquist Criterion for Zero ISI: For a system to be free of ISI, the combined end-to-end pulse
shape must have its peak value at the desired sampling instant and pass through zero at all other symbol
sampling instants.

• Pulse Shaping for ISI Prevention: To meet the Nyquist criterion, special pulse shapes are designed.
– Raised Cosine Filter: A practical and widely used family of Nyquist pulses. Its bandwidth is
controlled by a ”roll-off factor” α, where 0 ≤ α ≤ 1.
– Root-Raised Cosine (RRC) Filter: In a modern system, the Raised Cosine filtering responsibility
is split equally between the transmitter (which uses an RRC filter) and the receiver (which uses a
matched RRC filter). The cascade of these two filters results in the desired Raised Cosine response.

1.1 Advanced Technique: Combating ISI with Equalization

While pulse shaping is a preventative measure, real channels always introduce some unpredicted distortion.
Equalization is a form of signal processing at the receiver used to actively reverse this channel distortion.
• Zero-Forcing (ZF) Equalizer: A filter designed to be the mathematical inverse of the channel response
(E(f ) = 1/Hchan (f )).
– Problem: It suffers from severe noise enhancement. At frequencies where the channel is weak (a
”fade”), the equalizer’s gain becomes very high, amplifying the noise at those frequencies along with
the signal.
• Minimum Mean-Square Error (MMSE) Equalizer: A more robust approach that seeks to minimize
the total power of the error (residual ISI + noise). It provides a better balance between reducing ISI and
avoiding noise amplification, generally outperforming the ZF equalizer in practice.

38
Chapter 12: Practical System Design: The Link Budget
A link budget is a comprehensive accounting of all sources of gain and loss in a communication system. Engineers
use it to calculate the signal-to-noise ratio (SNR) at the receiver to ensure the communication link will be feasible
and reliable.

1 The Link Budget Equations

• The Master Equation (in dB): This equation sums up all gains and subtracts all losses.

Prx = Ptx + Gtx − Lf spl − Lmisc + Grx

Where:
– Prx , Ptx : Received and Transmitted Power (typically in dBW or dBm).
– Gtx , Grx : Transmit and Receive Antenna Gains (in dBi).
– Lf spl : Free Space Path Loss (in dB). This is usually the largest loss.
– Lmisc : All other miscellaneous losses (e.g., atmospheric absorption, cable loss, polarization mismatch)
(in dB).

• Free Space Path Loss (FSPL) Formula:

Lf spl (dB) = 20 log10 (d) + 20 log10 (f ) − 147.55

where d is the distance in meters and f is the frequency in Hertz.

2 Problem-Solving Technique: Performing a Link Budget

1. Identify All Gains and Losses from the system specification.

2. Convert All Parameters to dB. For power in Watts, use PdBW = 10 log10 (PWatts ).
3. Calculate Free Space Path Loss (FSPL) using the distance and frequency.
4. Sum Gains and Losses to find Received Power (Prx ) using the master equation.

5. Calculate Noise Power (N ) at the receiver: N = kT B, where k is Boltzmann’s constant (1.38 × 10−23
J/K), T is the system noise temperature in Kelvin, and B is the bandwidth in Hz. In dBW, this is
NdBW = 10 log10 (kT B).
6. Calculate the Signal-to-Noise Ratio (SNR): SN R(dB) = Prx (dBW) − N (dBW).
7. Compare: Check if the calculated SNR meets the requirement for the chosen modulation and coding
scheme to achieve the target BER.

Example: Deep Space Communication Link

Problem: A Mars rover (distance d = 225 × 109 m) transmits to a Deep Space Network antenna on Earth.

• Rover Transmit Power Ptx : 100 W

• Rover Antenna Gain Gtx : 25 dBi
• DSN Antenna Gain Grx : 70 dBi

• Frequency f : 8 GHz (8 × 109 Hz)

• Bandwidth B: 1 MHz (1 × 106 Hz)
• Receiver System Noise Temp T : 20 K
• Misc. Losses Lmisc : 2 dB

39
Is the link feasible if the system requires an SNR of at least 10 dB?
Solution:
1. Ptx = 100 W =⇒ 10 log10 (100) = 20 dBW.

2. Calculate FSPL:

Lf spl = 20 log10 (225 × 109 ) + 20 log10 (8 × 109 ) − 147.55

= 20(11.35) + 20(9.90) − 147.55
= 227.0 + 198.0 − 147.55 = 277.45 dB

3. Calculate Received Power Prx :

Prx = 20 + 25 + 70 − 277.45 − 2 = −164.45 dBW

4. Calculate Noise Power N :

N = (1.38 × 10−23 )(20)(1 × 106 ) = 2.76 × 10−16 W

NdBW = 10 log10 (2.76 × 10−16 ) = −155.6 dBW

5. Calculate SNR:
SN R(dB) = Prx − N = −164.45 − (−155.6) = −8.85 dB

Conclusion: The calculated SNR of -8.85 dB is well below the required 10 dB. Therefore, the link is not
feasible as designed. To make it work, engineers would need to use a combination of much more powerful
error-correcting codes (which can work at lower SNR), reduce the data rate (which reduces the bandwidth B
and thus the noise power N ), or use higher-gain antennas.

40