Digital

Modulation
and
Detection
PROF. ALI HUSSEIN MUQAIBEL

R E F E R E N C E B O O K B Y

A N D R E A G O L D S M I T H 5 . 1 U P T O T H E E N D O F 5 . 4
Geometric
Representation
of Signals
Dr. Ali Muqaibel
Ver. 3.0

Chapter 5 (Simon Haykin, Communication Systems) If you have any comments or corrections on
the slides , you may send to Dr. Ali Muqaibel,
Chapters 5 (Andrea Goldsmith) [5.1 – 5.5]
[email protected]
 Outlines

• Signal-space analysis

• Geometric representation of signals

• Euclidean distance

2 Dr. Ali Hussein Muqaibel

Signal-Space Analysis:

Digital Communication Systems

• General digital communication system

𝑚𝑖 𝑠𝑖 (𝑡) 𝑥(𝑡) 𝑚
ෝ
Message
Transmitter Channel Receiver
Source

𝑚:
ෝ estimate of 𝑚𝑖

• Message source emits a symbol from an alphabet of 𝑀 symbols (𝑚1 , 𝑚2 , … , 𝑚𝑀 ) every 𝑇 seconds.

• A priori probabilities 𝑝1 , 𝑝2 , … , 𝑝𝑀 specify the message source output.

• If emitted symbols of a source are equally likely, then

1
𝑝𝑖 = 𝑃 𝑚𝑖 = 𝑀 for 𝑖 = 1,2, … , 𝑀

3 Dr. Ali Hussein Muqaibel

 Signal-Space Analysis
𝑚𝑖 𝑠𝑖 (𝑡) 𝑥(𝑡) 𝑚
ෝ
Message
Transmitter Channel Receiver
Source
𝑤 𝑡
• Transmitter maps (encodes) the source output 𝑚𝑖 into a distinct signal (waveform) 𝑠𝑖 (𝑡) for
transmission over channel.
• 𝑠𝑖 𝑡 is a real-valued energy signal that occupies 𝑇 seconds allocated for the symbol 𝑚𝑖 ,
with its energy is
𝑇
𝐸𝑖 = න 𝑠𝑖2 𝑡 𝑑𝑡 , 𝑖 = 1,2, … , 𝑀
0
• Channel has the characteristics:
1. Linear with bandwidth wide enough to accommodate the signal 𝑠𝑖 (𝑡) with minor or no
distortion
2. Channel noise, 𝑤(𝑡), is a sample function of zero-mean white Gaussian noise random
process (AWGN channel)
4 Dr. Ali Hussein Muqaibel
Signal-Space Analysis Message
Source
𝑚𝑖
Transmitter
𝑠𝑖 (𝑡)
Channel
𝑥(𝑡)
Receiver
𝑚
ෝ

• Received signal, 𝑥(𝑡), is

0≤𝑡≤𝑇
𝑥 𝑡 = 𝑠𝑖 𝑡 + 𝑤 𝑡 ቊ
𝑖 = 1,2, … , 𝑀
• Receiver observes the received signal 𝑥(𝑡) for 𝑇 seconds and makes the best
estimate of transmitted signal 𝑠𝑖 (𝑡) (equivalently symbol 𝑚𝑖 )
• Due to the presence of noise, the receiver can make errors. It is required to design a
receiver minimizing the average probability of symbol error
𝑀

𝑃𝑒 = ෍ 𝑝𝑖 𝑃 𝑚
ෝ ≠ 𝑚𝑖 |𝑚𝑖
𝑖=1
𝑃 𝑚
ෝ ≠ 𝑚𝑖 |𝑚𝑖 : conditional error probability given that 𝑚𝑖 was sent

5 Dr. Ali Hussein Muqaibel

 Signal-Space Analysis
• Resulting receiver is optimum in the minimum probability of error sense
• Digital communication system model discussed provides the basis to design the optimum
receiver

𝑠𝑖 (𝑡) 𝑥(𝑡)
+
Transmitted Received
signal signal
𝑤(𝑡)
AWGN

• We will use geometric representation for the known set of transmitted signals {𝑠𝑖 (𝑡)}

6 Dr. Ali Hussein Muqaibel

 Geometric Representation of Signals
• Geometric representation: for any set of 𝑀 energy signals, {𝑠𝑖 (𝑡)} is a linear
combination of 𝑁 orthonormal basis functions where 𝑁 ≤ 𝑀
• Given a set of real-valued energy signals 𝑠1 𝑡 , 𝑠2 𝑡 , … , 𝑠𝑀 (𝑡)

Basis function
𝑁
0≤𝑡≤𝑇
𝑠𝑖 𝑡 = ෍ 𝑠𝑖𝑗 𝜙𝑗 (𝑡) , ቊ
𝑖 = 1,2, … , 𝑀
𝑗=1
• where the coefficients of the expansion are
𝑇
𝑖 = 1,2, … , 𝑀
𝑠𝑖𝑗 = න 𝑠𝑖 𝑡 𝜙𝑗 𝑡 𝑑𝑡 , ቊ
0
𝑗 = 1,2, … , 𝑁
• The real-valued basis functions 𝜙1 𝑡 , 𝜙2 𝑡 ,…, 𝜙𝑁 𝑡 are orthonormal

7 Dr. Ali Hussein Muqaibel

 Geometric Representation of Signals
Synthesizer for generating the signal 𝑠𝑖 (𝑡) Analyzer for generating the elements of 𝒔𝒊

Correlator 𝑇
𝑠𝑖1 × (product-integrator) × න 𝑑𝑡 𝑠𝑖1
0

𝜙1 (𝑡) 𝜙1 (𝑡)
𝑇
𝑠𝑖2 × + 𝑠𝑖 (𝑡) 𝑠𝑖 (𝑡) × 𝑠𝑖2
න 𝑑𝑡
0

𝜙2 (𝑡) 𝜙2 (𝑡)
: :
:
𝑇
𝑠𝑖𝑁 × × 𝑠𝑖𝑁
න 𝑑𝑡
𝑁 0
0≤𝑡≤𝑇
𝑠𝑖 𝑡 = ෍ 𝑠𝑖𝑗 𝜙𝑗 (𝑡) , ቊ
𝑖 = 1,2, … , 𝑀
𝑗=1
𝜙𝑁 (𝑡) 𝜙𝑁 (𝑡)
𝑇
𝑠𝑖𝑗 = න 𝑠𝑖 𝑡 𝜙𝑗 𝑡 𝑑𝑡 , ቊ
𝑖 = 1,2, … , 𝑀
0
𝑗 = 1,2, … , 𝑁
8 Dr. Ali Hussein Muqaibel
 Geometric Representation of Signals

• Orthonormal basis functions satisfy

𝑇
1, if 𝑖 = 𝑗
න 𝜙𝑖 𝑡 𝜙𝑗 𝑡 𝑑𝑡 = 𝛿𝑖𝑗 = ቊ
0
0, if 𝑖 ≠ 𝑗
where 𝛿𝑖𝑗 is the Kronecker delta.
• Orthonormal basis functions imply two conditions:
1. Each basis function is normalized to have unit energy
2. Basis functions 𝜙1 𝑡 , 𝜙2 𝑡 ,…, 𝜙𝑁 𝑡 are orthogonal with respect to each other over
the interval 0 ≤ 𝑡 ≤ 𝑇
• The set of coefficients {𝑠𝑖𝑗 }, 𝑗 = 1, … , 𝑁 can be visualized as an 𝑁-dimensional vector
(𝑠𝑖1 , 𝑠𝑖2 ,…, 𝑠𝑖𝑁 ) denoted by 𝒔𝒊 which has one-to-one relation with the transmitted signal 𝑠𝑖 (𝑡)

9 Dr. Ali Hussein Muqaibel

 Geometric Representation of Signals
• Each signal in the set {𝑠𝑖 (𝑡)} is determined by signal vector, 𝒔𝒊 given by

𝑠𝑖1
𝑠𝑖2
Column vector 𝒔𝒊 = : , 𝑖 = 1,2, … , 𝑀
𝑠𝑖𝑁
• Extend the conventional notion of 2D and 3D (Euclidean space) to an 𝑁 -dimensional
Euclidean space.
→ Set of signal vectors {𝒔𝒊 |𝑖 = 1,2, . . , 𝑀} represents set of 𝑀 points in the 𝑁-dimensional
Euclidean space (signal space)
• What is the importance of representing energy signals geometrically?
• For mathematical tractable analysis and providing the basis to noise consideration in digital
communication system.

10 Dr. Ali Hussein Muqaibel

 Example: Geometric Representation of Signals

• Geometrical representation for signals 𝒔𝟏 , 𝒔𝟐 , and 𝒔𝟑

𝑁 =? 𝜙2
𝑀 =?
Illustrating the geometric
representation of signals for the
case when 𝑁 = 2 and 𝑀 = 3.

𝜙1

11 Dr. Ali Hussein Muqaibel

 Geometric Representation of Signals
𝜙2

• 𝒔𝒊 represents the norm (length) of signal vector 𝒔𝒊

• Inner product (dot product) of 𝒔𝒊
𝑁
2 𝜙1
𝒔𝒊 2
= 𝒔𝑇𝒊 𝒔𝒊 = ෍ 𝑠𝑖𝑗
𝑗=1
where 𝑠𝑖𝑗 is the 𝑗𝑡ℎ element of 𝒔𝒊 , . 𝑇
denotes matrix transposition
𝟐
• Show that energy of 𝒔𝒊 𝒕 = 𝒔𝒊
𝑇 𝑇 𝑁 𝑁
𝐸𝑖 = න 𝑠𝑖2 𝑡 𝑑𝑡 = න ෍ 𝑠𝑖𝑗 𝜙𝑗 𝑡 ෍ 𝑠𝑖𝑘 𝜙𝑘 𝑡 𝑑𝑡
0 0 𝑗=1 𝑘=1
𝑁 𝑁 𝑇 𝑁
2 2
=෍ ෍ 𝑠𝑖𝑗 𝑠𝑖𝑘 න 𝜙𝑗 (𝑡)𝜙𝑘 (𝑡)𝑑𝑡 = ෍ 𝑠𝑖𝑗 = 𝒔𝒊
𝑗=1 𝑘=1 0 𝑗=1
• Energy of a signal 𝑠𝑖 (𝑡) equals to the squared length of signal vector 𝒔𝒊

12 Dr. Ali Hussein Muqaibel

Exercise: Geometric Representation of Signals
• Show that
𝑇
න 𝑠𝑖 𝑡 𝑠𝑘 𝑡 𝑑𝑡 = 𝒔𝑇𝒊 𝒔𝒌
0
• which is the inner product of 𝑠𝑖 (𝑡) and 𝑠𝑘 𝑡 over the interval [0, 𝑇]
• Note that the inner product of two signals 𝑠𝑥 (𝑡) and 𝑠𝑦 (𝑡) does not
depend on the choice of basis functions 𝜙𝑗 𝑡 , and it only depends
on the components of the signals 𝑠𝑥 (𝑡) and 𝑠𝑦 (𝑡) projected onto
each of the basis functions.

13 Dr. Ali Hussein Muqaibel

 Euclidean distance
• Euclidean distance (𝑑𝑖𝑘 ) between two points represented by the
𝜙2 𝑡
signal vectors 𝒔𝒊 and 𝒔𝒌 is
𝑁 𝑇
2 2 2 2
𝑑𝑖𝑘 = 𝒔𝒊 − 𝒔𝒌 = ෍ 𝑠𝑖𝑗 − 𝑠𝑘𝑗 = න 𝑠𝑖 𝑡 − 𝑠𝑘 𝑡 𝑑𝑡 𝜃12

𝑗=1 0
𝜙1 𝑡
• The angle, 𝜃𝑖𝑘 , between two signal vectors 𝒔𝒊 and 𝒔𝒌 is given by

𝒔𝑇𝒊 𝒔𝒌
cos𝜃𝑖𝑘 =
𝒔𝒊 𝒔𝒌
• Signal vectors 𝒔𝒊 and 𝒔𝒌 are orthogonal (perpendicular) if 𝒔𝑇𝒊 𝒔𝒌
= 0 ⇒ 𝜃𝑖𝑘 = 90°

14 Dr. Ali Hussein Muqaibel

Exercise 2: Geometric Representation of Signals
• For the geometric representation
shown and assuming orthonormal
𝜙2 𝑡
basis.
▪ Write 𝑆1 and 𝑆2 in terms of the basis functions. 5
4
▪ Find the norm of 𝑆1
𝜃12
▪ Find the energy in the in 𝑆2
▪ Find the angle between the two vectors, 𝑆1 and 𝑆2 . 𝜙1 𝑡
5 10

15 Dr. Ali Hussein Muqaibel

 Block diagram of a generic digital communication system

Gram-Schmidt Orthogonalization
Procedure
Dr. Ali Hussein Muqaibel
𝑠1 (𝑡) 𝑠2 (𝑡)

1 1
𝑡 𝑡
2 -1 2

𝑠3 (𝑡) 𝑠4 (𝑡)

1
𝑡 𝑡
-1 2 -1 2
Visit my Website and YouTube Channel for
more resources

1 Dr. Ali Hussein Muqaibel

 Outlines
• Gram-Schmidt procedure
Correlator 𝑇
(product-integrator) × න 𝑑𝑡 𝑠𝑖1
• Gram-Schmidt Example
0

• Signal Constellation Diagram

𝜙1 (𝑡)
𝑠1 (𝑡) 𝑠2 (𝑡) 𝑇
𝑠𝑖 (𝑡) × න 𝑑𝑡 𝑠𝑖2
0
1 1
𝑡 𝑡
2 2 𝜙2 (𝑡)
-1 :
:
𝑠3 (𝑡) 𝑠4 (𝑡) 𝑇
× න 𝑑𝑡 𝑠𝑖𝑁
0
1
𝑡 𝑡
2 2 𝜙𝑁 (𝑡)
-1 -1

2 Dr. Ali Hussein Muqaibel

 Gram-Schmidt Orthogonalization Procedure
• Gram-Schmidt procedure is a technique to find a complete set of orthonormal functions
• Suppose 𝑠1 𝑡 , 𝑠2 (𝑡), . . , 𝑠𝑀 𝑡 are energy signals.
• First basis function:
▪ Select 𝑠1 𝑡 (arbitrary) to find basis function
𝑠1 (𝑡)
𝜙1 𝑡 =
𝐸1
𝐸𝑖 : energy of the signal 𝑠𝑖 (𝑡)
▪ Then, 𝑠1 𝑡 = 𝐸1 𝜙1 𝑡 = 𝑠11 𝜙1 (𝑡) (why?)

3 Dr. Ali Hussein Muqaibel

 Gram-Schmidt Procedure
• Similarly using 𝑠2 (𝑡), define the coefficient 𝑠21
𝑇
𝑠21 = න 𝑠2 𝑡 𝜙1 (𝑡) 𝑑𝑡
0
• Let 𝑔2 𝑡 = 𝑠2 𝑡 − 𝑠21 𝜙1 (𝑡)
• Here, 𝑔2 𝑡 and 𝜙1 𝑡 are orthogonal over 0 ≤ 𝑡 ≤ 𝑇.
• Second basis function:
𝑔2 (𝑡) 𝑠2 𝑡 − 𝑠21 𝜙1 (𝑡)
𝜙2 𝑡 = =
2
𝑇 𝐸2 − 𝑠21
‫׬‬0 𝑔22 𝑡 𝑑𝑡
𝑇
Also, ‫׬‬0 𝜙1 𝑡 𝜙2 𝑡 𝑑𝑡 = 0 [Exercise: show that they are orthogonal]

4 Dr. Ali Hussein Muqaibel

 Gram-Schmidt Procedure
• In general,
𝑖−1

𝑔𝑖 𝑡 = 𝑠𝑖 𝑡 − ෍ 𝑠𝑖𝑗 𝜙𝑗 (𝑡)
𝑗=1
• where the coefficients 𝑠𝑖𝑗 are
𝑇
𝑠𝑖𝑗 = න 𝑠𝑖 𝑡 𝜙𝑗 𝑡 𝑑𝑡 , 𝑗 = 1,2, . . , 𝑖 − 1
0
• Given 𝑔𝑖 (𝑡), then
𝑔𝑖 (𝑡)
𝜙𝑖 𝑡 = , 𝑖 = 1,2, . . , 𝑁
𝑇
‫׬‬0 𝑔𝑖2 𝑡 𝑑𝑡
• which forms an orthonormal set of basis functions.

5 Dr. Ali Hussein Muqaibel

 Gram-Schmidt Procedure
• The dimension 𝑁 is less than or equal to the number of signals, 𝑀
→ If the signals 𝑠1 𝑡 , 𝑠2 𝑡 , . . , 𝑠𝑀 (𝑡) form a linearly independent set, then 𝑁 = 𝑀.
→ If the signals 𝑠1 𝑡 , 𝑠2 𝑡 , . . , 𝑠𝑀 (𝑡) are not linearly independent, then 𝑁 < 𝑀.
• The intermediate function 𝑔𝑖 (𝑡) is zero for 𝑖 > 𝑁 (why?)

• Example:
Symbol Amplitude Gray code
𝑠1 (𝑡) -3 00
𝑠2 (𝑡) -1 01
𝒔𝟏 𝒔𝟐 𝒔𝟑 𝒔𝟒 𝑠3 (𝑡) +1 11
𝜙1 (𝑡)
𝑠4 (𝑡) +3 10
-3 -1 0 +1 +3

6 Dr. Ali Hussein Muqaibel

Example: Gram-Schmidt Procedure
• Use Gram-Schmidt procedure to find the orthonormal
basis functions for the set of signals
𝑠1 (𝑡) 𝑠2 (𝑡)

1 1
𝑡 𝑡
2 -1 2

𝑠3 (𝑡) 𝑠4 (𝑡)

1
𝑡 𝑡
-1 2 2
-1

7 Dr. Ali Hussein Muqaibel

Example Solution:
Gram-Schmidt Procedure
• The energy for 𝑠1 (𝑡) is 𝜙1 (𝑡)
2
𝐸1 = න 𝑠12 𝑡 𝑑𝑡 = 2
0
• Hence, the first orthonormal basis function is 1/ 2
𝑠1 (𝑡) 𝑠1 (𝑡)
𝜙1 𝑡 = = 𝑡
𝐸1 2
• Projection of 𝑠2 (𝑡) over the basis function 𝜙1 (𝑡) is 2
2
𝑠21 = න 𝑠2 𝑡 𝜙1 𝑡 𝑑𝑡 = 0
0
• The intermediate function 𝜙2 (𝑡)
𝑔2 𝑡 = 𝑠2 𝑡 − 𝑠21 𝜙1 𝑡 = 𝑠2 (𝑡)
2
𝐸𝑔2 = න 𝑠22 𝑡 𝑑𝑡 = 2 1/ 2
0 𝑡
• → Second orthonormal basis function 2
𝑔2 𝑡 𝑠2 𝑡 −1/ 2
𝜙2 𝑡 = =
𝐸𝑔2 2

8 Dr. Ali Hussein Muqaibel

 Continue Example Solution
2 𝑠1 (𝑡) 𝑠2 (𝑡)
𝑠31 = න 𝑠3 𝑡 𝜙1 𝑡 𝑑𝑡 = 0
0 1 1
2
𝑡 𝑡
𝑠32 = න 𝑠3 𝑡 𝜙2 𝑡 𝑑𝑡 = − 2 2 2
0 -1
• The intermediate function 𝑠3 (𝑡) 𝑠4 (𝑡)
𝑔3 𝑡 = 𝑠3 𝑡 − 𝑠32 𝜙2 𝑡 = 0
1
• Hence, 𝑠3 𝑡 = 𝑠32 𝜙2 𝑡 [No new basis function] 𝑡 𝑡
2 2 2
-1 -1
𝑠41 = න 𝑠4 𝑡 𝜙1 𝑡 𝑑𝑡 = − 2
0
2 𝜙1 (𝑡)
𝑠42 = න 𝑠4 𝑡 𝜙2 𝑡 𝑑𝑡 = 0
0
1/ 2
• The intermediate function
𝑡
𝑔4 𝑡 = 𝑠4 𝑡 − 𝑠41 𝜙1 𝑡 = 0
2

𝜙2 (𝑡)
𝑠4 𝑡 = 𝑠41 𝜙1 𝑡 [No new basis function]
1/ 2
𝑡
2
−1/ 2

9 Dr. Ali Hussein Muqaibel

 Continue Example Solution
Gram-Schmidt Procedure
• There are only two orthonormal basis function to represent the
four signals 𝑠1 𝑡 , 𝑠2 𝑡 , 𝑠3 𝑡 , and 𝑠4 (𝑡) [𝑁 = 2 < 𝑀]
Signal Constellation Diagram (vector representation)
𝜙2 (𝑡)

− 2 2
𝜙1 (𝑡)

− 2
10 Dr. Ali Hussein Muqaibel
Practice: Gram Schmidt orthogonalization procedure

The figure below displays the waveforms of three signals.

a) Using Gram Schmidt orthogonalization procedure, find an orthonormal
basis for this set of signals.
b) Represent the three signals in terms of the basis and show them as
vectors in the signal-space diagram.

11 Dr. Ali Hussein Muqaibel

Correlation
and Matched
Filter Receivers

Ali Hussein Muqaibel

Visit my website and YouTube channel for more resources:
Muqaibel
 Outlines
• Correlation receiver and equivalence to matched filter
• Rotation of signal constellation
• Translation of signal constellation
• Minimum energy signals
• Union bound
• Pairwise error probability
• Bit and symbol error probabilities

2 Dr. Ali Hussein Muqaibel

 Optimal Receiver (Correlation Receiver)
For equally likely transmitted signals, 𝑠1 𝑡 , . . , 𝑠𝑀 (𝑡), and in AWGN channel, the optimal receiver has
two components:

1. Detector 2. Decoder
This part consists of a bank of 𝑁 product- Performs maximum likelihood decision rule on
integrators (or correlators), that has input 𝑥(𝑡) and observation vector 𝒙 to obtain the estimate 𝑚 ෝ
outputs observation vector 𝒙 of transmitted symbol 𝑚𝑖 , 𝑖 = 1, . . , 𝑀 such that
the average probability of error is minimized.
𝑇
× න 𝑑𝑡 𝑥1
0

Observation vector 𝒙
𝜙1 (𝑡)
𝑇
𝑥(𝑡) × න 𝑑𝑡 𝑥2
0

𝜙2 (𝑡)
:
:
𝑇
× න 𝑑𝑡 𝑥𝑁
0

𝜙𝑁 (𝑡)
Dr. Ali Hussein Muqaibel
 Equivalence of correlator and matched filter
receivers
• Detector part in correlation receiver has a set of 𝑁 correlators that
can be replaced by matched filters
• Let ℎ𝑗 (𝑡) be impulse response of a linear time-invariant (LTI) filter

𝑥(𝑡) ℎ𝑗 𝑡 = 𝜙𝑗 (𝑇 − 𝑡) 𝑦𝑗 (𝑡)

• ℎ𝑗 (𝑡) is matched to an input signal 𝜙𝑗 (𝑡)

→It is time reversed and delayed version by 𝑇
∞ ∞
𝑦𝑗 𝑡 = න 𝑥 𝜏 ℎ𝑗 𝑡 − 𝜏 𝑑𝜏 = න 𝑥 𝜏 𝜙𝑗 𝑇 − 𝑡 + 𝜏 𝑑𝜏
−∞ −∞

4 Dr. Ali Hussein Muqaibel

 Correlation & Matched Filter Receivers
∞ ∞
𝑦𝑗 𝑡 = න 𝑥 𝜏 ℎ𝑗 𝑡 − 𝜏 𝑑𝜏 = න 𝑥 𝜏 𝜙𝑗 𝑇 − 𝑡 + 𝜏 𝑑𝜏
−∞ −∞
• If we sample the output, 𝑦𝑗 (𝑡), at time 𝑡 = 𝑇, then
∞
𝑦𝑗 (𝑇) = න 𝑥 𝜏 𝜙𝑗 𝜏 𝑑𝜏
−∞
• For 𝜙𝑗 𝑡 zero outside 0 ≤ 𝑡 ≤ 𝑇, then 𝑦𝑗 (𝑇) is equivalent to the 𝑗𝑡ℎ correlator
output 𝑥𝑗
𝑇
𝑦𝑗 (𝑇) = න 𝑥 𝜏 𝜙𝑗 𝜏 𝑑𝜏
0
→ Detection part of correlator receiver can be implemented by a bank of 𝑁 matched
filters followed by samplers at 𝑡 = 𝑇

5 Dr. Ali Hussein Muqaibel

 Correlation & Matched Filter Receivers
• Detector part of matched filter receiver
𝑇
𝜙1 (𝑇 − 𝑡)
× න 𝑑𝑡 𝑥1
0

𝜙1 (𝑡)
𝑇
𝑥(𝑡) × න 𝑑𝑡 𝑥2 𝜙2 (𝑇 − 𝑡)
0

𝜙2 (𝑡)
:
:
𝑇
× න 𝑑𝑡 𝑥𝑁
0

𝜙𝑁 (𝑇 − 𝑡)

𝜙𝑁 (𝑡)

6 Dr. Ali Hussein Muqaibel

 Invariance of Probability of Error to Rotation
and Translation
• Partitions of signal space, 𝑍, under AWGN and maximum likelihood decision rule depends on the
signal constellation.
• Rotation of signal constellation does not affect 𝑃𝑒 (why?)
• AWGN is spherically symmetric in all directions in signal space
• If a signal constellation is rotated by orthonormal transformation
𝒔෤ 𝒊 = 𝑸𝒔𝒊 , 𝑖 = 1,2, . . , 𝑀
• where 𝑸 is orthonormal matrix, then probability of symbol error 𝑃𝑒 is the same (unchanged)
considering maximum likelihood detection over AWGN channel
• Example:

orthogonal matrix is a real square matrix whose columns/and rows are orthogonal unit vectors (orthonormal vectors)

7 Dr. Ali Hussein Muqaibel

 Principle of Translational Invariance
• Consider a translation of all signal points in constellation by 𝒂
𝒔ത 𝒊 = 𝒔𝒊 − 𝒂 𝑖 = 1, . . , 𝑀
• Translated observation vector is
ഥ = 𝒙−𝒂
𝒙
• In a similar manner, the Euclidean distance considering translated observation
vector is
ഥ − 𝒔ത 𝒊 = 𝒙 − 𝒔𝒊 for all 𝑖
𝒙
• If a signal constellation is translated by a constant vector amount, the probability of
symbol error 𝑃𝑒 is unchanged considering maximum likelihood detection over AWGN
channel
• Example: 4-level PAM signals
0 0 How is translation helpful?
𝜙1 𝜙1

−3𝛼/2 −𝛼/2 𝛼/2 3𝛼/2 𝛼 2𝛼 3𝛼

8 Dr. Ali Hussein Muqaibel
 Minimum Energy Signals
• Consider a set of symbols 𝑚1 , . . , 𝑚𝑀 representing the signals 𝒔𝟏 , . . , 𝒔𝑴 . The average energy
of signal constellation is
𝑀

𝐸𝑠 = ෍ 𝒔𝒊 2𝑝
𝑖
𝑖=1
where 𝑝𝑖 = 𝑃(𝑚𝑖 ).
• If the signals 𝒔𝟏 , . . , 𝒔𝑴 are translated by a vector 𝒂, then the average energy of translated
signal constellation is
𝑀

𝐸ത𝑠 = ෍ 𝒔𝒊 − 𝒂 2 𝑝𝑖
𝑖=1
• Minimum Energy Translate
• Given a signal constellation 𝒔𝒊 𝑀
𝑖=1 , the corresponding signal constellation with minimum
average energy is obtained by subtracting from each signal vector, 𝒔𝒊 , a constant 𝐸 𝒔

9 Dr. Ali Hussein Muqaibel

 Minimum Energy Signals
• Exercise:
• For equally probable symbols, find 𝐸𝑠,𝑚𝑖𝑛 for both constellations

(B1)
0
(A1) 𝜙1

−3𝛼/2 −𝛼/2 𝛼/2 3𝛼/2

(A2) 0
𝜙1 (B2)

𝛼 2𝛼 3𝛼

10 Dr. Ali Hussein Muqaibel

 Practice Problem
Consider the two 8-point QAM signal constellations shown in the Figure. The minimum distance between adjacent
points is 2A .
a) Determine the average transmitted power per symbol (then per bit) for each constellation, assuming that the
signal points equally probable.
b) Which constellation is more power-efficient over ASGN channel?

(A) (B) 𝑨=𝟏

11 Dr. Ali Hussein Muqaibel

 Union Bound on Probability of Error
𝜙2 (𝑡)
• Numerical computation of integral in average probability of symbol
error, 𝑃𝑒 , is sometimes difficult (or impractical).
→ Use bounds to predict the signal-to-noise ratio required to maintain
𝜙1 (𝑡)
certain error rate (𝑃𝑒 )
• Let 𝐴𝑖,𝑘 is the event that 𝒙 is closer to 𝒔𝒌 than 𝒔𝒊 where 𝑚𝑖 is sent,
𝑖, 𝑘 = 1, . . , 𝑀. Conditional error probability given 𝑚𝑖 is

𝑃𝑒 𝑚𝑖 = 𝑃(𝐴𝑖,1 ∪ ⋯ ∪ 𝐴𝑖,𝑖−1 ∪ 𝐴𝑖,𝑖+1 ∪ ⋯ ∪ 𝐴𝑖,𝑀 ) ≤ ෍ 𝑃(𝐴𝑖,𝑘 )

𝑘=1
𝑘≠𝑖
• which is called the union bound

1 Dr. Ali Hussein Muqaibel

 Union Bound on Probability of Error
• Geometrical representation

Note that Shaded region

𝟒
represents 𝑃𝑒 (𝑚𝑖 )
෍ 𝑷(𝑨𝟏,𝒌 ) ≥ 𝑷𝒆 (𝒎𝟏 )
𝒌=𝟐

Shaded region
represents 𝑃(𝐴1,2 )
𝑃(𝐴1,4 )
𝑃(𝐴1,3 )

2
Dr. Ali Hussein Muqaibel
 Union Bound on Probability of Error
• There is a difference between 𝑃 𝐴𝑖,𝑘 and 𝑃(𝑚
ෝ = 𝑚𝑘 |𝑚𝑖 )
▪ 𝑃(𝑚 ෝ = 𝑚𝑘 |𝑚𝑖 ) indicates that 𝒙 lies closer to 𝒔𝒌 compared to any other vector if 𝒔𝒊 was
sent.
▪ 𝑃(𝐴𝑖,𝑘 ) depends only on two signal vectors 𝒔𝒊 and 𝒔𝒌 (pairwise error probability,
𝑃2 (𝒔𝒊 , 𝒔𝒌 ))
• Pairwise error probability: It is the probability that a receiver mistakes 𝒔𝒌
for 𝒔𝒊 assuming the system uses only pair of signals (𝒔𝒊 and 𝒔𝒌 )
𝑀

𝑃𝑒 (𝑚𝑖 ) ≤ ෍ 𝑃2 (𝒔𝒊 , 𝒔𝒌 )
𝑘=1
𝑘≠𝑖

3 Dr. Ali Hussein Muqaibel

Pairwise Error Probability
• The probability 𝑃2 (𝒔𝒊 , 𝒔𝒌 ) when 𝑚𝑖 is sent,

𝑃2 (𝒔𝒊 , 𝒔𝒌 ) = 𝑃 𝒙 is closer to 𝒔𝒌 than 𝒔𝒊 , when 𝑚𝑖 is sent

∞
1 𝑣2
=න exp − 𝑑𝑣
𝑑𝑖𝑘 /2 𝜋𝑁0 𝑁 0 𝜙2 (𝑡)
where 𝑑𝑖𝑘 = 𝒔𝒊 − 𝒔𝒌

• Recall that
2 ∞
erfc 𝑢 = න exp −𝑧 2 𝑑𝑧
𝜋 𝑣
𝜙1 (𝑡)
• Let 𝑧 = 𝑣/ 𝑁0 , then
1 𝑑𝑖𝑘
𝑃2 (𝒔𝒊 , 𝒔𝒌 ) = erfc
2 2 𝑁0

4 Dr. Ali Hussein Muqaibel

 Union Bound and Probability of Error
𝜙2 (𝑡)
• The conditional error probability given 𝑚𝑖
𝑀
1 𝑑𝑖𝑘
𝑃𝑒 (𝑚𝑖 ) ≤ ෍ erfc
2 2 𝑁0
𝑘=1
𝑘≠𝑖
𝜙1 (𝑡)
• The average symbol error over all 𝑀 symbols
𝑀

𝑃𝑒 = ෍ 𝑝𝑖 𝑃𝑒 𝑚𝑖
𝑖=1
𝑀 𝑀
1 𝑑𝑖𝑘
≤ ෍ ෍ 𝑝𝑖 erfc
2 2 𝑁0
𝑖=1 𝑘=1
𝑘≠𝑖
• There are special forms that simplifies and bound the
expression of 𝑃𝑒

5 Dr. Ali Hussein Muqaibel

Pairwise Error Probability 1
𝑀 𝑀

𝑃𝑒 ≤ ෍ ෍ 𝑝𝑖 erfc
𝑑𝑖𝑘
2 2 𝑁0
𝑖=1 𝑘=1
𝑘≠𝑖
• Case 1: Circularly Symmetric
• If the signal constellation is circularly symmetric about the origin
→ 𝑃𝑒 𝑚𝑖 is the same for all 𝑖
𝑀 𝜙2 (𝑡)
1 𝑑𝑖𝑘
𝑃𝑒 ≤ ෍ erfc
2 2 𝑁0
𝑘=1
𝑘≠𝑖 𝜙1 (𝑡)

• Example:
• QPSK signal constellation

6 Dr. Ali Hussein Muqaibel

Minimum Distance and Pairwise Error
𝑀 𝑀

Probability 1
𝑃𝑒 ≤ ෍ ෍ 𝑝𝑖 erfc
2
𝑖=1 𝑘=1
𝑑𝑖𝑘
2 𝑁0
𝑘≠𝑖
• Case 2: Minimum Distance Bound
• Let the minimum distance of signal constellation, 𝑑𝑚𝑖𝑛 , be the smallest
Euclidean distance between any two signal points
𝑑𝑚𝑖𝑛 = min 𝑑𝑖𝑘
𝑖≠𝑘
• erfc(𝑢) is monotonically decreasing function of 𝑢, then

𝑑𝑖𝑘 𝑑𝑚𝑖𝑛
erfc ≤ erfc ∀𝑖, 𝑘 ∈ {1, . . , 𝑀}
2 𝑁0 2 𝑁0
• Hence,
(𝑀−1) 𝑑𝑚𝑖𝑛
𝑃𝑒 ≤ erfc
2 2 𝑁0
• Example:
• 16-QAM or signal constellations with large number of points

7 Dr. Ali Hussein Muqaibel

 Exponential Bound for the Probability of Error
• The complementary function can be further upper bounded by
2
𝑑𝑚𝑖𝑛 1 𝑑𝑚𝑖𝑛
erfc ≤ exp −
2 𝑁0 𝜋 4𝑁0
• So,
2
(𝑀 − 1) 𝑑𝑚𝑖𝑛
𝑃𝑒 ≤ exp −
2 𝜋 4𝑁0
→ Average probability of symbol error decreases exponentially as squared minimum
2
distance, 𝑑𝑚𝑖𝑛
• Average probability of symbol error is important for noise performance of digital
system. Another performance indicator is bit error rate (BER) (or average
probability of bit error)

8 Dr. Ali Hussein Muqaibel

 Bit and Symbol Error Probabilities
• There are no unique relationships between symbol and bit error probabilities (one cannot be
derived accurately from the other)
• Case 1 (Gray Code)
𝑃𝑒
≤ 𝐵𝐸𝑅 ≤ 𝑃𝑒
log 2 𝑀
• Case 2
• Let 𝑀 = 2𝐾 , assuming all symbol errors are equally probable,
𝑀 𝑃𝑒 2𝐾−1
𝐵𝐸𝑅 = = 𝐾 𝑃
2(𝑀 − 1) 2 −1 𝑒
• For large 𝑀, BER→ 𝑃𝑒 /2
• Bit error are not independent,
𝑃 𝑖 𝑡ℎ and 𝑗𝑡ℎ bits in error ≠ 𝐵𝐸𝑅 2

9 Dr. Ali Hussein Muqaibel

 FSK, MSK, Non-coherent
Orthogonal Modulation,
DPSK
Dr. Ali Muqaibel
Ver. 2.4

Visit my website and Video Chanel to get

more resources: Muqaibel
Material here can be found in most of the popular Communications textbooks, check for example Communication Systems 1
by Simon Haykin (Ch6) or Wireless Communications by Andrea Goldsmith (Ch 5)
 Outlines
• Linear and Non-Linear Modulation Techniques
• Frequency shift keying (FSK)
▪ Error probability of binary FSK
▪ MFSK and Spectral Efficiency
• Continuous-Phase-FSK (CPFSK)
▪ MSK (Minimum Shift Keying)
• Noncoherent Orthogonal Modulation
▪ Noncoherent Binary FSK
▪ DPSK
• Comparison between Digital Modulation Techniques.

2 Dr. Ali Hussein Muqaibel

 Linear and Non-Linear Modulation Techniques
Non-Linear Linear
• M-ASK, M-PSK and M-QAM are linear modulations Modulations Modulations

• Coherent FSK is a non-linear modulation

▪ Linear modulations
▪ ASK
▪ PSK

▪ Non-linear modulation
▪ FSK
▪ MSK
CPM : Continuous Phase Modulation
CPFSK : Continuous Phase FSK
CP-BFK : Continuous Phase-Binary FSK
MSK : Minimum Shift Keying
PSK : Phase Shift Keying
QPSK : Quadrature Phase Shift Keying
O-QPSK : Offset Quadrature Phase Shift Keying

3 Dr. Ali Hussein Muqaibel

 Binary Frequency Shift Keying (BFSK)
▪ In BFSK, the transmitted signals for binary symbols 0 or 1 are sinusoidal wave with a fixed difference in
frequency
2𝐸𝑏
cos 2𝜋𝑓𝑖 𝑡 , 0 ≤ 𝑡 ≤ 𝑇𝑏
𝑠𝑖 𝑡 = ൞ 𝑇𝑏
Phase continuity is always
0, elsewhere maintained (continuous-phase
frequency shift keying CPFSK)
and 𝑖 = 1,2. 𝐸𝑏 is transmitted signal energy per bit.
𝑓𝑖 = (𝑛𝑐 + 𝑖)/𝑇𝑏
where 𝑛𝑐 is a fixed integer.
▪ The symbol 1 is represented by 𝑠1 𝑡 , and binary symbol 0 by 𝑠2 (𝑡).
▪ 𝑠1 (𝑡) and 𝑠2 (𝑡) are orthogonal but not normalized and orthonormal basis functions are

2
𝜙𝑖 𝑡 = cos 2𝜋𝑓𝑖 𝑡 , 0 ≤ 𝑡 ≤ 𝑇𝑏
𝑇𝑏
0, elsewhere
where 𝑖 = 1,2.
▪ BFSK is a continuous phase signal

4 Dr. Ali Hussein Muqaibel

 Coherent Binary Frequency Shift Keying
• The coefficient 𝑠𝑖𝑗 for 𝑖 = 1, 2 and 𝑗 = 1, 2 is

𝑇𝑏
𝑠𝑖𝑗 = න 𝑠𝑖 𝑡 𝜙𝑗 𝑡 𝑑𝑡
0 𝑇
𝒔𝟐 = 0 𝐸𝑏
𝑇𝑏
2𝐸𝑏 2
=න cos 2𝜋𝑓𝑖 𝑡 cos 2𝜋𝑓𝑗 𝑡 𝑑𝑡
0 𝑇𝑏 𝑇𝑏
𝑇
𝐸 , 𝑖=𝑗 𝒔𝟏 = 𝐸𝑏 0
=൝ 𝑏
0, 𝑖≠𝑗

Euclidean distance between 𝒔𝟏 and 𝒔𝟐 is 2𝐸𝑏

5 Dr. Ali Hussein Muqaibel

 Error Probability of BFSK

1 𝐸𝑏
𝑃𝑒 = erfc
2 2𝑁0

• Compare BFSK and BPSK in terms of BER.

• Recall that the average error probability for BPSK is

1 𝐸𝑏
𝑃𝑒 = erfc
2 𝑁0

• Clearly, BPSK outperforms coherent BFSK.

• The bit energy-to-noise ratio, 𝐸𝑏 /𝑁0, in BFSK needs to be doubled to have same BER of BPSK
• In signal-space diagram, the Euclidean distance in BPSK is 𝑑𝑚𝑖𝑛 = 2 𝐸𝑏 . However, in BFSK 𝑑𝑚𝑖𝑛
= 2𝐸𝑏 → BPSK outperforms BFSK.

6 Dr. Ali Hussein Muqaibel

 Continuous-Phase-FSK (CPFSK) and
Minimum Shift Keying (MSK)
• In coherent BFSK, the phase of signals is not fully exploited (only synchronization)
• Minimum shift keying (MSK) can improve the performance exploiting the phase of signals
• The minimum-shift keying (MSK) scheme is used in GSM, a pioneer and a widely-used digital cellular
mobile system. MSK can be viewed as either a special case of binary continuous-phase frequency-shift
keying (CPFSK) or a special case of OQPSK.
• Consider a Continuous-Phase-FSK (CPFSK) signal defined as

2𝐸𝑏
𝑠𝑖 𝑡 = cos 2𝜋𝑓𝑖 𝑡 + 𝜃(0) , 0 ≤ 𝑡 ≤ 𝑇𝑏
𝑇𝑏
0, elsewhere
• where 𝜃 0 denotes phase value at 𝑡 = 0 (summing up the past), 𝑖 = 1,2, 𝐸𝑏 is transmitted energy per bit, and
𝑇𝑏 is bit duration.
▪ Binary symbol 1 → 𝑠1 𝑡 → 𝑓1
▪ Binary symbol 0 → 𝑠2 (𝑡) → 𝑓2

7 Dr. Ali Hussein Muqaibel

 Continuous-Phase-FSK (CPFSK)
• The signal 𝑠𝑖 (𝑡) can be expressed as

2𝐸𝑏
𝑠 𝑡 = cos 2𝜋𝑓𝑐 𝑡 + 𝜃(𝑡)
𝑇𝑏
4𝜋𝑑
• where the continuous-time phase is 3𝜋𝑑
2𝜋𝑑

𝜃 𝑡 − 𝜃(0)
𝜋𝑑
𝜃 𝑡 = 𝜃 0 ± 𝑇 𝑡, 0 ≤ 𝑡 ≤ 𝑇𝑏 𝜋𝑑
𝑏 𝑡
−𝜋𝑑
• and deviation ratio −2𝜋𝑑
−3𝜋𝑑
𝑑 = 𝑇𝑏 (𝑓1 − 𝑓2 ) −4𝜋𝑑

𝜋𝑑
• Binary symbol 1 ⇒ 𝜃 𝑡 = 𝜃 0 + 𝑇 𝑡 ⇒ 𝑠 𝑡 = 𝑠1 (𝑡) 2𝑇𝑏 4𝑇𝑏 6𝑇𝑏 8𝑇𝑏
𝑏

𝜋𝑑
• Binary symbol 0 ⇒ 𝜃 𝑡 = 𝜃 0 − 𝑇 𝑡 ⇒ 𝑠 𝑡 = 𝑠2 (𝑡)
𝑏

8 Dr. Ali Hussein Muqaibel

 Continuous-Phase-FSK (CPFSK)
4𝜋𝑑
3𝜋𝑑

𝜃 𝑡 − 𝜃(0)
2𝜋𝑑
𝜋𝑑 𝜋𝑑
• At 𝑡 = 𝑇𝑏 , 𝜃 𝑡 =𝜃 0 ± 𝑡, 0 ≤ 𝑡 ≤ 𝑇𝑏 𝑡
𝑇𝑏 −𝜋𝑑
−2𝜋𝑑
−3𝜋𝑑
𝜋𝑑, for symbol 1 −4𝜋𝑑
𝜃 𝑇𝑏 − 𝜃 0 = ቊ
−𝜋𝑑, for symbol 0
2𝑇𝑏 4𝑇𝑏 6𝑇𝑏 8𝑇𝑏

• Sending the symbol 1 increases the phase of CPFSK signal by 𝜋𝑑 radians; however, sending
the symbol 0 will reduce it by 𝜋𝑑
• Different variations of phase, 𝜃(𝑡), over time results in a phase tree (Phase Trellis)
1
→ CPFSK with 𝑑 = 1/2 is called minimum shift keying (MSK) , 𝑑 = 𝑇𝑏 (𝑓1 − 𝑓2 ), 𝑓1 − 𝑓2 =
2𝑇𝑏
𝜋
→ 𝜃 𝑡 = ± , 𝑡 = 𝑇𝑏
2

9 Dr. Ali Hussein Muqaibel

 𝜋𝑑
Minimum Shift Keying 𝜃 𝑡 =𝜃 0 ±
𝑇𝑏
𝑡, 0 ≤ 𝑡 ≤ 𝑇𝑏

• Example:
• Determine the phase trellis for a binary sequence 1101000 with 𝜃 0 = 0 and 𝑑 = 1/2.

• Frequency deviation is the minimum frequency spacing between 𝑓1 and 𝑓2 to be coherently orthogonal
.
1
→ CPFSK with 𝑑 = 1/2 is called minimum shift keying (MSK) , 𝑑 = 𝑇𝑏 (𝑓1 − 𝑓2 ), 𝑓1 − 𝑓2 = 2𝑇
𝑏
𝜋
→ 𝜃 𝑡 = ±2, 𝑡 = 𝑇𝑏

10 Dr. Ali Hussein Muqaibel

 Signal-Space of MSK
• Signal constellation for MSK is two-dimensional ( 𝑁 = 2 ) with four
possible message points (𝑀 = 4)
• What is the difference between QPSK and MSK?
• In QPSK, the symbol can be any of the 4 message points; however, in
MSK only 2 possible message points can be selected based on the
value of 𝜃(0)
• Signal-space diagram of MSK 𝜋𝑑
𝜃 𝑡 =𝜃 0 ± 𝑡, 0 ≤ 𝑡 ≤ 𝑇𝑏
𝑇𝑏

Transmitted Phase States Coordinates of Message

Symbol (radians) Points
𝜽(𝟎) 𝜽(𝑻𝒃 ) 𝒔𝟏 𝒔𝟐
0 0 −𝜋/2 + 𝐸𝑏 + 𝐸𝑏
1 𝜋 −𝜋/2 − 𝐸𝑏 + 𝐸𝑏
0 𝜋 𝜋/2 − 𝐸𝑏 − 𝐸𝑏
1 0 𝜋/2 + 𝐸𝑏 − 𝐸𝑏
11 Dr. Ali Hussein Muqaibel
 Power Spectra and Bandwidth Efficiency of 𝑀-ary FSK
2 log 2 𝑀
• If the 𝑀 FSK signals are separated by 1/2𝑇 to maintain 𝜌=
𝑀
orthogonality , the baseband PSD of 𝑀-FSK for 𝑀 = 2,4,8 are
shown in the figure.
• Channel bandwidth required to transmit 𝑀-ary FSK signals
𝑀
𝐵 = 2𝑇
1
• Note that 𝑇 = 𝑇𝑏 log 2 𝑀 , 𝑅𝑏 = 𝑇
𝑏
𝑅𝑏 2 log2 𝑀
• Therefore, the bandwidth efficiency [𝑏𝑝𝑠/𝐻𝑧] 𝜌 = 𝐵
= 𝑀

𝑀 2 4 8 16 32 64
𝜌 1 1 0.75 0.5 0.312 0.187

12 Dr. Ali Hussein Muqaibel

 Noncoherent Orthogonal Modulation
• Consider two orthogonal signals, 𝑠1 𝑡 and 𝑠2 (𝑡) with equal energy, 0 ≤ 𝑡 ≤ 𝑇,
• 𝑠1 𝑡 and 𝑠2 (𝑡) are sent over an imperfect channel (shifting the carrier phase).
• 𝑔1 (𝑡) and 𝑔2 (𝑡) are the phase-shifted signals (orthogonal and with same
energy 𝐸)
• Received signal is

𝑔1 𝑡 + 𝑤 𝑡 , 𝑠1 𝑡 sent
𝑥 𝑡 =ቊ
𝑔2 𝑡 + 𝑤 𝑡 , 𝑠2 𝑡 sent
where 0 ≤ 𝑡 ≤ 𝑇, 𝑤(𝑡) is additive white Gaussian noise (zero mean and PSD of 𝑁0 /2)

13 Dr. Ali Hussein Muqaibel

 Noncoherent Orthogonal Modulation
• Generalized binary receiver (noncoherent)

• Carrier phase is unknown → receiver relies on amplitude

• Error occurs if larger amplitude caused by noise alone results in a wrong decision

14 Dr. Ali Hussein Muqaibel

 1 𝐸𝑏
Noncoherent Binary FSK 𝑃𝑒 = exp −
2 2𝑁0

• In BFSK, transmitted signal is

2𝐸𝑏
𝑠𝑖 𝑡 = cos 2𝜋𝑓𝑖 𝑡 , 0 ≤ 𝑡 ≤ 𝑇𝑏
𝑇𝑏
0, elsewhere

• where 𝑓𝑖 = 𝑛𝑖 /𝑇𝑏 , 𝑖 = 1,2.

• Noncoherent BFSK is a special case of noncoherent
orthogonal modulation with 𝑇 = 𝑇𝑏 , 𝐸 = 𝐸𝑏
• Average bit error probability is

1 𝐸𝑏
𝑃𝑒 = exp −
2 2𝑁0 Noncoherent receiver for BFSK signals

15 Dr. Ali Hussein Muqaibel

 Differential Phase Shift Keying (DPSK)
• Differential phase shift keying (DPSK) is an example of noncoherent orthogonal modulation
• DPSK transmitter consists of
1. Differential encoding of binary wave
2. Phase shift keying
• Transmitted signal for a binary symbol 1 (𝑇𝑏 ≤ 𝑡 ≤ 2𝑇𝑏 )

2𝐸𝑏
cos 2𝜋𝑓𝑐 𝑡 , 0 ≤ 𝑡 ≤ 𝑇𝑏
𝑇𝑏
𝑠1 𝑡 =
2𝐸𝑏
cos 2𝜋𝑓𝑐 𝑡 , 𝑇𝑏 ≤ 𝑡 ≤ 2𝑇𝑏
𝑇𝑏

16 Dr. Ali Hussein Muqaibel

 1 𝐸𝑏
Differential Phase Shift Keying 𝑃𝑒 =
2
exp −
𝑁0

• Transmitted signal for a binary symbol 0 (𝑇𝑏 ≤ 𝑡 ≤ 2𝑇𝑏 )

2𝐸𝑏
cos 2𝜋𝑓𝑐 𝑡 , 0 ≤ 𝑡 ≤ 𝑇𝑏
𝑇𝑏
𝑠2 𝑡 =
2𝐸𝑏
cos 2𝜋𝑓𝑐 𝑡 + 𝜋 , 𝑇𝑏 ≤ 𝑡 ≤ 2𝑇𝑏
𝑇𝑏

• 𝑠1 (𝑡) and 𝑠2 𝑡 are orthogonal over the two-bit interval 0 ≤ 𝑡 ≤ 2𝑇𝑏

→ Special case of noncoherent modulation with 𝑇 = 2𝑇𝑏 , 𝐸 = 2𝐸𝑏

• Bit error probability for DPSK is

1 𝐸𝑏
𝑃𝑒 = exp −
2 𝑁0

17 Dr. Ali Hussein Muqaibel

 Differential Phase Shift Keying
• Generation and detection of DPSK
• If binary symbol, 𝑏𝑘 , is 1 → leave 𝑑𝑘 unchanged
• If binary symbol 𝑏𝑘 is 0 → change 𝑑𝑘
• Differentially encoded sequence, 𝑑𝑘 , introduces a phase shift in carrier with angles 0 (symbol
1) or 𝜋 (symbol 0)
• Example:
• Find the differentially encoded data for binary sequence 10010011

{𝒃𝒌 } 1 0 0 1 0 0 1 1
{𝑑𝑘−1 } 1 1 0 1 1 0 1 1
{𝑑𝑘 } 1 1 0 1 1 0 1 1 1
𝜃 0 0 𝜋 0 0 𝜋 0 0 0

18 Dr. Ali Hussein Muqaibel

 Performance of Digital
Modulation Techniques
over AWGN Channels
Dr. Ali Muqaibel
Ver. 2.4

Visit my website and Video Chanel to get

more resources: Muqaibel
Material here can be found in most of the popular Communications textbooks, check for example Communication Systems 1
by Simon Haykin (Ch6) or Wireless Communications by Andrea Goldsmith (Ch 5)
 Comparison of Digital Modulation
• A summary of bit error rate for Signaling Scheme Bit Error Rate
different digital modulation Coherent binary PSK
1 𝐸𝑏
schemes over AWGN channel Coherent QPSK erfc
2 𝑁0
Coherent MSK
Coherent binary FSK
1 𝐸𝑏
erfc
2 2𝑁0

DPSK 1
exp −𝐸𝑏 /𝑁0
2
Noncoherent binary FSK 1
exp −𝐸𝑏 /2𝑁0
2

2 Dr. Ali Hussein Muqaibel

 Bit error rate comparison
• Bit error rates decreases monotonically as 𝐸𝑏
/𝑁0 increases [waterfall curve]
• Coherent BPSK, QPSK, MSK results in less
BER than other schemes
• Coherent BPSK and DPSK requires less SNR
by 3 dB compared with coherent BFSK
• At high SNR, noncoherent BFSK performance
approaches coherent BFSK within 1 dB
– [similarly DPSK and coherent BPSK]

• MSK makes decision based on observation

over 2𝑇𝑏 → memory

3 Dr. Ali Hussein Muqaibel

 Bandwidth Efficient 𝑀-ary Digital Modulation
• A summary of power-bandwidth requirement for 𝑀-ary PSK (𝑃𝑒 = 10−4 )

Value of M 𝑩𝒂𝒏𝒅𝒘𝒊𝒅𝒕𝒉 𝑴−𝒂𝒓𝒚 𝑨𝒗𝒆𝒓𝒂𝒈𝒆 𝒑𝒐𝒘𝒆𝒓 𝑴−𝒂𝒓𝒚

𝑩𝒂𝒏𝒅𝒘𝒊𝒅𝒕𝒉 𝑩𝒊𝒏𝒂𝒓𝒚 𝑨𝒗𝒆𝒓𝒂𝒈𝒆 𝒑𝒐𝒘𝒆𝒓 𝑩𝒊𝒏𝒂𝒓𝒚

4 0.5 0.34 dB
8 0.333 3.91 dB
16 0.25 8.52 dB
32 0.2 13.52 dB
For the same Data rate

• QPSK is widely used in practice

→ It gives best trade-off between power and bandwidth

4 Dr. Ali Hussein Muqaibel

 Bandwidth Efficient 𝑀-ary Digital Modulation
• M-PSK and M-QAM have similar spectral and bandwidth
characteristics

• For 𝑀 > 4
▪ M-PSK is circular
▪ M-QAM is rectangular
• Distance between points is smaller in M-PSK than M-QAM

5 Dr. Ali Hussein Muqaibel

 King Fahd University of Petroleum & Minerals
Electrical Engineering Department
EE419: Wireless Communications
In Class Group Work
Signal Representation and Passband Communication (Ch5&Ch6)
Developed by Dr. Ali Muqaibel
Names (Group):

A signal constellation diagram is shown in the figure. Let the basis functions be as follows:
2 2
1 = cos ( 2 f c t ) , 0  t  T
T 2
where f c = b) 01 a) 00
2 T
2 = sin ( 2 f c t ) , 0  t  T
T
1
a) What type of modulation is being used (ASK, PSK, or FSK)? Why?

c) 10 11

b) Build (sketch) the block diagram for correlators based coherent receiver.

c) Let a=1, If at the output of the two correlators are as follows

Out of correlator with 1 0.92 -0.5 0.3

d) t
Out of correlator with  2 -0.7 -0.2 2

What is the received sequence of bits? Show how you get the answer?

d) Let a=1, Assuming that all decisions where correct in the previous question. Which one of the
three symbols was exposed to the largest amount of noise (Show how you find the answer)? How
much is this noise energy?
e) Re-draw the previous constellation diagram and re-assign the bits to the constellation points in a
better way. Justify your answer?

f) A new constellation diagram is shown. Assign the bits and find the values of x and y as
function of a such that the new constellation diagram exhibit the same probability of error as the
original diagram.
2
x

1
y

-x

g) On the new constellation diagram (part f), draw the decision boundaries.

h) Calculate the average energy per bit for the original constellation diagram and for the diagram
introduced in part (f). Which one of the two constellations do you prefer? Justify your answer.

i) Show that the two basis functions are orthogonal. (show all details)
2sin  cos = sin ( 2 )

________________________________________________________________________
Good Luck, Dr. Ali Muqaibel