Chapter 3: Physical-Layer Transmission Techniques: Section 3.1: Digital Modulations

Digital modulation techniques
Signal Space Analysis

Receiver Structure and Sufficient Statistics
Error Probability Analysis and the Union Bound
Passband modulation
Chapter 3: Physical-layer transmission techniques
Section 3.1: Digital modulations
Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 1

Passband modulation
Outline of the lecture notes

1 Digital modulation techniques
Advantages over analog modulation
Main considerations in digital modulation techniques
Typical types of digital modulation techniques
2 Signal Space Analysis

Rational
Signal and system model
Geometric representation of signals
Practical examples
Signal space representation
3 Receiver Structure and Sufficient Statistics

General results
Proofs of sufficient statistics for optimal detection
Decision regions and criterion
4 Error Probability Analysis and the Union Bound

Error probability
The union bound on error probability
5 Passband modulation
General principles
Amplitude and phase modulation
Pulse amplitude modulation (MPAM)
Phase shift keying (MPSK)
Quadrature amplitude modulation (MQAM)

Signal Space Analysis Advantages over analog modulation
Receiver Structure and Sufficient Statistics Main considerations in digital modulation techniques
Error Probability Analysis and the Union Bound Typical types of digital modulation techniques
Passband modulation
Advantages over analog modulation
The advances over the last several decades in hardware and

digital signal processing have made digital transceivers much
cheaper, faster, and more power-efficient than analog
transceivers.
More importantly, digital modulation offers a number of other
advantages over analog modulation, including:
higher data rates,
powerful error correction techniques,
resistance to channel impairments,
more efficient multiple access strategies, and
better security and privacy.

Passband modulation
Advantages over analog modulation (cont.)
Digital transmissions consist of transferring information in the

form of bits over a communications channel.
The bits are binary digits taking on the values of either 1 or 0.
These information bits are derived from the information
source, which may be a digital source or an analog source that
has been passed through an A/D converter.
Both digital and A/D converted analog sources may be
compressed to obtain the information bit sequence.

Passband modulation
Main considerations in digital modulation techniques
Digital modulation consists of mapping the information bits

into an analog signal for transmission over the channel.
Detection consists of determining the original bit sequence
based on the signal received over the channel.
The main considerations in choosing a particular digital
modulation technique are:
high data rate
high spectral efficiency (minimum bandwidth occupancy)
high power efficiency (minimum required transmit power)
robustness to channel impairments (minimum probability of bit
error)
low power/cost implementation

Passband modulation
Typical types of digital modulation techniques
Often the previous ones are conflicting requirements, and the

choice of modulation is based on finding the technique that
achieves the best tradeoff between these requirements.
There are two main categories of digital modulation:
amplitude/phase modulation
frequency modulation
Frequency modulation typically has a constant signal envelope
and is generated using nonlinear techniques, this modulation
is also called constant envelope modulation or nonlinear
modulation

Passband modulation
Typical types of digital modulation techniques (cont.)
Amplitude/phase modulation is also called linear modulation.

Linear modulation generally has better spectral properties
than nonlinear modulation, since nonlinear processing leads to
spectral broadening.
However, amplitude and phase modulation embeds the
information bits into the amplitude or phase of the
transmitted signal, which is more susceptible to variations
from fading and interference.

Passband modulation
Typical types of digital modulation techniques (cont.)
In addition, amplitude and phase modulation techniques

typically require linear amplifiers, which are more expensive
and less power efficient than the nonlinear amplifiers that can
be used with nonlinear modulation.
Thus, the general tradeoff of linear versus nonlinear
modulation is one of better spectral efficiency for the former
technique and better power efficiency and resistance to
channel impairments for the latter technique.

Digital modulation techniques Rational
Signal Space Analysis Signal and system model
Receiver Structure and Sufficient Statistics Geometric representation of signals
Error Probability Analysis and the Union Bound Practical examples
Passband modulation Signal space representation
Rational
Digital modulation encodes a bit stream of finite length into

one of several possible transmitted signals.
Intuitively, the receiver minimizes the probability of detection
error by decoding the received signal as the signal in the set of
possible transmitted signals that is closest to the one received.
Determining the distance between the transmitted and
received signals requires a metric for the distance between
signals.

Rational (cont.)
By representing signals as projections onto a set of basis

functions, we obtain a one-to-one correspondence between the
set of transmitted signals and their vector representations.
Thus, we can analyze signals in finite-dimensional vector
space instead of infinite-dimensional function space, using
classical notions of distance for vector spaces.
In this section we show:
how digitally modulated signals can be represented as vectors
in an appropriately-defined vector space, and
how optimal demodulation methods can be obtained from this
vector space representation.
This general analysis will then be applied to specific
modulation techniques in later sections.

Transmitted signal
AWGN Channel
n(t)
m ={b ,...,b } s(t) r(t) ^ ={b^ ,...,b^ }
i 1 K m
Transmitter + Receiver 1 K
Figure 1: Communication system model over AWGN channel (i.e., a

special case of wireless channel).
Consider a communication system model as shown in the

above figure.
Every 𝑇 seconds, the sytem sends 𝐾 = log2 𝑀 bits of
information through the channel for a data rate of 𝑅 = 𝐾/𝑇
bits per second (bps).
Transmitted signal (cont.)
There are 𝑀 = 2𝐾 possible sequences of 𝐾 bits and each bit

sequence of length 𝐾 comprises a message
𝑚𝑖 = {𝑏1 , ..., 𝑏𝐾 } ∈ 𝕄, where 𝕄 = {𝑚1 , ..., 𝑚𝑀 } is the set of
all such messages.
The message 𝑚𝑖 has∑ probability 𝑝𝑖 of being selected for
transmission, where 𝑀𝑖=1 𝑝𝑖 = 1.
Suppose that message 𝑚𝑖 is to be transmitted over the
AWGN channel during the time interval [0, 𝑇 ). Since the
channel is analog, the message must be embedded into an
analog signal for channel transmission.

Transmitted signal (cont.)
Therefore, each message 𝑚𝑖 ∈ 𝕄 is mapped to a unique

analog signal 𝑠𝑖 (𝑡) ∈ 𝕊 = {𝑠1 (𝑡), ..., 𝑠𝑀 (𝑡)} where 𝑠𝑖 (𝑡) is
defined on the time interval [0, 𝑇 ) and has energy
∫ 𝑇
𝐸𝑠𝑖 = 𝑠2𝑖 (𝑡)𝑑𝑡, 𝑖 = 1, ..., 𝑀. (1)
0
When messages 𝑚𝑖 are sent sequentially, the transmitted

signal becomes a sequence of the corresponding analog signals
as follows ∑
𝑠(𝑡) = 𝑠𝑖 (𝑡 − 𝑘𝑇 ). (2)
𝑘

Transmitted and received signals
In the aforementioned model, the transmitted signal is sent

through an AWGN channel where a white Gaussian noise
process 𝑛(𝑡) of power spectral density 𝑁𝑜 /2 is added to form
the received signal
𝑟(𝑡) = 𝑠(𝑡) + 𝑛(𝑡). (3)
s(t)
m m m
1 1 1
s (t) s (t−2T) s (t−3T) ...
1 1 1
s (t−T)
0 T 2 2T 3T 4T
m
2

Received signal
Given 𝑟(𝑡), the receiver must determine the best estimate of

which 𝑠𝑖 (𝑡) ∈ 𝕊 was transmitted during each transmission
interval [0, 𝑇 ).
This best estimate of 𝑠𝑖 (𝑡) is mapped to a best estimate of
the message 𝑚{𝑖 (𝑡) ∈ 𝕄 and} the receiver produces this best
estimate 𝑚 ˆ ˆ
ˆ = 𝑏1 , ..., 𝑏𝐾 of the transmitted bit sequence.
The goal of the receiver design in estimating the transmitted
message is to minimize the probability of message error
𝑀
∑
𝑃𝑒 = 𝑝 (𝑚
ˆ ∕= 𝑚𝑖 ∣𝑚𝑖 sent) 𝑝 (𝑚𝑖 sent) (4)
𝑖=1
over each time interval [0, 𝑇 ).

Introduction
By representing the signals {𝑠𝑖 (𝑡), 𝑖 = 1, ..., 𝑀 } geometrically,

one can solve for the optimal receiver design in AWGN
channels based on a minimum distance criterion.
Note that, wireless channels typically have a time-varying
impulse response in addition to AWGN. We will consider the
effect of an arbitrary channel impulse response on digital
modulation performance in the next sections.
The basic premise behind a geometrical representation of
signals is the notion of a basis set.

Basis function representation of signals
Specifically, using a Gram-Schmidt orthogonalization

procedure, it can be shown that any set of 𝑀 real energy
signals 𝑆 = {𝑠1 (𝑡), ..., 𝑠𝑀 (𝑡)} defined on [0, 𝑇 ) can be
represented as a linear combination of 𝑁 ≤ 𝑀 real
orthogonal basis functions {𝜙1 (𝑡), ..., 𝜙𝑀 (𝑡)}.
We say that these basis functions span the set 𝕊.
Each signal {𝑠𝑖 (𝑡) ∈ 𝕊} can be represented by
𝑁
∑
𝑠𝑖 (𝑡) = 𝑠𝑖,𝑗 𝜙𝑗 (𝑡), 0 ≤ 𝑡 < 𝑇, (5)
𝑗=1
where ∫ 𝑇
𝑠𝑖,𝑗 = 𝑠𝑖 (𝑡)𝜙𝑗 (𝑡)𝑑𝑡 (6)
0
is a real coefficient representing the projection.
Basis function representation of signals (cont.)
These basis functions have the following property

∫ 𝑇 {
1 𝑖 = 𝑗,
𝜙𝑖 (𝑡)𝜙𝑗 (𝑡)𝑑𝑡 = (7)
0 0 𝑖 ∕= 𝑗.
The basis set consists of the sine and cosine functions

√
2
𝜙1 (𝑡) = cos (2𝜋𝑓𝑐 𝑡) (8)
𝑇
and √
2
𝜙2 (𝑡) =sin (2𝜋𝑓𝑐 𝑡) . (9)
𝑇
√ ∫𝑇
2
where 𝑇 is used to obtain 0 𝜙2𝑖 (𝑡)𝑑𝑡 = 1, 𝑖 = 1, 2.

Basis functions in linear passband modulation techniques
With these basis functions, one only obtain an approximation

to (7), since
∫ 𝑇 ∫ 𝑇
2 sin (4𝜋𝑓𝑐 𝑡)
𝜙21 (𝑡)𝑑𝑡 = 0.5 [1 + cos (4𝜋𝑓𝑐 𝑡)] 𝑑𝑡 = 1+
0 𝑇 0 4𝜋𝑓𝑐 𝑡
(10)
The numerator in the second term of (10) is bounded by 1,
and for 𝑓𝑐 𝑇 ≫ 1 the denominator of this term is very large.
As a result, this second term can be neglected.

Basis functions in linear passband modulation (cont.)
With these basis functions, one can have

∫ 𝑇 ∫
2 𝑇 − cos (4𝜋𝑓𝑐 𝑡)
𝜙1 (𝑡)𝜙2 (𝑡)𝑑𝑡 = 0.5 sin (4𝜋𝑓𝑐 𝑡) 𝑑𝑡 = ≈0
0 𝑇 0 4𝜋𝑓𝑐 𝑡
(11)
where the approximation is taken as an equality as 𝑓𝑐 𝑇 ≫ 1.
√
With the√basis set 𝜙1 (𝑡) = 2/𝑇 cos (2𝜋𝑓𝑐 𝑡) and
𝜙2 (𝑡) = 2/𝑇 sin (2𝜋𝑓𝑐 𝑡), the basis function representation
(5) corresponds to the complex representation of 𝑠𝑖 (𝑡) in
terms of its in-phase
√ and quadrature components with an
extra factor of 2/𝑇 as follows
√ √
2 2
𝑠𝑖 (𝑡) = 𝑠𝑖,1 cos (2𝜋𝑓𝑐 𝑡) + 𝑠𝑖,2 sin (2𝜋𝑓𝑐 𝑡) . (12)
𝑇 𝑇
Basis functions in linear passband modulation (cont.)
In practice, the basis set may include a baseband pulse-shaping

filter 𝑔(𝑡) to improve the spectral characteristics of the transmitted
signal:
𝑠𝑖 (𝑡) = 𝑠𝑖,1 𝑔(𝑡) cos (2𝜋𝑓𝑐 𝑡) + 𝑠𝑖,2 𝑔(𝑡) sin (2𝜋𝑓𝑐 𝑡) (13)
where the simplest pulse

√ shape that satisfy (7) is the rectangular
pulse shape 𝑔(𝑡) = 2/𝑇 , 0 ≤ 𝑡 < 𝑇.

Definitions used in signal space representation
We denote the coefficients {𝑠𝑖,𝑗 } as a vector

s𝑖 = [𝑠𝑖,1 , ..., 𝑠𝑖,𝑁 ] ∈ ℛ𝑁 which is called the signal
constellation point corresponding to the signal 𝑠𝑖 (𝑡).
The signal constellation consists of all constellation points
{s1 , ..., s𝑀 }.
Given the basis functions {𝜙1 (𝑡), ..., 𝜙𝑁 (𝑡)} there is a
one-to-one correspondence between the transmitted signal
𝑠𝑖 (𝑡) and its constellation point s𝑖 .
The representation of 𝑠𝑖 (𝑡) in terms of its constellation point
s𝑖 ∈ ℛ𝑁 is called:
its signal space representation and
the vector space containing the constellation is called the
signal space.

Definitions used in signal space representation (cont.)
A two-dimensional signal space is illustrated in the below

figure, where we show s𝑖 ∈ ℛ2 with the 𝑖th axis of ℛ2
corresponding to the basis function 𝜙𝑖 (𝑡), 𝑖 = 1, 2.
s i2 s i2
M=4, K=2 M=8, K=3
01 011
010 001
11 00 110 000
si1 si1
110 100
10 101

With this signal space representation we can analyze the

infinite-dimensional functions 𝑠𝑖 (𝑡) as vectors s𝑖 in
finite-dimensional vector space ℛ2 .
This greatly simplifies the analysis of the system performance
as well as the derivation of the optimal receiver design.
Signal space representations for common modulation
techniques like MPSK and MQAM are two-dimensional
(corresponding to the in-phase and quadrature basis
functions).
In order to analyze signals via a signal space representation,
we need to use some definitions for the vector characterization
in the vector space ℛ𝑁 .

In particular, the length of a vector in ℛ𝑁 is defined as

v
u𝑁
u∑
∥s𝑖 ∥ = ⎷ 𝑠2𝑖,𝑗 . (14)
𝑗=1
The distance between two signal constellation points s𝑖 and s𝑘

is thus
v √
u𝑁 ∫ 𝑇
u∑
∥s𝑖 − s𝑘 ∥ = ⎷ 2
(𝑠𝑖,𝑗 − 𝑠𝑘,𝑗 ) = (𝑠𝑖 (𝑡) − 𝑠𝑘 (𝑡))2 𝑑𝑡.
𝑗=1 0
(15)

Finally, the inner product ⟨𝑠𝑖 (𝑡), 𝑠𝑘 (𝑡)⟩ between two real
signals 𝑠𝑖 (𝑡) and 𝑠𝑘 (𝑡) on the interval [0, 𝑇 ) is defined as
∫ 𝑇
⟨𝑠𝑖 (𝑡), 𝑠𝑘 (𝑡)⟩ = 𝑠𝑖 (𝑡)𝑠𝑘 (𝑡)𝑑𝑡. (16)
0
Similarly, the inner product ⟨s𝑖 , s𝑘 ⟩ between two real vectors is

∫ 𝑇
⟨s𝑖 , s𝑘 ⟩ = s𝑖 s𝑇𝑘 = 𝑠𝑖 (𝑡)𝑠𝑘 (𝑡)𝑑𝑡 = ⟨𝑠𝑖 (𝑡), 𝑠𝑘 (𝑡)⟩. (17)
0
It is noted that two signals are orthogonal if their inner

product is zero.

Signal Space Analysis General results
Receiver Structure and Sufficient Statistics Proofs of sufficient statistics for optimal detection
Error Probability Analysis and the Union Bound Decision regions and criterion
Passband modulation
Receiver structure and sufficient statistics
Given the channel output 𝑟(𝑡) = 𝑠𝑖 (𝑡) + 𝑛(𝑡), 0 ≤ 𝑡 < 𝑇 , we

now investigate the receiver structure to determine which
constellation point s𝑖 or, equivalently, which message 𝑚𝑖 , was
sent over the time interval [0, 𝑇 ).
A similar procedure is done for each time interval
[𝑘𝑇, (𝑘 + 1)𝑇 ).
We would like to convert the received signal 𝑟(𝑡) over each
time interval into a vector, since it allows us to work in
finite-dimensional vector space to estimate the transmitted
signal.
However, this conversion should not compromise the
estimation accuracy. For this conversion, consider the receiver
structure shown in the next figure.
Passband modulation
Receiver structure and sufficient statistics (cont.)
T si ,1 ! n1 r1
"0 ()dt
r (t ) ! si (t ) n(t ) 1 (t ) mˆ mi
Find i
T si , N ! n N rN
()dt
0
N (t )
As shown in the above figure, the components of signal and

noise vectors are determined by
∫ 𝑇
𝑠𝑖,𝑗 = 𝑠𝑖 (𝑡)𝜙𝑗 (𝑡)𝑑𝑡, (18)
0
and ∫ 𝑇
𝑛𝑗 = 𝑛(𝑡)𝜙𝑗 (𝑡)𝑑𝑡. (19)
0
Passband modulation
Receiver structure and sufficient statistics (cont.)
We can rewrite 𝑟(𝑡) as

𝑁
∑ 𝑁
∑
𝑟(𝑡) = (𝑠𝑖,𝑗 + 𝑛𝑗 ) 𝜙𝑗 (𝑡) + 𝑛𝑟 (𝑡) = 𝑟𝑗 𝜙𝑗 (𝑡) + 𝑛𝑟 (𝑡),
𝑗=1 𝑗=1
∑𝑁 (20)
where 𝑟𝑗 = 𝑠𝑖,𝑗 + 𝑛𝑗 and 𝑛𝑟 (𝑡) = 𝑛(𝑡) − 𝑗=1 𝑛𝑗 𝜙𝑗 (𝑡)
denotes the remainder noise.

Passband modulation
Proofs of sufficient statistics for optimal detection
If we can show that the optimal detection of the transmitted

signal constellation point s𝑖 given received signal 𝑟(𝑡) does not
make use of the remainder noise 𝑛𝑟 (𝑡), then the receiver can
make its estimate 𝑚 ˆ of the transmitted message 𝑚𝑖 as a
function of r = (𝑟1 , ..., 𝑟𝑁 ) alone.
In other words, r = (𝑟1 , ..., 𝑟𝑁 ) is a sufficient statistic for 𝑟(𝑡)
in the optimal detection of the transmitted messages.
Let exam the distribution of r. Since 𝑛(𝑡) is a Gaussian
random process, if we condition on the transmitted signal
𝑠𝑖 (𝑡) then the channel output 𝑟(𝑡) = 𝑠𝑖 (𝑡) + 𝑛(𝑡) is also a
Gaussian random process and r = [𝑟1 , ..., 𝑟𝑁 ] is a Gaussian
random vector.

Passband modulation
Proofs of sufficient statistics for optimal detection (cont.)

Recall that 𝑟𝑖 = 𝑠𝑖,𝑗 + 𝑛𝑗 . Thus, conditioned on the
transmitted constellation s𝑖 , we have
𝜇𝑟𝑗 ∣s𝑖 = 𝐸 [𝑟𝑗 ∣s𝑖 ] = 𝐸 [𝑠𝑖,𝑗 + 𝑛𝑗 ∣𝑠𝑖,𝑗 ] = 𝑠𝑖,𝑗 (21)
since n(t) has zero mean, and
[ ]2 [ ]
𝜎𝑟𝑗 ∣s𝑖 = 𝐸 𝑟𝑗 − 𝜇𝑟𝑗 ∣s𝑖 = 𝐸 [𝑠𝑖,𝑗 + 𝑛𝑗 − 𝑠𝑖,𝑗 ∣𝑠𝑖,𝑗 ]2 = 𝐸 𝑛2𝑗 .
(22)
[( ) ]
With Cov [𝑟𝑗 𝑟𝑘 ∣s𝑖 ] = 𝐸 𝑟𝑗 − 𝜇𝑟𝑗 (𝑟𝑘 − 𝜇𝑟𝑘 ) ∣s𝑖 = 𝐸 [𝑛𝑗 𝑛𝑘 ]
and some manipulations, one can have
∫ {
𝑁0 𝑇 𝑁0 /2 𝑗 = 𝑘
𝐸 [𝑛𝑗 𝑛𝑘 ] = 𝜙𝑗 (𝑡)𝜙𝑘 (𝑡)𝑑𝑡 = . (23)
2 0 0 𝑗 ∕= 𝑘.

Passband modulation
Thus, conditioned on the transmitted constellation s𝑖 , the 𝑟𝑗 ’s

are uncorrelated and, since [they] are Gaussian and also
2
independent. Moreover, 𝐸 𝑛𝑗 = 𝑁0 /2.
We have shown that, conditioned on the transmitted
constellation s𝑖 , 𝑟𝑗 is a Gauss-distributed random variable that
is independent of 𝑟𝑘 , 𝑘 ∕= 𝑗and has mean 𝑠𝑖,𝑗 and variance
𝑁0 /2.
Thus, the conditional distribution of r is given by
⎡ ⎤
𝑁
∏ ∑𝑁
1 1 2
𝑝 (r∣s𝑖 sent) = 𝑝 (𝑟𝑗 ∣𝑚𝑖 ) = 𝑁/2
exp ⎣− (𝑟𝑗 − 𝑠𝑖,𝑗 ) ⎦ .
𝑗=1 (𝜋𝑁0 ) 𝑁0 𝑗=1
(24)

Passband modulation
It is also straightforward to show that 𝐸 [𝑟𝑗 𝑛𝑟 (𝑡)∣s𝑖 ] = 0 for

any 𝑡, 0 ≤ 𝑡 < 𝑇 . Thus, since 𝑟𝑗 conditioned on s𝑖 and 𝑛𝑟 (𝑡)
are Gaussian and uncorrelated, they are independent.
Also, since the transmitted signal is independent of the noise,
𝑠𝑖,𝑗 is independent of the process 𝑛𝑟 (𝑡).
We now discuss the receiver design criterion and show it is not
affected by discarding 𝑛𝑟 (𝑡).
The goal of the receiver design is to minimize the probability
of error in detecting the transmitted message 𝑚𝑖 given
received signal 𝑟(𝑡).
To minimize 𝑃𝑒 = 𝑝 (𝑚
ˆ ∕= 𝑚𝑖 ∣𝑟(𝑡)) = 1 − 𝑝 (𝑚
ˆ = 𝑚𝑖 ∣𝑟(𝑡)), we
maximize 𝑝 (𝑚
ˆ = 𝑚𝑖 ∣𝑟(𝑡)).

Passband modulation
Therefore, the receiver output 𝑚

ˆ given received signal 𝑟(𝑡)
should correspond to the message 𝑚𝑖 that maximizes
𝑝 (𝑚𝑖 sent∣𝑟(𝑡)).
Since there is a one-to-one mapping between messages and
signal constellation points, this is equivalent to maximizing
𝑝 (𝑚𝑖 sent∣𝑟(𝑡)).

Passband modulation
Recalling that 𝑟(𝑡) is completely described by 𝑟 = (𝑟1 , ..., 𝑟𝑁 )

and 𝑛𝑟 (𝑡), we have
𝑝 (s𝑖 sent∣𝑟(𝑡)) = 𝑝 ((𝑠𝑖,1 , ..., 𝑠𝑖,𝑁 ) sent∣(𝑟1 , ..., 𝑟𝑁 , 𝑛𝑟 (𝑡)))

𝑝 ((𝑠𝑖,1 , ..., 𝑠𝑖,𝑁 ) sent, (𝑟1 , ..., 𝑟𝑁 ), 𝑛𝑟 (𝑡))
=
𝑝 ((𝑟1 , ..., 𝑟𝑁 ), 𝑛𝑟 (𝑡))
𝑝 ((𝑠𝑖,1 , ..., 𝑠𝑖,𝑁 ) sent, (𝑟1 , ..., 𝑟𝑁 )) 𝑝 (𝑛𝑟 (𝑡))
=
𝑝 ((𝑟1 , ..., 𝑟𝑁 )) 𝑝 (𝑛𝑟 (𝑡))
= 𝑝 ((𝑠𝑖,1 , ..., 𝑠𝑖,𝑁 ) sent∣(𝑟1 , ..., 𝑟𝑁 )) . (25)
where the third equality follows from the fact that the 𝑛𝑟 (𝑡) is
independent of both (𝑟1 , ..., 𝑟𝑁 ) and of (𝑠𝑖,1 , ..., 𝑠𝑖,𝑁 ).

Passband modulation
This analysis shows that (𝑟1 , ..., 𝑟𝑁 ) is a sufficient statistic for

𝑟(𝑡) in detecting 𝑚𝑖 , in the sense that the probability of error
is minimized by using only this sufficient statistic to estimate
the transmitted signal and discarding the remainder noise.
Since r is a sufficient statistic for the received signal 𝑟(𝑡), we
call r the received vector associated with 𝑟(𝑡).

Passband modulation
Decision regions
As aforementioned, the optimal receiver minimizes error

probability by selecting the detector output 𝑚
ˆ that maximizes
the probability of correct detection
1 − 𝑃𝑒 = 𝑝 (𝑚ˆ sent∣r received).
In other words, given a received vector r, the optimal receiver
ˆ = 𝑚𝑖 corresponding to the constellation s𝑖 that
selects 𝑚
satisfies
𝑝 (s𝑖 ∣r) > 𝑝 (s𝑗 ∣r) , ∀𝑗 ∕= 𝑖 (26)
where 𝑝 (s𝑖 ∣r) ≜ 𝑝 (s𝑖 sent∣r received) for the sake of
notational simplicity.

Passband modulation
Decision regions(cont.)
Thus, the decision regions (𝑍1 , ..., 𝑍𝑀 ) corresponding to

(s1 , ..., s𝑀 ) are the subsets of the signal space ℛ𝑁 and
defined by
𝑍𝑖 = (r : 𝑝 (s𝑖 ∣r) > 𝑝 (s𝑗 ∣r) , ∀𝑗 ∕= 𝑖) . (27)
Once the signal space has been partitioned by decision

regions, for a received vector r ∈ 𝑍𝑖 , the optimal receiver
outputs the message estimate 𝑚 ˆ = 𝑚𝑖
The receiver processing consists of 𝑎) computing the received
vector r from 𝑟(𝑡), 𝑏) finding which decision region 𝑍𝑖
contains r, and 𝑐) outputting the corresponding message 𝑚𝑖 .

Passband modulation
An example on decision regions

This process is illustrated in the below figure, that shows a
two-dimensional signal space with four decision regions
𝑍1 , ..., 𝑍4 corresponding to four constellations s1 , ..., s4 .
The received vector r lies in region 𝑍1 , so the receiver will
output the message 𝑚1 as the best message estimate given
received vector r.
φ2(t)
Z
s 2
2
x
s
3 s
1
φ1(t)
Z Z
3 1
s Z
4
4

Passband modulation
Decision criterion
Using Bayes rule, one can have
𝑝 (r∣s𝑖 ) 𝑝 (s𝑖 )
𝑝 (s𝑖 ∣r) = . (28)
𝑝 (r)
To minimize error probability, the receiver output 𝑚 ˆ = 𝑚𝑖
corresponds to the constellation point s𝑖 that maximizes
𝑝 (s𝑖 ∣r), i.e., the detected transmitted constellation point ŝ
can be determined by
𝑝 (r∣s𝑖 ) 𝑝 (s𝑖 )
ŝ = arg max = arg max 𝑝 (r∣s𝑖 ) 𝑝 (s𝑖 ) , 𝑖 = 1, ..., 𝑀
s𝑖 𝑝 (r) s𝑖
(29)
where the second equality follows from the fact that 𝑝(r) is
not a function of s𝑖
Passband modulation
Decision criterion (cont.)
Assuming that transmitted messages 𝑚𝑖 are equally likely

(i.e., 𝑝(s𝑖 ) = 1/𝑀 ), (29) becomes
ŝ = arg max 𝑝 (r∣s𝑖 ) , 𝑖 = 1, ..., 𝑀. (30)

s𝑖
Let define the likelihood function associated with the receiver

as
𝑓 (s𝑖 ) = 𝑝 (r∣s𝑖 ) . (31)
Given a received vector r, a maximum likelihood (ML)
receiver outputs 𝑚
ˆ = 𝑚𝑖 corresponding to the constellation
point s𝑖 that maximizes 𝑓 (s𝑖 ).

Passband modulation
Since the log function is increasing in its argument,

maximizing s𝑖 is equivalent to maximizing the log likelihood
function, defined as 𝑓𝑀 𝐿 (s𝑖 ) = log 𝑓 (s𝑖 ). Using (24) for
𝑓𝑀 𝐿 (s𝑖 ) = log 𝑓 (s𝑖 ) yields
𝑁
1 ∑ 2 1 2
𝑓𝑀𝐿 (s𝑖 ) = − (𝑟𝑗 − 𝑠𝑖,𝑗 ) +constant = − ∥r − s𝑖 ∥ +constant.
𝑁0 𝑗=1 𝑁0
(32)
Based on (30), the detected transmitted constellation point ŝ can
be determined by the ML criterion as
2
ŝ = arg min ∥r − s𝑖 ∥ , 𝑖 = 1, ..., 𝑀. (33)
s𝑖

Passband modulation
Under the aforementioned assumption of equiprobable

(transmitted) messages 𝑚𝑖 , the ML structure minimizes the
probability of detection error at the receiver.
Under the aforementioned assumption of equiprobable
(transmitted) messages 𝑚𝑖 , the ML structure minimizes the
probability of detection error at the receiver.

Error probability
Passband modulation
Error probability of ML detection

With 𝑝 (𝑚𝑖 sent) = 1/𝑀 , the error probability of the ML receiver:
𝑀
∑
𝑃𝑒 = 𝑝 (r ∈
/ 𝑍𝑖 ∣𝑚𝑖 sent) 𝑝 (𝑚𝑖 sent)
𝑖=1
𝑀
1 ∑
= 𝑝 (r ∈
/ 𝑍𝑖 ∣𝑚𝑖 sent)
𝑀 𝑖=1
𝑀
s3 1 ∑
= 1− 𝑝 (r ∈ 𝑍𝑖 ∣𝑚𝑖 sent)
s4 s2
𝑀 𝑖=1
d min
𝑀 ∫
s5 s1
Z1 1 ∑
0 = 1− 𝑝 (r∣𝑚𝑖 ) 𝑑r
r=s 1+n 𝑀 𝑖=1 𝑍𝑖
s6
𝑀 ∫
s8
s7 1 ∑
= 1− 𝑝 (r = s𝑖 + n∣s𝑖 ) 𝑑r
𝑀 𝑖=1 𝑍𝑖
Mobile communications-Chapter 3: Physical-layer transmissions ∑
Section 3.1: ∫ modulations
𝑀Digital 44
Error probability
Passband modulation
As observed, (34) gives an exact solution to the error

probability but it is impossible to solve this error probability in
closed-form. Therefore, the union bound on error probability
is investigated.
Let 𝐴𝑖,𝑘 denote the event that ∥r − s𝑘 ∥ < ∥r − s𝑖 ∥ given that
the constellation point s𝑖 was sent.
If the event 𝐴𝑖,𝑘 occurs, then the constellation point will be
decoded in error since the transmitted constellation point s𝑖 is
not the closest constellation point to the received vector r.
However, event 𝐴𝑖,𝑘 does not necessarily imply that s𝑘 will be
decoded instead of s𝑖 , since there may be another
constellation point s𝑙 with ∥r − s𝑙 ∥ < ∥r − s𝑘 ∥ < ∥r − s𝑖 ∥.

Error probability
Passband modulation
The union bound on error probability (cont.)

The constellation point is decoded correctly if
∥r − s𝑖 ∥ < ∥r − s𝑘 ∥ ∀𝑘 ∕= 𝑖. Therefore,
⎛ ⎞
𝑀 𝑀
⎜∪ ⎟ ∑
𝑃𝑒 (𝑚𝑖 sent) = 𝑝 ⎜
⎝ 𝐴 ⎟
𝑖,𝑘 ⎠ ≤ 𝑝 (𝐴𝑖,𝑘 ) . (35)
𝑘=1 𝑘=1
𝑘∕=𝑖 𝑘∕=𝑖
where the inequality follows from the union bound on

probability.
More specifically, 𝑝 (𝐴𝑖,𝑘 ) can be determined by
𝑝 (𝐴𝑖,𝑘 ) = 𝑝 (∥s𝑘 − r∥ < ∥s𝑖 − r∥ ∣𝑚𝑖 sent)
= 𝑝 (∥s𝑘 − (s𝑖 + n)∥ < ∥s𝑖 − (s𝑖 + n)∥)
= 𝑝 (∥n + s𝑖 − s𝑘 ∥ < ∥n∥)
( )
= 𝑝 2 ∥n∥ ∥s𝑖 − s𝑘 ∥ cos 𝛼 < − ∥s𝑖 − s𝑘 ∥2
Error probability
Passband modulation
Since ⟨n, s𝑖 − s𝑘 ⟩ = ∥n∥ ∥s𝑖 − s𝑘 ∥ cos 𝛼, one will have

𝑛 = ∥n∥ cos 𝛼 = ⟨n,s 𝑖 −s𝑘 ⟩
∥s𝑖 −s𝑘 ∥ is a Gaussian random variable with
zero-mean and variance 𝑁0 /2
As a result, 𝑝 (𝐴𝑖,𝑘 ) can be simplified to
( ) ( )
∥s𝑖 − s𝑘 ∥ 𝑑𝑖,𝑘
𝑝 (𝐴𝑖,𝑘 ) = 𝑝 𝑛 < − =𝑝 𝑛>
2 2
∫ ∞ ( 2
) ( )
1 −𝑛 𝑑𝑖,𝑘
= √ exp 𝑑𝑛 = 𝑄 √ .
𝑑𝑖,𝑘 /2 𝜋𝑁0 𝑁0 2𝑁0

Error probability
Passband modulation

Substituting (36) into (35), one can have
∑𝑀 ( )
𝑑𝑖,𝑘
𝑃𝑒 (𝑚𝑖 sent) ≤ 𝑄 √ . (36)
𝑘=1
2𝑁0
𝑘∕=𝑖
where the 𝑄 function, 𝑄(𝑧), is defined as the probability that

a Gaussian random variable 𝑥 with zero-mean and variance of
1 is bigger than 𝑧, i.e.,
∫ ∞ ( )
1 −𝑥2
𝑄 (𝑧) = 𝑝 (𝑥 > 𝑧) = √ exp 𝑑𝑥. (37)
𝑧 2𝜋 2
Summing (36) over all possible messages yields the union bound
∑𝑀 𝑀 𝑀 ( )
1 ∑∑ 𝑑𝑖,𝑘
𝑃𝑒 = 𝑝 (𝑚𝑖 ) 𝑃𝑒 (𝑚𝑖 sent) ≤ 𝑄 √ . (38)
𝑖=1
𝑀 𝑖=1 2𝑁0
𝑘=1
𝑘∕=𝑖
Error probability
Passband modulation

Note that the 𝑄 function cannot be solved in closed-form. It
can be obtained
( from
) the complementary error function as
𝑄(𝑧) = 12 erfc √𝑧2 .
One can upper bound 𝑄(𝑧) with the closed-form expression
1 ( )
𝑄(𝑧) ≤ √ exp −𝑧 2 /2 . (39)
𝑧 2𝜋
and this bound is quite tight for 𝑧 ≫ 0.
Let define the minimum distance of the constellation as
𝑑𝑚𝑖𝑛 = min𝑖,𝑘 𝑑𝑖,𝑘 , one can simplify (41) with looser bound
( )
𝑑𝑚𝑖𝑛
𝑃𝑒 ≤ (𝑀 − 1)𝑄 √ . (40)
2𝑁0

Error probability
Passband modulation
Using (39) for the 𝑄 function yields a closed-form bound

( 2 )
𝑀 −1 −𝑑𝑚𝑖𝑛
𝑃𝑒 ≤ √ exp . (41)
𝜋 4𝑁0
Note that for binary modulation (𝑀 = 2), there is only one

way to make an error and 𝑑𝑚𝑖𝑛 is the distance between the
two signal constellation points, so the bound is exact
( )
𝑑𝑚𝑖𝑛
𝑃𝑏 = 𝑄 √ . (42)
2𝑁0

Error probability
Passband modulation
Bit error rate
Recall that 𝑃𝑒 is the probability of a symbol (message) error:

𝑃𝑒 = 𝑝 (𝑚ˆ ∕= 𝑚𝑖 ∣𝑚𝑖 sent) where 𝑚𝑖 corresponds to a message
with 𝑙𝑜𝑔2 𝑀 bits.
However, system designers are typically more interested in the
bit error probability, also called the bit error rate (BER), than
in the symbol error probability, since bit errors drive the
performance of higher layer networking protocols and
end-to-end performance.
Thus, we would like to design the mapping of 𝑀 possible bit
sequences to a message 𝑚𝑖 , 𝑖 = 1, ..., 𝑀 so that a symbol
error associated with an adjacent decision region, which is the
most likely way to make an error, corresponds to only one bit
error.
Error probability
Passband modulation
Bit error rate (cont.)
With such a mapping, assuming that mistaking a signal

constellation for a constellation other than its nearest
neighbors has a very low probability, we can make the
approximation
𝑃𝑒
𝑃𝑏 ≈ . (43)
𝑙𝑜𝑔2 𝑀
The most common form of mapping with the property is
called Gray coding, which will be discussed in more detail.
Signal space concepts are applicable to any modulation where
bits are encoded as one of several possible analog signals,
including the amplitude, phase, and frequency modulations as
discussed later.

Digital modulation techniques General principles
Signal Space Analysis Amplitude and phase modulation
Receiver Structure and Sufficient Statistics Pulse amplitude modulation (MPAM)
Error Probability Analysis and the Union Bound Phase shift keying (MPSK)
Passband modulation Quadrature amplitude modulation (MQAM)
General principles
The basic principle of passband digital modulation is to

encode an information bit stream into a carrier signal which is
then transmitted over a communications channel.
Demodulation is the process of extracting this information bit
stream from the received signal. Corruption of the transmitted
signal by the channel can lead to bit errors in the
demodulation process.
The goal of modulation is to send bits at a high data rate
while minimizing the probability of data corruption.
In general, modulated carrier signals encode information in the
amplitude 𝛼(𝑡), frequency 𝑓 (𝑡), or phase 𝜃(𝑡) of a carrier
signal.

General principles (cont.)
Thus, the modulated signal can be represented as
𝑠(𝑡) = 𝛼(𝑡) cos [2𝜋 (𝑓𝑐 + 𝑓 (𝑡)) + 𝜃(𝑡) + 𝜙0 ] = 𝛼(𝑡) cos [2𝜋𝑓𝑐 𝑡 + 𝜙(𝑡) + 𝜙0 ]
where 𝜙(𝑡) = 2𝜋𝑓 (𝑡)𝑡 + 𝜃(𝑡) and 𝜙0 is the phase offset of the
carrier. This representation combines frequency and phase
modulation into angle modulation.
One can rewrite the right-hand side of (44) in terms of its in-phase
and quadrature components as:
𝑠(𝑡) = 𝛼(𝑡) cos 𝜙(𝑡) cos [2𝜋𝑓𝑐 𝑡] − 𝛼(𝑡) sin 𝜙(𝑡) sin [2𝜋𝑓𝑐 𝑡]
= 𝑠𝐼 (𝑡) cos [2𝜋𝑓𝑐 𝑡] − 𝑠𝑄 (𝑡) sin [2𝜋𝑓𝑐 𝑡] (44)
where 𝑠𝐼 (𝑡) = 𝛼(𝑡) cos 𝜙(𝑡) is called the in-phase component of 𝑠(𝑡)
and 𝑠𝑄 (𝑡) = 𝛼(𝑡) sin 𝜙(𝑡) is called its quadrature component.

General principles (cont.)
We can write 𝑠(𝑡) in its complex baseband representation as

[ ]
𝑠(𝑡) = Re 𝑢(𝑡)𝑒𝑗2𝜋𝑓𝑐 𝑡 (45)
where 𝑢(𝑡) = 𝑠𝐼 (𝑡) + 𝑗𝑠𝑄 (𝑡).

This representation is useful since receivers typically process
the in-phase and quadrature signal components separately.

Amplitude and phase modulation

In amplitude and phase modulation the information bit stream
is encoded in the amplitude and/or phase of the transmitted
signal.
Specifically, over a time interval of 𝑇𝑠 , 𝐾 = log2 𝑀 bits are
encoded into the amplitude and/or phase of the transmitted
signal 𝑠(𝑡), 0 ≤ 𝑡 < 𝑇𝑠 .
The transmitted signal over this period
𝑠(𝑡) = 𝑠𝐼 (𝑡) cos [2𝜋𝑓𝑐 𝑡] − 𝑠𝑄 (𝑡) sin [2𝜋𝑓𝑐 𝑡] can be written in
terms of its signal space representation as
𝑠(𝑡) = 𝑠𝑖,1 𝜙1 (𝑡) + 𝑠𝑖,2 𝜙2 (𝑡) (46)
where basis functions 𝜙1 (𝑡) = 𝑔(𝑡) cos (2𝜋𝑓𝑐 𝑡 + 𝜙0 ) and

𝜙2 (𝑡) = −𝑔(𝑡) sin (2𝜋𝑓𝑐 𝑡 + 𝜙0 ), where 𝑔(𝑡) is a shaping pulse.
Amplitude and phase modulation (cont.)
To send the 𝑖th message over the time interval [𝑘𝑇, (𝑘 + 1)𝑇 ),
we set 𝑠𝐼 (𝑡) = 𝑠𝑖,1 𝑔(𝑡) and 𝑠𝑄 (𝑡) = 𝑠𝑖,2 𝑔(𝑡). These in-phase
and quadrature signal components are baseband signals with
spectral characteristics determined by the pulse shape 𝑔(𝑡).
In particular, their bandwidth 𝐵 equals the bandwidth of g(t),
and the transmitted signal 𝑠(𝑡) is a passband signal with
center frequency fc and passband bandwidth 2𝐵.
In practice we take 𝐵 = 𝐾𝑔 /𝑇𝑠 where 𝐾𝑔 depends on the
pulse shape: for rectangular pulses 𝐾𝑔 = .5 and for raised
cosine pulses .5 ≤ 𝐾𝑔 ≤ 1.
Thus, for rectangular pulses the bandwidth of 𝑔(𝑡) is .5/𝑇𝑠
and the bandwidth of 𝑠(𝑡) is 1/𝑇𝑠 .

Since the pulse shape 𝑔(𝑡) is fixed, the signal constellation for
amplitude and phase modulation is defined based on the
constellation point: (𝑠𝑖,1 , 𝑠𝑖,2 ) ∈ ℝ2 , 𝑖 = 1, ..., 𝑀 .
The complex baseband representation of 𝑠(𝑡) is
[ ]
𝑠(𝑡) = Re 𝑥(𝑡)𝑒𝑗𝜙0 𝑒𝑗2𝜋𝑓𝑐 𝑡 (47)
where:
𝑥(𝑡) = 𝑠𝐼 (𝑡) + 𝑗𝑠𝑄 (𝑡) = (𝑠𝑖,1 + 𝑗𝑠𝑖,2 ) 𝑔(𝑡).
The constellation point s𝑖 = (𝑠𝑖,1 , 𝑠𝑖,2 ) is called the symbol
associated with the log2 𝑀 bits and
𝑇𝑠 is called the symbol time and the bit rate for this
modulation is 𝐾 bits per symbol or 𝑅 = log2 𝑀/𝑇𝑠 bits per
second.
There are three main types of amplitude/phase modulation:

Pulse Amplitude Modulation (MPAM): information encoded in
amplitude only.
Phase Shift Keying (MPSK): information encoded in phase
only.
Quadrature Amplitude Modulation (MQAM): information
encoded in both amplitude and phase.
The number of bits per symbol 𝐾 = log2 𝑀 , signal
constellation (𝑠𝑖,1 , 𝑠𝑖,2 ) ∈ ℝ2 , 𝑖 = 1, ..., 𝑀 , and the choice of
shaping pulse 𝑔(𝑡) determines the digital modulation design.
The pulse shape 𝑔(𝑡) is designed to improve spectral efficiency
and combat inter-symbol-interference (ISI).

Amplitude and phase modulation over a given symbol period

can be generated using the modulator structure shown in the
next Figure.
Note that the basis functions in this figure have an arbitrary
phase 𝜙0 associated with the transmit oscillator.
Demodulation over each symbol period is performed using the
demodulation structure of Figure xx +1, which is equivalent
to the structure of for 𝜙1 (𝑡) = 𝑔(𝑡) cos (2𝜋𝑓𝑐 𝑡 + 𝜙) and
𝜙2 (𝑡) = −𝑔(𝑡) sin (2𝜋𝑓𝑐 𝑡 + 𝜙).
Typically the receiver includes some additional circuitry for
carrier phase recovery that matches the carrier phase 𝜙 at the
receiver to the carrier phase 𝜙0 at the transmitter, which is
called coherent detection.

If 𝜙 − 𝜙0 = △𝜙 ∕= 0 then the in-phase branch will have an
unwanted term associated with the quadrature branch and
vice versa, i.e., 𝑟1 = 𝑠𝑖,1 cos (△𝜙) + 𝑠𝑖,2 sin (△𝜙) + 𝑛1 and
𝑟2 = 𝑠𝑖,1 sin (△𝜙) + 𝑠𝑖,2 cos (△𝜙) + 𝑛2 can result in significant
performance degradation.
The receiver structure also assumes that the sampling function
every 𝑇𝑠 seconds is synchronized to the start of the symbol
period, which is called synchronization or timing recovery.
Receiver synchronization and carrier phase recovery are
complex receiver operations that can be highly challenging in
wireless environments.
We will assume perfect carrier recovery in our discussion of
MPAM, MPSK and MQAM, and therefore set 𝜙 = 𝜙0 = 0 for
their analysis.
Pulse amplitude modulation (MPAM)
We will start by looking at the simplest form of linear

modulation, one-dimensional MPAM, which has no
quadrature component (𝑠𝑖,2 = 0).
For MPAM all of the information is encoded into the signal
amplitude 𝐴𝑖 . The transmitted signal over one symbol time is
given by
[ ]
𝑠𝑖 (𝑡) = Re 𝐴𝑖 𝑔(𝑡)𝑒𝑗2𝜋𝑓𝑐 𝑡 = 𝐴𝑖 𝑔(𝑡) cos(2𝜋𝑓𝑐 𝑡), 0 ≤ 𝑡 ≤ 𝑇𝑠 ≫ 1/𝑓𝑐 ,
(48)
where 𝐴𝑖 = (2𝑖 − 1 − 𝑀 )𝑑, 𝑖 = 1, 2, ..., 𝑀 defines the signal
constellation, parameterized by the distance 𝑑 which is
typically a function of the signal energy, and 𝑔(𝑡) is the
shaping pulse.
Pulse amplitude modulation (cont.)

In−Phase branch
s i1 Shaping s i1g(t)
Filter
g(t)
cos(2πf c t+ φ0)
cos(2πf c t+ φ0)
s(t)
π
2
−sin(2π f c t+ φ0)
s i2g(t)
s i2 Shaping
Filter
g(t)
Quadrature Branch
Figure 2: Amplitude and phase modulator.

The minimum distance between constellation points is

𝑑𝑚𝑖𝑛 = 𝑚𝑖𝑛𝑖,𝑗 ∣ 𝐴𝑖 − 𝐴𝑗 ∣= 2𝑑. The amplitude of the
transmitted signal takes on M different values, which implies
that each pulse conveys 𝑙𝑜𝑔2 𝑀 = 𝐾 bits per symbol time 𝑇𝑠 .
Over each symbol period the MPAM signal associated with
the 𝑖th constellation has energy
∫ 𝑇𝑠 ∫ 𝑇𝑠
𝐸𝑠𝑖 = 𝑠2𝑖 (𝑡)𝑑𝑡 = 𝐴2𝑖 𝑔2 (𝑡) cos2 (2𝜋𝑓𝑐 𝑡)𝑑𝑡 = 𝐴2𝑖 . (49)
0 0
It is noted that the energy is not the same for each signal
𝑠𝑖 (𝑡), 𝑖 = 1, ..., 𝑀 .


Assuming equally likely symbols, the average energy is
𝑀
1 ∑ 2
𝐸𝑠 = 𝐴𝑖 . (50)
𝑀
𝑖=1
In−Phase branch
Ts
r 1=s i1+n 1
g(T−t)
cos (2πf c t+ φ )
^
m=mi
r(t)=s i(t)+n(t)
Find i: x Z i
π/2
−sin (2π fc t+φ)
Ts
r 2=s i2+n 2
g(T−t)
Quadrature branch


The constellation mapping is usually done by Gray encoding,
where the messages associated with signal amplitudes that are
adjacent to each other differ by one bit value, as illustrated in
the below figure.
With this encoding method, if noise causes the demodulation
process to mistake one symbol for an adjacent one (the most
likely type of error), this results in only a single bit error in the
sequence of K bits. Gray codes can be designed for MPSK and
square MQAM constellations, but not rectangular MQAM.
M=4, K=2
00 01 11 10
2d
M=8, K=3
000 001 011 010 110 111 101 100
2d

√
An example: For 𝑔(𝑡) = 2/𝑇𝑠 , 0 ≤ 𝑡 < 𝑇𝑠 a rectangular
pulse shape, find the average energy of 4PAM modulation.
Solution: For 4PAM, the 𝐴𝑖 values are 𝐴𝑖 = {3𝑑, −𝑑, 𝑑, 3𝑑}.
Hence, the average is
𝑑2
(9 + 1 + 1 + 9) = 5𝑑2 .
𝐸𝑠 = (51)
4
The decision regions 𝑍𝑖 , 𝑖 = 1, ..., 𝑀 associated with the
pulse amplitude 𝐴𝑖 = (2𝑖 − 1 − 𝑀 )𝑑 for 𝑀 = 4 and 𝑀 = 8
as shown in the next figure. Mathematically, for any 𝑀 , these
decision regions are defined by
⎧

⎨(−∞, 𝐴𝑖 + 𝑑) 𝑖 = 1,
𝑍𝑖 = [𝐴𝑖 − 𝑑, 𝐴𝑖 + 𝑑) 2 ≤ 𝑖 ≤ 𝑀 − 1 (52)

⎩
[𝐴𝑖 − 𝑑, ∞) 𝑖=𝑀
One can see that MPAM has only a single basis function
𝜙1 (𝑡) = 𝑔(𝑡) cos(2𝜋𝑓𝑐 𝑡).
Thus, the coherent demodulator for MPAM reduces to the
demodulator shown in the next figure, where the
multi-threshold device maps 𝑥 to a decision region 𝑍𝑖 and
outputs the corresponding bit sequence
𝑚
ˆ = 𝑚𝑖 = {𝑏1 , ..., 𝑏𝐾 }.


Z1 Z2 Z3 Z4
A A A A
1 2 3 4
2d
Z1 Z2 Z3 Z4 Z5 Z6 Z7 Z8
A A A A A A A A
1 2 3 4 5 6 7 8
2d
Figure 5: Decision Regions for MPAM.
Multithreshold Device
(M−2)d
4d
Ts
s i (t)+n(t) 2d ^
m=m i=b1b2...bK
g (Ts−t) x
X 0
−2d
} Zi
−4d
cos(2πf c t)
−(M−2)d

Phase shift keying (MPSK)
For MPSK modulation, all of the information is encoded in

the phase of the transmitted signal.
Thus, the transmitted signal over one symbol time is given by
[ ]
𝑠𝑖 (𝑡) = Re 𝐴𝑔(𝑡)𝑒𝑗2𝜋(𝑖−1)/𝑀 𝑒𝑗2𝜋𝑓𝑐 𝑡 , 0 ≤ 𝑡 ≤ 𝑇𝑠 (53)
[ ]
2𝜋(𝑖 − 1)
= 𝐴𝑔(𝑡) cos 2𝜋𝑓𝑐 𝑡 +
𝑀
[ ]
2𝜋(𝑖 − 1)
= 𝐴𝑔(𝑡) cos cos(2𝜋𝑓𝑐 𝑡)
𝑀
[ ]
2𝜋(𝑖 − 1)
− 𝐴𝑔(𝑡) sin sin(2𝜋𝑓𝑐 𝑡).
𝑀

Phase shift keying (cont.)
The constellation
[ points
] or symbols (𝑠𝑖,1[ , 𝑠𝑖,2 ) are
] given by
2𝜋(𝑖−1) 2𝜋(𝑖−1)
𝑠𝑖,1 = 𝐴 cos 𝑀 and 𝑠𝑖,2 = 𝐴 sin 𝑀 for
𝑖 = 1, ..., 𝑀 . 𝜃𝑖 = 2𝜋(𝑖−1)
𝑀 , 𝑖 = 1, 2, ..., 𝑀 = 2𝐾 are the
different phases in the signal constellation points that convey
the information bits.
The minimum distance between constellation points is
𝑑𝑚𝑖𝑛 = 2𝐴 sin(𝜋/𝑀 ), where 𝐴 is typically a function of the
signal energy.
2PSK is often referred to as binary PSK or BPSK, while 4PSK
is often called quadrature phase shift keying (QPSK), and is
the same as MQAM with 𝑀 = 4 which is defined below.

All possible transmitted signals 𝑠𝑖 (𝑡) have equal energy:

∫ 𝑇𝑠
𝐸𝑠𝑖 = 𝑠2𝑖 (𝑡)𝑑𝑡 = 𝐴2 . (54)
0
√
Note that for 𝑔(𝑡) = 2/𝑇𝑠 , 0 ≤ 𝑡 ≤ 𝑇𝑠 , i.e., a rectangular
pulse, this signal has constant envelope, unlike the other
amplitude modulation techniques MPAM and MQAM.
However, rectangular pulses are spectrally-inefficient, and
more efficient pulse shapes make MPSK nonconstant
envelope.


Analogous to MPAM, MPSK constellation mapping is usually
done by Gray encoding, where the messages associated with
signal phases that are adjacent to each other differ by one bit
value, as illustrated in the below Figure.
With this encoding method, mistaking a symbol for an
adjacent one causes only a single bit error.
s i2 s i2
M=4, K=2 M=8, K=3
01 011
010 001
11 00 110 000
si1 si1
110 100
10 101
Figure 7: Gray Encoding for MPSK.

Phase shift keying: Decision region
The decision regions 𝑍𝑖 , 𝑖 = 1, ..., 𝑀 , associated with MPSK

for 𝑀 = 8 are shown in the next figure.
If we represent r = 𝑟𝑒𝑗𝜃 ∈ ℝ2 in polar coordinates then these
decision regions for any 𝑀 are defined by
{ }
𝑍𝑖 = 𝑟𝑒𝑗𝜃 : 2𝜋(𝑖 − .5)/𝑀 ≤ 𝜃 ≤ 2𝜋(𝑖 + .5)/𝑀 . (55)
For the special case of BPSK, the decision regions simplify to

𝑍1 = (r : r > 0) and 𝑍2 = (r : r ≤ 0).
Moreover BPSK has only a single basis function
𝜙1 (𝑡) = 𝑔(𝑡) cos(2𝜋𝑓𝑐 𝑡) and, since there is only a single bit
transmitted per symbol time 𝑇𝑠 , the bit duration 𝑇𝑏 = 𝑇𝑠 .

Phase shift keying: Decision region (cont.)
Thus, the coherent demodulator for BPSK reduces to the

demodulator shown in the next figure, where the threshold
device maps 𝑥 to the positive or negative half of the real line,
and outputs the corresponding bit value.
We have assumed in this figure that the message
corresponding to a bit value of 1, 𝑚1 = 1, is mapped to
constellation point 𝑠1 = 𝐴 and the message corresponding to
a bit value of 0, 𝑚2 = 0, is mapped to the constellation point
𝑠2 = −𝐴.

Phase shift keying: Decision region (cont.)

Z3
Z4
Z2
Z3 Z2
Z5
Z1 Z1
Z6
Z4
Z8
Z7
Figure 8: Decision Regions for MPSK.
Threshold Device
^
m=1
Tb ^
s i (t)+n(t)
X g (Tb−t) r
0
} Z :r>0
1
m=1 or 0
^
m=0 } Z :r<0
2
cos(2π f c t)
Figure 9: Coherent Demodulator for BPSK.

Quadrature amplitude modulation (MQAM)
For MQAM, the information bits are encoded in both the

amplitude and phase of the transmitted signal.
Thus, whereas both MPAM and MPSK have one degree of
freedom in which to encode the information bits (amplitude or
phase), MQAM has two degrees of freedom.
As a result, MQAM is more spectrally-efficient than MPAM
and MPSK, in that it can encode the most number of bits per
symbol for a given average energy.

Quadrature amplitude modulation (cont.)

The transmitted signal is given by
[ ]
𝑠𝑖 (𝑡) = Re 𝐴𝑖 𝑒𝑗𝜃𝑖 𝑔(𝑡)𝑒𝑗2𝜋𝑓𝑐 𝑡 (56)
= 𝐴𝑖 cos(𝜃𝑖 )𝑔(𝑡) cos(2𝜋𝑓𝑐 𝑡) − 𝐴𝑖 sin(𝜃𝑖 )𝑔(𝑡) sin(2𝜋𝑓𝑐 𝑡).
where 0 ≤ 𝑡 ≤ 𝑇𝑠 .
The energy in 𝑠𝑖 (𝑡) is
∫ 𝑇𝑠
𝐸𝑠𝑖 = 𝑠2𝑖 (𝑡)𝑑𝑡 = 𝐴2𝑖 . (57)
0
that is the same as for MPAM.
The distance between any pair of symbols in the signal
constellation is
√
𝑑𝑖,𝑘 = ∥s𝑖 − s𝑘 ∥ = (𝑠𝑖,1 − 𝑠𝑘,1 )2 + (𝑠𝑖,2 − 𝑠𝑘,2 )2 . (58)

Quadrature amplitude modulation (cont.)
For square signal constellations, where 𝑠𝑖,1 and 𝑠𝑖,2 take

values on (2𝑖 − 1 − 𝐿)𝑑, 𝑖 = 1, 2, ..., 𝐿 = 2𝑙 , the minimum
distance between signal points reduces to 𝑑𝑚𝑖𝑛 = 2𝑑, the
same as for MPAM.
In fact, MQAM with square constellations of size 𝐿2 is
equivalent to MPAM modulation with constellations of size 𝐿
on each of the in-phase and quadrature signal components.
Common square constellations are 4QAM and 16QAM, which
are shown in the below figure.

Quadrature amplitude modulation: constellation and

decision regions
Z1 Z2 Z3 Z4
Z5 Z6 Z7 Z8
Z9 Z 10 Z 11 Z 12
Z 13 Z 14 Z 15 Z 16
4−QAM 16−QAM
Figure 10: 4QAM and 16QAM

Figure 11: Decision Regions for
Constellations.
MQAM withM = 16.

Quadrature amplitude modulation: constellation and

decision regions
These square constellations have 𝑀 = 22𝑙 = 𝐿2 constellation

points, which are used to send 2𝑙 bits/symbol, or 𝑙 bits per
dimension.
It can be shown that the average power of a square signal
constellation with 𝑙 bits per dimension, 𝑆𝑙 , is proportional to
4𝑙 /3, and it follows that the average power for one more bit
per dimension 𝑆𝑙+1 ≈ 4𝑆𝑙 .
Thus, for square constellations it takes approximately 6 dB
more power to send an additional 1 bit/dimension or 2
bits/symbol while maintaining the same minimum distance
between constellation points.

Chapter 3: Physical-Layer Transmission Techniques: Section 3.1: Digital Modulations

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chapter 3: Physical-Layer Transmission Techniques: Section 3.1: Digital Modulations

Uploaded by

Copyright:

Available Formats

Digital modulation techniques

Signal Space Analysis

Chapter 3: Physical-layer transmission techniques

Section 3.1: Digital modulations

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 1

Outline of the lecture notes

2 Signal Space Analysis

3 Receiver Structure and Suﬃcient Statistics

4 Error Probability Analysis and the Union Bound

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 2

Advantages over analog modulation

The advances over the last several decades in hardware and

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 3

Advantages over analog modulation (cont.)

Digital transmissions consist of transferring information in the

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 4

Main considerations in digital modulation techniques

Digital modulation consists of mapping the information bits

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 5

Typical types of digital modulation techniques

Often the previous ones are conﬂicting requirements, and the

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 6

Typical types of digital modulation techniques (cont.)

Amplitude/phase modulation is also called linear modulation.

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 7

Typical types of digital modulation techniques (cont.)

In addition, amplitude and phase modulation techniques

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 8

Digital modulation encodes a bit stream of ﬁnite length into

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 9

By representing signals as projections onto a set of basis

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 10

Figure 1: Communication system model over AWGN channel (i.e., a

Consider a communication system model as shown in the

Transmitted signal (cont.)

There are 𝑀 = 2𝐾 possible sequences of 𝐾 bits and each bit

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 12

Transmitted signal (cont.)

Therefore, each message 𝑚𝑖 ∈ 𝕄 is mapped to a unique

When messages 𝑚𝑖 are sent sequentially, the transmitted

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 13

Transmitted and received signals

In the aforementioned model, the transmitted signal is sent

𝑟(𝑡) = 𝑠(𝑡) + 𝑛(𝑡). (3)

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 14

Given 𝑟(𝑡), the receiver must determine the best estimate of

over each time interval [0, 𝑇 ).

By representing the signals {𝑠𝑖 (𝑡), 𝑖 = 1, ..., 𝑀 } geometrically,

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 16

Basis function representation of signals

Speciﬁcally, using a Gram-Schmidt orthogonalization

Basis function representation of signals (cont.)

These basis functions have the following property

The basis set consists of the sine and cosine functions

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 18

Basis functions in linear passband modulation techniques

With these basis functions, one only obtain an approximation

Mobile communications-Chapter 3: Physical-layer transmissions Section 3.1: Digital modulations 19

Basis functions in linear passband modulation (cont.)

With these basis functions, one can have

Basis functions in linear passband modulation (cont.)

In practice, the basis set may include a baseband pulse-shaping

where the simplest pulse