EE321

Ajit Chaturvedi

Ajit Chaturvedi EE321

EE321 Course Introduction

Objectives: This is the second UG course on

Communications. Its focus is on fundamentals of digital
communication systems. The foundations of digital
transmitter and receivers will be covered. We will also briefly
cover the notion of capacity of digital communication systems
and how to achieve them using codes.
Contents Tentatively we propose to cover the following:
Modulation
Complex baseband representation
Spectral description of random processes
Modulation degrees of freedom
Linear modulation
Orthogonal and biorthogonal modulation
Differential modulation

2 / 39
Demodulation
Gaussian basics
Hypothesis testing basics
Signal space concepts
Optimal reception in AWGN
Performance analysis of ML reception
Bit-level demodulation
Elements of link budget analysis
Synchronization and noncoherent communication
Receiver design requirements
Parameter estimation basics
Parameter estimation for synchronization
Noncoherent communication
Performance of noncoherent communication
Channel equalization
The channel model
Receiver front end
Eye diagrams
Maximum likelihood sequence estimation

3 / 39
Channel Equalization Continued
Linear Equalization
Decision Feedback Equalization
Performance Analysis of MLSE
Numerical Comparison of Equalization Techniques
Information Theory
Capacity of AWGN Channel
Shannon Theory Basics
Some Capacity Computations
Coding Theory
Basics of Block and Convolutions Codes

4 / 39
Grading:
Quiz (best three out of four) - 20 %
MATLAB assignments (four) - 15 %
Mid-sem exam - 25 %
End-sem exam - 40 %
Make-up exam only for end-sem exam if approved by
Academic Office.
Book: Fundamentals of Digital Communication, Upamanyu
Madhow, Cambridge university press.
Instructors: Rohit Budhiraja ([email protected]) and Ajit K
Chaturvedi ([email protected])

5 / 39
Complex Baseband Representation
Communication systems more often than not have passband
(bandpass) channels.
Hence we must be able to transmit and receive passband
signals.
We will show that all the information in a real-valued
passband signal is contained in a corresponding
complex-valued baseband signal.
This equivalence between passband and complex baseband
has profound practical significance because complex envelope
can be represented accurately in discrete time using a much
smaller sampling rate than the corresponding passband signal.
Modern transceivers implement complicated signal processing
algorithms digitally on complex baseband signals, keeping the
analog processing of passband signals to a minimum.
Sophisticated algorithms can be developed in complex
baseband, independent of the physical frequency band that
will be used for communication.
6 / 39
In-phase and Quadrature Components

Any passband signal sp (t) can be expressed as:

√ √
sp (t) = 2sc (t) cos(2πfc t) − 2ss (t) sin(2πfc t)

Here, sc (t) and ss (t) represent the in-phase (I) and

quadrature (Q) components of the signal, respectively.
The waveforms sc (t) and ss (t) are also referred to as the
in-phase (or I) component and the quadrature (or Q)
component of the passband signal sp (t), respectively.
Example of a passband signal:
√ √
sp (t) = 2I[0,1] (t) cos(300πt)− 2(1−|t|)I[−1,1] (t) sin(300πt)

I component: sc (t) = I[0,1] (t).

Q component: ss (t) = (1 − |t|)I[−1,1] (t).

7 / 39
Complex Envelope

The complex envelope (or complex baseband representation)

of sp (t) is defined as:

s(t) = sc (t) + jss (t)

For the given example, the complex envelope is given by:

s(t) = I[0,1] (t) + j(1 − |t|)I[−1,1] (t)

8 / 39
Time Domain Relationship between Passband and
Complex Baseband

The relationship between passband and complex baseband

signals can be expressed as:
√
sp (t) = Re( 2s(t)e j2πfc t )

To verify this, plug in the equation s(t) = sc (t) + jss (t) and
Euler’s identity on the RHS of the above equation.

9 / 39
Envelope and Phase of a Passband Signal

The complex envelope s(t) can also be represented in polar

form, defining the envelope e(t) and phase θ(t) as:
 
−1 ss (t)
q
2 2
e(t) = |s(t)| = sc (t) + ss (t), θ(t) = tan
sc (t)
√
Substituting s(t) = e(t)e jθ(t) into sp (t) = Re( 2s(t)e j2πfc t ),
another representation of sp (t) is obtained as:

sp (t) = e(t) cos(2πfc t + θ(t))

These equations provide three equivalent representations of

the relationship between passband and complex baseband in
the time domain.

10 / 39
Example: Modeling Frequency or Phase Offsets in
Complex Baseband

Consider the passband signal sp (t), with complex baseband

representation s = sc + jss .
A phase-shifted version of the passband signal is:
√ √
s̃p (t) = 2sc (t) cos(2πfc t + θ(t)) − 2ss (t) sin(2πfc t + θ(t))

Here, θ(t) may vary slowly with time, e.g., θ(t) = 2πat + b
for carrier frequency offset a and phase offset b.

11 / 39
Complex Envelope after Phase Offset
To determine the complex envelope s̃(t) of s̃p (t) with respect
to fc , observe that s̃p (t) can be expressed as:
√
s̃p (t) = Re( 2s(t)e jθ(t) e j2πfc t )
Comparing the RHS with the standard representation, we can
see that
s̃(t) = s(t)e jθ(t)
Hence we can express s̃(t) as (time dependence is suppressed):
s̃c + j s̃s = (sc + jss )(cos θ + j sin θ)
This demonstrates how frequency or phase offsets affect the
complex envelope.
s̃c = sc cos θ − ss sin θ
s̃s = sc sin θ + ss cos θ
This is a typical illustration of the advantage of working in the
complex baseband.
12 / 39
Orthogonality of I and Q Channels

The passband
√ waveform corresponding to the I component i.e.
xc (t) = 2sc (t) cos 2πfc t and the passband waveform
corresponding
√ to the Q component i.e.
xs (t) = 2ss (t) sin 2πfc t are orthogonal. That is:

⟨xc , xs ⟩ = 0.

Using Parseval’s identity, the inner product can be expressed

in the frequency domain:
Z ∞
⟨xc , xs ⟩ = ⟨Xc (f ), Xs (f )⟩ = Xc (f )Xs∗ (f ) df .
−∞

13 / 39
Orthogonality Contd.

We need the expressions of Xc and Xs .

We can see that:
1
xc (t) = √ sc (t)e j2πfc t + sc (t)e −j2πfc t


2
1

↔ Xc (f ) = √ Sc (f − fc ) + Sc (f + fc ) ,
2
Similarly,
1
xs (t) = √ ss (t)e j2πfc t − ss (t)e −j2πfc t


2j
1

↔ Xs (f ) = √ Ss (f − fc ) − Ss (f + fc ) .
2j

14 / 39
Orthogonality Contd.

Substituting these into the inner product we get:

Z ∞
1

⟨xc , xs ⟩ = Sc (f − fc ) + Sc (f + fc )
2j −∞

∗ ∗


× Ss (f − fc ) − Ss (f + fc ) df

Cross product terms have no overap, hence their product is

zero.
This yields:
Z ∞ Z ∞ 
1
⟨xc , xs ⟩ = Sc (f )Ss∗ (f ) df − Sc (f )Ss∗ (f ) df =0
2j −∞ −∞

15 / 39
Example

Consider the signal:

u(t) = xc (t)xs (t) = 2sc (t)ss (t) cos 2πfc t sin 2πfc t

Is it a passband signal?
Around which frequency?
What is the value of:
Z ∞
u(t) dt =?
−∞

It is equal to U(0).
Since it is a passband signal,
Z ∞
u(t) dt = U(0) = 0.
−∞

16 / 39
Passband and Complex Baseband Inner Products

For real passband signals up and vp with complex envelopes u

and v , we will show that the inner product satisfies:

⟨up , vp ⟩ = ⟨uc , vc ⟩ + ⟨us , vs ⟩ = Re(⟨u, v ⟩).

To show√the first equality, substitute

√ the standard form
sp (t) = 2sc (t) cos(2πfc t) − 2ss (t) sin(2πfc t) for up and vp
and use the orthogonality of I and Q channels.
For the second equality, consider the complex inner product
⟨u, v ⟩ as:
Z ∞
⟨u, v ⟩ = (uc (t) + jus (t))(vc (t) − jvs (t)) dt.
−∞

17 / 39
Energy of Complex Envelope
Simplifying:


⟨u, v ⟩ = ⟨uc , vc ⟩ + ⟨us , vs ⟩ + j − ⟨uc , vs ⟩ + ⟨us , vc ⟩ .

The real part gives the desired term.

We now show that the energy of the complex envelope is
equal to that of the corresponding passband signal.
In terms of inner products, the energy of the passpand signal
can be obtained by letting up = vp = sp as:

⟨sp , sp ⟩ = ||sp ||2 .

Using the above relation between inner products, we can write

⟨sp , sp ⟩ = Re(⟨s, s⟩) = Re(||s||2 ) = ||s||2 .

Thus,
||sp ||2 = ||s||2
18 / 39
Frequency Domain Relationships
Define:
√ √
v (t) = 2S(t)e j2πfc t ↔ V (f ) = 2S(f − fc ).

From the time-domain relationship between sp (t) and s(t):

v (t) + v ∗ (t)
sp (t) = Re(v (t)) =
2
V (f ) + V ∗ (−f ) S(f − fc ) + S ∗ (−f − fc )
↔ Sp (f ) = = √ .
2 2
Observe that the symmetry condition Sp (f ) = Sp∗ (−f ) holds.
This implies that sp (t) is real-valued.
If S(f ) has energy concentrated in the baseband, then the
energy of V (f ) is around fc and V ∗ (−f ) is around −fc .
Thus, as expected, Sp (f ) is indeed passband,
19 / 39
Frequency Domain Relationships Contd.

In the figure on the next slide, we show the relationship

between the spectra of Sp (f ), V (f ) and S(f ).
All of these spectra are, in general, complex-valued.
Starting from an arbitrary baseband signal S(f ) at the bottom
of the figure, we first construct V (f ) in the middle of the
figure.
Then use V (f ) to construct a conjugate symmteric passband
signal Sp (f ) at the top of the figure.
Please note that S(f ) does not, in general, obey conjugate
symmetry.
This is why the baseband signal s(t) is complex-valued.

20 / 39
Frequency Domain Relationships Contd.

21 / 39
General Applicability of Complex Baseband Representation

We have seen that given a complex baseband signal (or a pair

of real signals) we can generate a real-valued passband signal.
But do these representation apply to any real-valued passband
signal?
Consider a real-valued passband signal sp (t) implying that
Sp (f ) = Sp∗ (−f ).
Thus, knowing the values of Sp (f ) for positive frequencies is
enough to characterize the values for negative frequencies.
Define a scaled version of Sp (f ) for positive frequencies:
(
+ 2Sp (f ), f > 0,
V (f ) = 2Sp (f ) =
0, else.

We can see that this definition is in ∗conformity with the earlier

derived relationship Sp (f ) = V (f )+V
2
(−f )
.
22 / 39
V (f ) and Complex Envelope

Note that the energy of V (f ) is concentrated around fc .

Now, let us define the complex envelope
√ of Sp by inverting the
earlier stated relationship V (f ) = 2S(f − fc ).
Thus, S(f ) is related to V (f ) as:

1
S(f ) = √ V (f + fc ).
2
Since V (f ) is concentrated around +fc , Sf , which is obtained
by translating it to the left by fc , is at baseband.
Thus, starting from an arbitrary passband signal Sp (f ), we
have obtained a baseband signal S(f ) that satisfies the earlier
stated relationships.
We need to go through the same figure but this time from top
to bottom.
23 / 39
Upconversion and Downconversion

24 / 39
Downconversion
To downconvert from passband to baseband:
√
2sp (t) cos(2πfc t) = 2sc (t) cos2 (2πfc t)
−2ss (t) sin(2πfc t) cos(2πfc t)
= sc (t) + sc (t) cos(4πfc t) − ss (t) sin(4πfc t).

The first term is the I component, while the second and third
terms are passband signals at 2fc .
Similarly, we can obtain the Q component.
√
− 2sp (t) sin(2πfc t) = 2ss (t) sin2 (2πfc t)
−2sc (t) sin(2πfc t) cos(2πfc t)
= ss (t) − ss (t) cos(4πfc t) − sc (t) sin(4πfc t).

The first term is the Q component, while the second and third
terms are passband signals at 2fc .
25 / 39
Information Resides in Complex Baseband

The complex baseband representation removes the rapid but

predictable phase variation due to the fixed reference
frequency fc .
This allows us to focus on the slower amplitude and phase
variations induced by baseband modulation.
Since the phase variation due to fc is predictable and does not
convey any information, all the information in a passband
signal is contained in its complex envelope.

26 / 39
Example: Linear Modulation

Suppose information is encoded into a complex number:

b = bc + jbs = re jθ ,

where bc and bs are the real and imaginary parts, and r ≥ 0,

θ are real-valued and correspond to its polar form.
In practice, we encode information in a stream of complex
symbols {b[n]}.
Let p(t) be a real-valued baseband pulse.
The linearly modulated complex baseband waveform:

s(t) = bp(t).

can be used to convey the information in b over a passband

channel by upconverting to an arbitrary frequency fc .

27 / 39
Linear Modulation Contd.

The corresponding passband signal is given by:

√ 
sp (t) = Re 2s(t)e j2πfc t
√

= 2 bc p(t) cos(2πfc t) − bs p(t) sin(2πfc t)
√
= 2r cos(2πfc t + θ).

This can be viewed as amplitude modulation by bc and bs of

the I and Q components, respectively.
Alternately, it can be viewed as amplitude and phase
modulation by r and θ of the passband waveform.
In practice, a stream of {b[n]} linearly modulates time shifts
of the basic waveform p(t) i.e. p(t − T ), p(t − 2T ) . . . and
generates
P the corresponding complex baseband waveform
n b[n]p(t − nT ).

28 / 39
Complex Baseband Equivalent of Passband Filtering

There is a need for passband filtering in several situations

such as:
At the transmitter (e.g., to conform to spectral masks).
At the receiver (e.g., to filter out noise).
Broad class of channels modeled as linear filters.
Suppose a passband signal sp (t) is passed through a passband
filter with impulse response hp (t):

yp (t) = (sp ∗ hp )(t).

Let y , s, and h denote the complex envelopes for yp , sp , and

hp , respectively.

29 / 39
Frequency Domain Filtering Contd.

In frequency domain, the filtering operation can be

represented as:
Yp (f ) = Sp (f )Hp (f ).
Since passband signals are completely characterized by their
positive frequencies, we can focus only on the positive
frequency components:

Y+ (f ) = Yp (f ) f >0
, S+ (f ) = Sp (f ) f >0
, H+ (f ) = Hp (f ) f >0
.

These satisfy:
Y+ (f ) = S+ (f )H+ (f ).

30 / 39
Filtering in Complex Baseband

The complex envelope in the frequency domain is given by:

√
Y (f ) = 2Y (f + fc )
√ +
= 2S+ (f + fc )H+ (f + fc )
1
= √ S(f )H(f ).
2
This shows that passband filtering is equivalent to complex
baseband filtering.

31 / 39
Frequency Domain Filtering

32 / 39
Time Domain Passband Filtering in Complex Baseband

Use the complex baseband representation of the impulse

response of the desired passband filter.
The filtered output is given by:
1
y (t) = √ (s ∗ h)(t).
2
Writing out the real and imaginary parts:
1 1
yc = √ (sc ∗ hc − ss ∗ hs ), ys = √ (sc ∗ hs + ss ∗ hc ).
2 2
This required four real baseband filters.

33 / 39
Four Real Baseband Filters

34 / 39
Transceiver Implementations

In modern passband transceivers, most of the intelligence is

moved to the baseband and is implemented using Digital
Signal Processors.
The digital algorithms used are independent of the band over
which communications eventually occurs.
They can be implemented using low-cost technologies like:
Very Large Scale Integrated (VLSI) circuits.
Field Programmable Gate Arrays (FPGA).
General-purpose DSP engines.

35 / 39
Transceiver Implementations

On the other hand, analog components like local oscillators,

power amplifiers, and low-noise amplifiers need to be
optimized for the band of interest and are often bulky.
Baseband processing allows use of relatively sloppy low-cost
analog filters in the passband.
For example, transmitted passband waveforms can conform to
spectral masks (mandated by regulatory agencies) using
baseband DSP filtering.
Similarly, analog phase-locked loops for carrier synchronization
have been eliminated in many modern receivers.
Instead, the receiver employs a fixed analog local oscillator for
down conversion followed by a digital phase locked loop
implemented in baseband.

36 / 39
37 / 39
Example: Handling Carrier Frequency and Phase Offsets

A communication receiver uses a local oscillator with a fixed

carrier frequency fc to demodulate an incoming passband
signal:
√
yp (t) = 2[yc (t) cos(2π(fc +a)t+b)−ys (t) sin(2π(fc +a)t+b)],

where a and b are carrier frequency and phase offsets,

respectively.
Denote the I and Q components at the output of the
downconverter as ỹc , ỹs , and the corresponding complex
envelope as:
ỹ = ỹc + j ỹs .
We wish to recover yc , ys , and hence y = yc + jys .

38 / 39
Complex Envelope Relationship
We have seen that the relationship between y and ỹ is:

ỹ (t) = y (t)e j(2πat+b) .

This relation can be inverted to recover y from ỹ :

y (t) = ỹ (t)e −j(2πat+b) .

Thus,


yc (t)+jys (t) = (ỹc (t)+j ỹs (t)) cos(2πat+b)−j sin(2πat+b) .

Equating real and imaginary parts, we obtain:

yc = ỹc cos(2πat + b) + ỹs sin(2πat + b),

ys = −ỹc sin(2πat + b) + ỹs cos(2πat + b).

These computations allow the receiver to recover the actual
yc and ys which were transmitted.
39 / 39