Introduction To Analog and Digital Communications 2.2

Introduction to Analog and Digital
Communications
By:
Phil Schniter
Introduction to Analog and Digital
Communications
By:
Phil Schniter
Online:
< http://cnx.org/content/col10968/1.2/ >
CONNEXIONS
Rice University, Houston, Texas

This selection and arrangement of content as a collection is copyrighted by Phil Schniter. It is licensed under the
Creative Commons Attribution 3.0 license (http://creativecommons.org/licenses/by/3.0/).
Collection structure revised: September 14, 2009
PDF generated: October 28, 2012
For copyright and attribution information for the modules contained in this collection, see p. 61.
Table of Contents
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Preview of Comm System Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4 The Communications Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
5 Analog Communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
6 Random Signals and Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
7 Complex-Baseband Equivalent Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
8 Digital Communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
9 Discrete-Time Implementation of Digital Communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
10 Error Analysis of Digital Communications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Attributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
iv
Available for free at Connexions <http://cnx.org/content/col10968/1.2>

Chapter 1
Introduction
1
Goal: Transmit a message from one location to another.

When message is a...
continuous waveform, use analog comm (e.g., FM radio),
sequence of numbers, use digital comm (e.g., mp3 le),
but keep in mind that the sequence of numbers might represent a sampling of a continuous waveform (as in
the case of digital audio).
Typical communication media:
twisted pair wire (e.g., telephoneA )
coaxial cable (e.g., TVA,D , dataD )
ber optic cable (e.g., ethernetD )
EM waves (e.g., cellular phonesA,D , WiFiD , TVA,D )
water waves (e.g., underwater networkA,D )
power linesA,D
compact discD
hard driveD
magnetic tapeA,D
Table 1.1
where A = analog and D = digital.

Note that, whether the message signal is discrete-time or continuous-time, the transmitted signal is
continuous-time!
1 This content is available online at <http://cnx.org/content/m31807/1.4/>.
1
2 CHAPTER 1. INTRODUCTION
1.1 Analog Communication
Figure 1.1
• Perfect recovery is impossible in the presence of noise!
1.2 Digital Communication
Figure 1.2
• A digital message is converted to an analog message coding and pulse-shaping, and then transmitted
using analog modulation. To recover the message, the received signal is demodulated, sampled, and
digitally processed.
• Perfect recovery is possible even in the presence of noise!

Chapter 2
Preview of Comm System Components

1
The Modulator:
• Translates baseband analog message to passband:
Figure 2.1
where fc is the carrier frequency.

• There are two principal motivations for doing this:
a. Often we want to communicate several messages simultaneously (e.g., TV, radio, voice). It's
dicult or impossible to do this if they overlap in frequency!
b. Wireless EM transmission/reception is much easier at higher frequencies, since need antenna
λ c
length > 10 . (λ = fc is wavelength and c =3e8 m/s speed of light.)
system transmission band λ/10

VHF (TV) 30300 MHz 10.1 m
UHF (TV) 0.33 GHz 101 cm
cellular 824960 MHz 3 cm
WiFi 2.4 GHz 1 cm
Table 2.1
Notice that practical antenna length determines where dierent signal types can be transmitted.
The Coder/Mapper:
• Coder transforms sequence of message bits into an error-resiliant sequence of coded bits.
3
4 CHAPTER 2. PREVIEW OF COMM SYSTEM COMPONENTS
• Mapper transforms coded bits into discrete symbols. Ex: If the symbol alphabet is {−3, −1, 1, 3}
and the symbol mapping is
bits symbol
00 3
01 -1
10 1
11 -3
Table 2.2
then ASCII text would be transmitted via
letter ASCII code symbol sequence
a 01 10 00 01 -1 1 -3 -1
b 01 10 00 10 -1 1 -3 1
c 01 10 00 11 -1 1 -3 3
d 01 10 01 00 -1 1 -1 -3
. . .
. . .
. . .
Table 2.3
Pulse Shaper:
• Converts symbol sequence into a continuous waveform.

• In linear modulation schemes, the time-n symbol a [n] scales a nT -delayed version of pulse g (t):
P
m (t) = n a [n] g (t − nT ) "baseband message"
(2.1)
T = "symbol period"
Ex: Say symbol sequence is [1, 3, −1, 1, 3]. Then
Figure 2.2

Chapter 3
Preliminaries
1
3.1 Fourier Transform (FT)
Denition:
R∞
W (f ) = −∞
w (t) e−j2πf t dt = F{w (t)}
R∞ (3.1)
w (t) = −∞
W (f ) ej2πf t df = F −1 {W (f )}.
Properties:
• Linearity: F{c1 w1 (t) + c2 w2 (t)} = c1 W1 (f ) + c2 W2 (f ).

conjugate symmetricW (f )
• Real-valued w (t) ⇒ { .
|W (f ) |symmetric aroundf =0
3.2 Bandwidth
Figure 3.1
3.3 Dirac Delta (or continuous impulse) δ (·)

An innitely tall and thin waveform with unit area:
5
6 CHAPTER 3. PRELIMINARIES
Figure 3.2
that's often used to kick a system and see how it responds.
Key properties:
R∞
1. Sifting:
−∞
w (t) δ (t − q) dt = w (q).
2. Time-domain impulse δ (t) has a at spectrum:
Z ∞
F{δ (t)} = δ (t) e−j2πf t dt = 1 (for all f ). (3.2)
−∞
3. Freq-domain impulse δ (f ) corresponds to a DC waveform:
Z ∞
F −1 {δ (f )} = δ (f ) ej2πf t df = 1 (for all t). (3.3)
−∞
3.4 Frequency-Domain Representation of Sinusoids
Notice from the sifting property that
Z ∞
−1
F {δ (f − fo )} = δ (f − fo ) ej2πf t df = ej2πfo t . (3.4)
−∞
Thus, Euler's equations
cos (2πfo t) = 1 j2πfo t

2e + 12 e−j2πfo t
(3.5)
1 j2πfo t 1 −j2πfo t
sin (2πfo t) = 2j e − 2j e
and the Fourier transform pair ej2πfo t ↔ δ (f − fo ) imply that
1
F{cos (2πfo t)} = 2 δ (f − fo ) + 12 δ (f + fo )
(3.6)
1 1
F{sin (2πfo t)} = 2j δ (f − fo ) − 2j δ (f + fo ) .
Often we draw this as

7
Figure 3.3
3.5 Frequency Domain via MATLAB
Fourier transform requires evaluation of an integral. What do we do if we can't dene/solve the integral?
1
1. Generate (rate-
Ts ) sampled signal in MATLAB.
2. Plot magnitude of Discrete Fourier Transform (DFT) using plottf.m (from course webpage).
Square-wave example:
(a) (b)
Figure 3.4

Noise-wave example
(a) (b)
Figure 3.5
h
Notice that plottf.m only plots frequencies f ∈ − 2T1 s , 2T1 s .
3.6 Linear Time-Invariant (LTI) Systems
An LTI system can be described by either its impulse response h (t) or its frequency response H (f ) =
F{h (t)}.
Figure 3.6
Input/output relationships:
• Time-domain: Convolution with impulse response h (t)

9
Figure 3.7
• Freq-domain: Multiplication with freq response H (f )
Figure 3.8
3.7 Linear Filtering
Freq-domain illustration of LPF, BPF, and HPF:
Figure 3.9
3.8 Lowpass Filters

sin(πx)
Ideal non-causal LPF (using sinc (x) := πx ):

Figure 3.10
Ideal LPF with group-delay to :
Figure 3.11
A causal linear-phase LPF with group-delay to :
Figure 3.12
but MATLAB can give better causal linear-phase LPFs...

1
In MATLAB, generate
Ts -sampled LPF impulse response via
h = firls(Lf, [0,fp,fs,1], [G,G,0,0])/Ts;
where...

11
Figure 3.13
The commands firpm and fir2 have the same interface, but yield slightly dierent results (often worse
for our apps).
In MATLAB, perform ltering on
1
Ts -sampled signal x via
y = Ts*filter(h,1,x); or y = Ts*conv(h,x);
(a) (b)
Figure 3.14
important: The routines firls,firpm,fir2 generate causal linear-phase lters with group delay
= Lf
2 samples. Thus, the ltered output y will be delayed by Lf
2 samples relative to x .


Chapter 4
The Communications Channel

1
The eects of signal propagation are usually modeled as:
Figure 4.1
where
H (f ): linear ltering due to multipath propagation
w (t): additive noise/interference.
Noise/interference sources include:
• electronic circuitry (thermal/shot noise or quantization noise); usually broadband,

• other comm systems (multi-access interference or co-channel interference); broadband or narrow-
band.
SNR can be improved with appropriate ltering at receiver:
Es
signal energy
• signal-to-noise ratio (SNR) = En
noise energy
=
• roughly speaking, average SNR can be improved by ltering out the frequencies dominated by noise:
Figure 4.2
13
14 CHAPTER 4. THE COMMUNICATIONS CHANNEL
4.1 Filtering due to Multipath Propagation
The signal may propagate along paths with dierent lengths:
Figure 4.3
Since dierent path lengths imply dierent path delays:
x (t) = c1 s (t − τ1 ) + c2 s (t − τ2 ) + · · · + cN s (t − τN ) , (4.1)
which can be written as x (t) = s (t) ∗ h (t) for
h (τ ) = c1 δ (τ − τ1 ) + c2 δ (τ − τ2 ) + · · · + cN δ (τ − τN ) . (4.2)
The result is an |H (f ) | that varies with f, implying frequency-dependent signal attenuation:
(a) (b)
Figure 4.4

Chapter 5
Analog Communication
1
1. Amplitude modulation (AM)

2. Quadrature amplitude modulation (QAM)
3. Vestigial sideband modulation (VSB)
4. Frequency modulation (FM)
5.1 AM with suppressed carrier
AM of real-valued message m (t) (e.g., music) is
Figure 5.1
1
+ e−j2πfc t
j2πf t
Euler's cos (2πfc t) = 2 e c
then implies
R∞
S (f ) = −∞
m (t) cos (2πfc t) e−j2πf t dt
1 ∞ −j2π(f −fc )t
R∞
dt + 12 −∞ m (t) e−j2π(f +fc )t dt
R
= m (t) e (5.1)
2 −∞
1
= 2M (f − fc ) + 21 M (f + fc ) .
15
16 CHAPTER 5. ANALOG COMMUNICATION
Figure 5.2
Because m (t) ∈ R, know |M (f ) | symmetric around f = 0, implying the AM transmitted spectrum

below fc is redundant! This motivates the QAM and VSB modulation schemes...
With fc known, AM demodulation can be accomplished by:
Figure 5.3
For a trivial noiseless channel, we have r (t) = s (t), so that
v (t) = LPF{s (t) · 2cos (2πfc t)}

= LPF{m (t) · 2cos2 (2πfc t)}
| {z }
1+cos(2π·2fc t) (5.2)
= LPF{m (t) + m (t) cos (2π · 2fc t)}

= m (t) ,
assuming a LPF with passband cuto Bp ≥ W Hz and stopband cuto Bs ≤ 2fc − W Hz:
Figure 5.4
Note that we've assumed perfectly synchronized oscillators!

17
When the receiver oscillator has {freq,phase} oset {γ, φ}:
v (t) = LPF{m (t) cos (2πfc t) · 2cos (2π (fc + γ) t + φ)}

| {z }
cos(2πγt+φ) + cos(2π(2fc +γ)t+φ)
(5.3)
= m (t) cos (2πγt + φ) .
| {z }
time-varying attenuation!
νfc
note: a freq oset of λ = c Hz can occur when there is relative velocity of ν m/s between
transmitter and receiver.
5.2 AM with pilot tone or carrier tone
It's common to include a pilot/carrier tone with frequency| fc :
s (t) = m (t) cos (2πfc t) + Acos (2πfc t)

| {z }
pilot/carrier tone
(5.4)
= [m (t) + A] cos (2πfc t)
1
S (f ) = 2 [M (f −fc ) +M (f +fc ) +Aδ (f −fc ) +Aδ (f +fc )]
Figure 5.5
Advantage: aids receiver with carrier synchronization.

Disadvantage: consumes transmission power.
While modern systems choose A max|m (t) |, many older systems use A > max|m (t) |, known as
large carrier AM, allowing reception based on envelope detection:
π
v (t) = 2 LPF{ |r (t) | } −A
(5.5)
≈ m (t) (with a trivial channel)
where |·| can be easily implemented using a diode.

π
The gain
2 above makes up for the loss incurred when LPFing the rectied signal:

Figure 5.6
5.3 Quadrature Amplitude Modulation (QAM)
QAM is motivated by unwanted redundancy in the AM spectrum, which was symmetric around fc .
QAM sends two real-valued signals {mI (t) , mQ (t)} simultaneously, resulting in a non-symmetric spec-
trum.
Figure 5.7
QAM demodulation is accomplished by:

19
Figure 5.8
where the LPF specs are the same as in AM, i.e., passband edge Bp ≥ W Hz and stopband edge
Bs ≤ 2fc − W Hz.
For a trivial channel, we have r (t) = s (t), so that
vI (t) = LPF{r (t) · 2cos (2πfc t)}

= LPF{mI (t) 2cos2 (2πfc t)
| {z }
1+cos(4πfc t)
−mQ (t) 2sin (2πfc t) cos (2πfc t)}

| {z }
sin(4πfc t)
= mI (t)
(5.6)
vQ (t) = LPF{−r (t) · 2sin (2πfc t)}
= LPF{−mI (t) 2cos (2πfc t) sin (2πfc t)
| {z }
sin(4πfc t)
+mQ (t) 2sin2 (2πfc t)}

| {z }
1−cos(4πfc t)
= mQ (t) ,
assuming synchronized oscillators.
When the oscillators are not synchronized, one gets coupling between the I&Q components as well as
attenuation of each. Writing the I&Q signals in the complex-baseband form
m̃ (t) = mI (t) + jmQ (t)

(5.7)
ṽ (t) = vI (t) + jvQ (t)
yields a much simpler description of QAM:

Figure 5.9
F
note: Re{u (t)} = 1
2 [u (t) + u∗ (t)] ↔ 1
2 [U (f ) + U ∗ (−f )].
We now verify the complex-baseband model for modulation:
Re{m̃ (t) ej2πfc t }

=
Re{(mI (t) + jmQ (t)) (cos (2πfc t) + jsin (2πfc t))} (5.8)
=
mI (t) cos (2πfc t) − mQ (t) sin (2πfc t) = s (t) ,
as well as for demodulation (assuming r (t) = s (t)):
ṽ (t) = LPF{s (t) · 2e−j2πfc t }

= LPF{(mI(t) cos (2πfc t) − mQ (t) sin (2πfc t)) · 2e−j2πfc t }
+ e−j2πfc t e−j2πfc t
j2πfc t

= LPF{mI (t) e
(5.9)
−mQ (t) je−j2πfc t − jej2πfc t e−j2πfc t }

−j4πfc t
− mQ (t) je−j4πfc t − j }

= LPF{mI (t) 1 + e
= mI (t) + jmQ (t) .

The convenience of complex-baseband results in widespread use of complex-valued signals for comm systems!
note: To get the complex baseband formulation for AM, we simply set mQ (t) = 0 and mI (t) =
m (t).
5.4 Vestigial Sideband Modulation (VSB)
VSB is another way to restore regain the spectral eciency lost in AM. It's used to transmit North American
terrestrial TV, both analog (NTSC) and digital (ATSC) formats.

21
Like AM, it can operate with or without a carrier tone.

Basically, VSB suppresses most of the redundant AM spectrum by ltering it:
Figure 5.10
The passband VSB lter is a BPF C (f ) where
C (f − fc ) + C (f + fc ) = 2 for |f | ≤ W, (5.10)
which implies its inside rollo is symmetric around f = fc :
Figure 5.11

For VSB modulation, we have
s (t) = m (t) cos (2πfc t) ∗ c (t)

(5.11)
1
S (f ) = 2 [M (f + fc ) + M (f − fc )] C (f ) .
It turns out that VSB demod is identical to AM demod:
v (t) = LPF{r (t) · 2cos (2πfc t)}

= LPF{s (t) · 2cos (2πfc t)} (trivial channel)
V (f ) = LPF{S (f − fc ) + S (f + fc )}
1
= 2 LPF{[M (f ) + M (f − 2fc )] C (f − fc )
(5.12)
+ [M (f + 2fc ) + M (f )] C (f + fc )}
1
= M (f ) [C (f − fc ) + C (f + fc )]
|2 {z }
=1 for f ∈[−W,W ]
= M (f ) .
We note that the property
1
F{cos (2πfc t) c (t)} = 2 [C (f − fc ) + C (f + fc )] (5.13)
may be convenient, e.g., for testing whether a given lter c (t) satises the passband VSB criterion. VSB
∼
ltering can also be implemented at baseband using a complex-valued lter response c (t) which satises
C̃ (f ) + C̃ ∗ (−f ) = 2 for |f | ≤ W, (5.14)
generating the complex-baseband message signal m̃ (t). The message can be recovered by simplying ignoring
the imaginary part of the complex-baseband output ṽ (t).
Figure 5.12
Motivation: ltering at baseband is usually much cheaper than ltering at passband.

23
5.5 Frequency Modulation (FM)
While AM modulated the carrier amplitude, FM modulates the carrier frequency.
(a) (b)
Figure 5.13
In particular, FM modulates the real-valued message m (t) via
 
 Z t 
s (t) = cos 2πfc t + 2πkf m (τ ) dτ . (5.15)
 
 0 
| {z }
ϕ(t)"instantaneous modulation phase"
where kf is called the frequency-sensitivity factor. Since the instantaneous modulation frequency
dϕ (t)
= 2πkf m (t) (5.16)
dt
is a scaled version of the message m (t), it is tting to call this scheme frequency modulation.
Using the peak frequency deviation ∆f = kf max|m (t) |, the modulation index D is dened as
∆f
D = (5.17)
W ← denominator is the single-sided BW of m (t) .
Increasing D decreases spectral eciency but increases robustness to noise/interference.
D1 : narrowband FM ,

(5.18)
D1 : wideband FM .
Carson's Rule approximates the FM passband signal-BW as
BW99 ≈ 2 (∆f + W ) = 2 (D + 1) W. (5.19)

Example: Mono FM radio:
• Message signal ltered to freq interval [30,15k] Hz.

75
• FCC limits ∆f ≤ 75 kHz (channels 200 kHz apart). ⇒D = 15 =5
FM stereo uses smaller D due to message spectrum:
Figure 5.14
There are various FM demodulators, but the discriminator is one of the best known. Recalling that
d dϕ (t)
cos (ϕ (t)) = − sin (ϕ (t)) , (5.20)
dt dt
we see that
Rt
d d
dt s (t) = dt cos2πfc t + 2πkf 0 m (τ ) dτ
Rt (5.21)
= − [2πfc + 2πkf m (t)] sin 2πfc t + 2π 0 m (τ ) dτ
is a form of large-carrier AM (assuming fc > kf m (t)), which can be demodulated using an envelope detector
as follows:
Figure 5.15

Chapter 6
Random Signals and Noise

1
When designing a comm system, it is impossible to know exactly what the signal and noise waveforms will
be. But usually we know average characteristics such as energy distribution across frequency. It is exactly
these statistics that are most often used for comm system design.
We will consider random waveforms known as zero-mean wide-sense stationary random processes. Such
an x (t) is completely completely characterized by its power spectral density Sx (f ), or average signal power
versus frequency f. The technical denition of PSD accounts for the fact that power is dened as energy per
unit time:
Z 2
1 T
Sx (f ) := lim E{ x (t) ej2πf t dt }, (6.1)

T →∞ 2T −T
Above, E{·} denotes expectation, i.e., statistical average.

A common random waveform is white noise. Saying that w (t) is white is equivalent to saying that its
PSD is at:
Sw (f ) = No for all f. (6.2)
Broadband noise is often modelled as white random noise. A fundamentally important question is: What
does ltering do to the power spectrum of a signal? The answer comes with the aid of the autocor-
relation function Rx (τ ), dened as
Rx (τ ) := E{x (t) x∗ (t − τ )} (6.3)
and having the important property
F
Rx (τ ) ↔ Sx (f ) . (6.4)
Note that, for white noise w (t) with PSD No , we have
Sw (f ) = No ⇒ Rw (τ ) = No δ (τ )
(6.5)
⇒ E{w (t) w∗ (t − τ )} = No δ (τ ) .
Here we see that w (t1 ) and w (t2 ) |t2 6= t1 are "uncorrelated" because E{w (t1 ) w∗ (t2 )} = 0, and that w (t)
2
has average energy Ew = E{|w (t) | } = ∞, which may be surprising. We could have found the same via
R∞
Ew = −∞ Sw (f ) df .
25
26 CHAPTER 6. RANDOM SIGNALS AND NOISE
R ∞ Now say that white noise w (t) is ltered with non-random h (t), yielding output y (t) =
−∞
w (q) h (t − q) dq . We can nd Sy (f ) by rst nding Ry (τ ) and then taking the FT. Recall:
Ry (τ ) = E{y (t) y ∗ (t − τ )}. (6.6)
Plugging in the expression for y (t), we nd
R∞ R∞
Ry (τ ) = E{ −∞ w (q1 ) h (t − q1 ) dq1 −∞ w∗ (q2 ) h∗ (t − τ − q2 ) dq2 }
R∞ R∞ (6.7)
= E{ −∞ −∞ w (q1 ) w∗ (q2 ) h (t − q1 ) h∗ (t − τ − q2 ) dq1 dq2 }.
Because averaging is a linear operation, and because the h (·) terms are xed (non-random) quantities, we
can write
R∞ R∞
Ry (τ ) = −∞ −∞
E{w (q1 ) w∗ (q2 )} h (t − q1 ) h∗ (t − τ − q2 ) dq1 dq2 . (6.8)
Since w (t) is white,
E{w (q1 ) w∗ (q2 )} = E{w (q1 ) w∗ (q1 − (q1 − q2 ))} = No δ (q1 − q2 ) , (6.9)
which allows use of the sifting property on the inner integral:
R∞ R∞
Ry (τ ) = −∞ −∞
No δ (q1 − q2 ) h (t − q1 ) h∗ (t − τ − q2 ) dq1 dq2
R∞
= No −∞ h (t − q2 ) h∗ (t − τ − q2 ) dq2
R −∞ (6.10)
= No ∞
h (q) h∗ (q − τ ) (−dq) using q := t − q2
R∞
= No −∞ h (q) h∗ (q − τ ) dq.
To summarize, the autocorrelation of ltered white noise is
Z ∞
Ry (τ ) = No h (q) h∗ (q − τ ) dq. (6.11)
−∞
Having Ry (τ ), the nal step is nding Sy (f ) = F{Ry (τ )}:

R∞
Sy (f ) = Ry (τ ) e−j2πf τ dτ
−∞
R∞ h R∞ i
= −∞
No −∞ h (q) h∗ (q − τ ) dq e−j2πf τ dτ
R∞ hR i
∞
= No −∞ h (q) e−j2πf q −∞ h∗ (q − τ ) ej2πf (q−τ ) dτ dq
R∞ hR i (6.12)
∞
= No −∞ h (q) e−j2πf q −∞ h∗ (t) ej2πf t dt dq
R∞ hR i∗
∞
= No −∞ h (q) e−j2πf q dq −∞ h (t) e−j2πf t dt
2
= No H (f ) H ∗ (f ) = No |H (f ) |
To summarize, the power spectrum of ltered white noise is
2
Sy (f ) = No |H (f ) | . (6.13)
More generally, the power spectrum of a ltered random process x (t) can be shown to be
2
Sy (f ) = Sx (f ) |H (f ) | , (6.14)
which is quite intuitive.

Chapter 7
Complex-Baseband Equivalent Channel

1
Linear communication schemes (e.g., AM, QAM, VSB) can all be represented (using complex-baseband
mod/demod) as:
Figure 7.1
It turns out that this diagram can be greatly simplied...

First, consider the signal path on its own:
Figure 7.2
Since s (t) is a bandpass signal, we can replace the wideband channel response h (t) with its bandpass
equivalent hbp (t):
27
28 CHAPTER 7. COMPLEX-BASEBAND EQUIVALENT CHANNEL
Figure 7.3
Then, notice that
[s (t) ∗ hbp (t)] 2e−j2πfc t s (τ ) hbp (t − τ ) dτ · 2e−j2πfc t

R
=
s (τ ) 2e−j2πfc τ hbp (t−τ ) e−j2πfc (t−τ ) dτ
R
= (7.1)
s (t) 2e−j2πfc t ∗ hbp (t) e−j2πfc t ,

=
which means we can rewrite the block diagram as
Figure 7.4
We can now reverse the order of the LPF and hbp (t) e−j2πfc t (since both are LTI systems), giving
Figure 7.5
Since mod/demod is transparent (with synched oscillators), it can be removed, simplifying the block
diagram to

29
Figure 7.6
Now, since m̃ (t) is bandlimited to W Hz, there is no need to model the left component of Hbp (f + fc ) =
F{hbp (t) e−j2πfc t }:
Figure 7.7
Replacing hbp (t) e−j2πfc t with the complex-baseband response h̃ (t) gives the complex-baseband equiv-
alent signal path:
Figure 7.8
The spectrums above show that hbp (t) = Re{h̃ (t) · 2ej2πfc t }.
Next consider the noise path on it's own:

Figure 7.9
Figure 7.10
From the diagram, ṽn (t) is a baseband version of the bandpass noise spectrum that occupies the frequency
range f ∈ [fc − Bs , fc + Bs ].
Since ṽn (t) is complex-valued, Ṽn (f ) is non-symmetric.
Say that w (t) is real-valued white noise with power spectral density (PSD) Sw (f ) = N0 . Since Sw (f )
is constant over all f, the PSD of the complex noise ṽn (t) will be constant over the LPF passband, i.e.,

31
f ∈ [−Bp , Bp ]:
Figure 7.11
A well-designed communications receiver will suppress all energy outside the signal bandwidth W, since
it is purely noise. Given that the noise spectrum outside f ∈ [−W, W ] will get totally suppressed, it doesn't
matter how we model it! Thus, we choose to replace the lowpass complex noise ṽn (t) with something simpler
to describe: white complex noise w̃ (t) with PSD Sw̃ (f ) = N0 :
Figure 7.12
We'll refer to w̃ (t) as "complex baseband equivalent " noise.

Putting the signal and noise paths together, we arrive at the complex baseband equivalent channel model :

Figure 7.13
The diagrams above should convince you of the utility of the complex-baseband representation in sim-
plifying the system model!

Chapter 8
Digital Communication
1
Transmission consists of
P
1. pulse shaping: m̃ (t) = n a [n] g (t − nT ),
2. modulation: s (t) = Re{m̃ (t) ej2πfc t }.
Reception consists of
1. demodulation: ṽ (t) = LPF{2r (t) e−j2πfc t },

2. ltering: y (t) = ṽ (t) ∗ q (t),
3. sampling: y [m] = y (mT ).
Figure 8.1
Building on analog QAM mod/demod components, digital mod adds pulse shaping & demod adds lter-
ing/sampling.
Simplifying via the complex-baseband equivalent channel:
33
34 CHAPTER 8. DIGITAL COMMUNICATION
Figure 8.2
Transmitter pulse shaping is used to convert the symbol sequence {a [n]} into the continuous message
m̃ (t):
P
m̃ (t) = n a [n] g (t − nT ) "baseband message"
(8.1)
T = "symbol period"
Thus, m̃ (t) can be seen to be a superposition of scaled and time-shifted copies of the pulse waveform g (t).
Example, if the symbol sequence [a [0] , a [1] , a [2] , a [3] , a [4]] equals [1, 3, −1, 1, 3], then the square pulse
g (t) shown below left yields the message m̃ (t) shown below right.
Figure 8.3
Receiver ltering (via q (t)) has two goals:
1. noise suppression (i.e., SNR improvement),

2. inter-symbol interference (ISI) prevention.
Noise suppression was briey discussed in Preliminaries (Chapter 3) and will soon be revisited in more detail.
Next we describe ISI.
Realize that, in the ideal digital comm system, the nth output y [n] would simply equal the nth input
a [n]. But in practice, y [n] can be corrupted by interference from the other symbols {a [m]}m6=n , known as
inter-symbol interference, and noise.

35
8.1 ISI-prevention for the noiseless trivial channel
Consider the idealized system
Figure 8.4
R P
y (t) = q (τ ) m̃ (t − τ ) dτ for m̃ (t) = n a [n] g (t − nT )
P R
n a [n] q (τ ) g (t − nT − τ ) dτ
= (8.2)
P
= n a [n] p (t − nT ) for p (t) = g (t) ∗ q (t) .
Thus, the idealized system can be re-drawn as
Figure 8.5
where
P P
y [m] = y (mT ) = n a [n] p (mT − nT ) = n a [n] p ((m − n) T ) . (8.3)
To make y [m] = a [m] (i.e., prevent ISI), we need

Figure 8.6
which is known as the Nyquist Criterion. This criterion can be simply stated as p (mT ) = δ [m] using
1 m=0 "distcrete-time impulse,"

δ [m] = {
0 m 6= 0 or "Kronecker delta." (8.4)
P∞
a [n] = m=−∞ a [m] δ [n − m] "sifting property."
Examples of Nyquist, and non-Nyquist, combined-pulses p (t) for [a [0] , a [1] , a [2] , a [3] , a [4]] =
[1, 3, −1, 1, 3]:
Figure 8.7

37
There is an interesting frequency-domain interpretation. Since
1
P∞ k
F P∞
T k=−∞ δ f− T ↔ m=−∞ δ (t − mT ) , (8.5)
we can see that
∞ ∞
1 X k F
X
P (f ) ∗ δ f− ↔ p (t) · δ (t − mT ).
T T m=−∞
k=−∞ (8.6)
| {z } | {z }
∞ ∞
1
P( k
)
P P
T f− T p(mT )δ(t−mT )
k=−∞ m=−∞
So, the time-domain Nyquist criterion p (mT ) = δ [m] implies
1
P∞ k
F
T k=−∞ P f− T ↔ δ (t) , (8.7)
which in turn implies
Figure 8.8
{ T1 P f −
k

In other words, the superposition of
T }k∈Z must sum to one. This frequency-domain version
of the Nyquist Criterion will soon come in handy...
A popular choice of combined pulse p (t) = g (t) ∗ q (t) is the raised-cosine pulse with rollo parameter
α ∈ [0, 1]:
cos(παt/T ) sin(πx)
pRC (t) = 1−(2αt/T )2
sinc (t/T ) , sinc (x) := πx
T |f | ≤ (1−α)
2T
(8.8)
(1−α)
πT 1−α
≤ (1+α)
2

PRC (f ) = { T cos 2α |f | − 2T 2T ≤ |f | 2T
.
(1+α)
0 2T ≤ |f |
Tradeo: larger α ⇒ less time-spread but more freq-spread:

Figure 8.9
So, we now know how to design the combined pulse p (t).

But what about the individual pulses g (t) and q (t)?
8.2 Maximizing SNR for ISI-free Pulses in White Noise
Now let's bring the noise back into consideration. Given
Figure 8.10
E{w̃ (t) w̃∗ (t − τ )} = N0 δ (τ ) (complex white noise) (8.9)
we want a {g (t) , q (t)} pair that maximizes the SNR of y [m].

Separating the noise and signal contributions to y [m] via

39
Figure 8.11
the SNR can be written
Es E{|ys [m]|2 }
= = (8.10)
SNR
En E{|yn [m]|2 } ,
where Es and En are average signal and noise energies.

Here, we treat both w̃ (t) and a [n] as random, implying that ys [m] and yn [m] are both random.
Notice that, with an ISI-free combined pulse p (t), we get
P
ys [m] = n a [n] p ((m − n) T ) = a [m] p (0)
R∞ (8.11)
p (0) = −∞
q (τ ) g (0 − τ ) dτ,
so that
R∞ 2
2
Es = E{|ys [m] | } = E{a [m] −∞ q (τ ) g (−τ ) dτ }

R 2
2 ∞ (8.12)
= E{|a [m] | } −∞ q (τ ) g (−τ ) dτ ,

| {z }
2
σa
where σa 2 denotes average symbol energy. Next, notice that
R∞
yn [m] = yn (mT ) = −∞
q (τ ) w̃ (mT − τ ) dτ, (8.13)
so that
R 2
∞
En = E{|yn [m] |2 } = E{ −∞ q (τ ) w̃ (mT − τ ) dτ }

R∞ R∞
= E{ −∞ q (τ ) w̃ (mT − τ ) dτ −∞ q ∗ τ ' w̃∗ mT − τ ' dτ ' }

R∞ R∞ (8.14)
q (τ ) −∞ q ∗ τ ' E{w̃ (mT − τ ) w̃∗ mT − τ ' }dτ ' dτ

= −∞
| {z }
N0 δ(τ ' −τ )
R∞ 2
= N0 −∞
|q (τ ) | dτ.
Putting these together, we nd
R∞ 2
2
σa | −∞
q(τ )g(−τ )dτ |
SNR = R∞ 2 . (8.15)
N0 −∞
|q(τ )| dτ

Cauchy-Schwarz says
R 2
∞ R∞ 2 R∞ 2
−∞ b (τ ) c (τ ) dτ ≤ |b (τ ) | dτ · |c (τ ) | dτ

−∞ −∞
(8.16)
with equality i b (τ ) = Kc∗ (τ ) for any K,
which implies
2 R∞
σa 2
SNR ≤ N0 −∞
|g (−τ ) | dτ
(8.17)
with equality i q (τ ) = Kg ∗ (−τ ) for any K.
Noting that SNR doesn't depend on K, we choose K = 1. Thus, given pulse g (t), the SNR-maximizing
receiver lter is
q (τ ) = g ∗ (−τ ) known as a "matched lter". (8.18)
We can write this in the frequency domain as
R∞ R∞ ∗
Q (f ) = −∞
q (τ ) e−j2πf τ dτ = g (−τ ) e−j2πf τ dτ
R∞ hR−∞ i∗
∞
= −∞
g ∗ (t) ej2πf t dt = −∞
g (t) e −j2πf t
dt (8.19)
= G∗ (f ) .
Summary: For SNR-maximizing ISI-free pulses, we need
1. G (f ) Q (f ) = P (f ) satisfying the Nyquist criterion,

2. Q (f ) = G∗ (f ),
2
which together imply |G (fp) | must satisfy the Nyquist criterion.
One option is G (f ) = PRC (f ), since PRC (f ) was Nyquist. We call this the square-root raised cosine
−1
(SRRC) pulse. Working out the details of F {GSRRC (f )}, we nd
(1−α)sinc( Tt (1−α)) 4αcos(π Tt (1+α))
gSRRC (t) = 2 + “ 2
” . (8.20)
1−(4α Tt ) π 1−(4α Tt )

41
Figure 8.12
∗
At the receiver, we would use qSRRC (t) = gSRRC (−t) = gSRRC (t); the latter equality is due to gSRRC (t)
being real and symmetric.


Chapter 9
Discrete-Time Implementation of Digital

Communication
1
Digital implementation of transmitter pulse-shaping and receiver ltering is much more practical than analog.
First, recall the analog implementation of our digital communication system: from the module Digital
Communication (Chapter 8).
Figure 9.1
To proceed further, we need an important DSP concept called sinc reconstruction:
1
If waveform x (t) is bandlimited to Hz, then (9.1)
2Ts
∞
X 1
x (t) = x [n] sinc (t−nTs ) for x [n] = x (nTs ) . (9.2)
n=−∞
Ts
In other words, a bandlimited waveform can be reconstructed from its samples via sinc pulse shaping.
43
CHAPTER 9. DISCRETE-TIME IMPLEMENTATION OF DIGITAL
44
COMMUNICATION
Figure 9.2
9.1 Discrete-Time Pulse-Shaping

T 1+α
Applying
P -sampling and reconstruction to g (t) (where the SRRC pulse bandwidth
2T requires the use of
P ≥ 2),
P
τ − l PT g [l] = g l PT
P
g (τ ) = l g [l] sinc T for
P
m̃ (t) = n a [n] g (t − nT )
P T
P P
= n a [n] l g [l] sinc T t − nT − l P
(9.3)
P T
P P
= n a [n] k g [k − nP ] sinc T t − k P via k = nP − l
P X
a [n] g [k − nP ]sinc T t − k PT
P

= k n
| {z }
:=m̃[k]
The sequence m̃ [k], a weighted sum of P -shifted pulses g [k], can be generated by P -upsampling a [n] (i.e.,
inserting P −1 zeros between every pair of samples) and ltering with g [k]:
Figure 9.3

45
note: sinc-pulse shaping = digital-to-analog conversion (DAC).
9.2 Discrete-Time Receiver Filtering

T
Applying
P -sampling and reconstruction to bandlimited q (τ ):
∞
P
τ − l PT q [l] = q l PT
P
q (τ ) = q [l] sinc T for (9.4)
l=−∞
where again we need P ≥ 2, yields
R∞
y k PT = −∞ q (τ ) ṽ k PT − τ dτ

y↑ [k] =
R∞ P ∞
q [l] sinc PT τ − l PT ṽ k PT − τ dτ

= −∞
l=−∞ Z ∞
∞

P ' T
τ ṽ (k − l) − τ dτ ' '
P
= q [l] sinc (9.5)
l=−∞ −∞ T P
| {z }
{sinc( P t )∗ṽ(t)} =ṽ[k−l]
T t=(k−l) T
P
∞
P
= q [l] ṽ [k − l] = q [k] ∗ ṽ [k] ,
l=−∞
from which y [m] is obtained by keeping only every P th sample:
Figure 9.4
P

i.e., downsampling. Here sinc T t does anti-alias ltering.
9.3 Discrete-Time Complex-Baseband Channel
Finally, we derive a discrete-time representation of the channel between m̃ [k] and ṽ [k]:
Figure 9.5

46
COMMUNICATION
Using w̃ [k] to refer to the noise component of ṽ [k], it can be seen from the block diagram that
R∞ P
k PT − τ

w̃ [k] = w̃ (τ ) sinc dτ (9.6)
−∞ T
m̃ [k] is eectively pulse-shaped by sinc PT t ∗ h̃ (t) ∗

To model the signal component of ṽ [k], realize that
sinc PT t . But since the frequency response of sinc PT t has a at gain of PT over the signal bandwidth, and

thus the bandwidth of h̃ (t),
T2

P P
sinc t ∗ h̃ (t) ∗ sinc t = 2 h̃ (t) . (9.7)
T T P
T T2
So, with
P -sampling and reconstruction of P 2 h̃ (t), i.e.,
∞
T2 P T2
t − i PT h i PT
P
P2 h̃ (t) = h [i] sinc T for h [i] = P2 (9.8)
i=−∞
we can write ṽ [k] as
2
w̃ [k] + l m̃ [l] PT 2 h̃ k PT − l PT
P
ṽ [k] =
P P
= w̃ [k] + l m̃ [l] i h̃ [i] sinc (k − l − i)
| {z } (9.9)
δ[k−l−i]
P
= w̃ [k] + l m̃ [l] h̃ [k − l]
yielding the discrete-time channel
Figure 9.6
Merging the discrete-time channel with the discrete-time modulator and demodulator yields
Figure 9.7

47
known as the fractionally sampled system model. This model is very convenient for MATLAB simulation
and acts as a foundation for further analysis.

48
COMMUNICATION

Chapter 10
Error Analysis of Digital

Communications
1
Recall the gure below, from the module Discrete-Time Implementation of Digital Communication (Chap-
ter 9). When the channel is trivial and noiseless and the pulses satisfy the Nyquist criterion (i.e.,
g [k] ∗ q [k] = δ [k]), the digital comm system will work perfectly, yielding y [n] = a [n].
Figure 10.1
In practice, however,
• the pulses g [k] and q [k] will be truncated to nite length,

• the channel will not be trivial (i.e., h̃ [k] 6= δ [k]), and
• the channel will not be noiseless (i.e., w̃ [k] 6= 0),
leading to y [n] 6= a [n], in which case we must infer the value of a [n] from the received samples {y [m]}∞
m=−∞ .
For now, we consider using only the single sample y [n] to infer a [n].
Key question: What are the mechanisms by which errors are made?
To better understand error behavior, we can plot the eye diagram or the constellation diagram and
calculate the symbol error rate (SER).
10.1 Eye Diagrams
Usually used when a [n] ∈ R

, the eye diagram
is a plot which superimposes T -second segments of Re{y (t)}
over the time intervals t ∈ nT − T2 , nT + T2 for many integers n.
49
50 CHAPTER 10. ERROR ANALYSIS OF DIGITAL COMMUNICATIONS
Figure 10.2
In MATLAB, the eye diagram can be made by superimposing P -sample segments of Re{y↑ [k]} corre-
P P −1
sponding to intervals k ∈ {nP − 2 , · · · , nP + 2 } for many n. (Usually P ≥ 8.)
Figure 10.3
If a [n] ∈ C, eye diagrams can be plotted for both the I and Q channels using Re{y (t)} and Im{y (t)},
respectively. MATLAB for digital mod/demod with eye diagram:

51
(a) (b)
Figure 10.4
10.2 Constellation Diagrams
The constellation diagram is a plot of Im{y [n]} vs. Re{y [n]} for many integers n. When the comm system
is working well, the points cluster around the symbol alphabet values:

Figure 10.5
Recall that y [n] ∈ C due to the complex-baseband channel model, regardless of whether a [n] ∈ R or
a [n] ∈ C.
Sometimes it is instructive to superimpose a plot of Im{y↑ [n]} vs. Re{y↑ [n]}, which approximates the
trajectory of y (t) in the complex plane:
Figure 10.6
MATLAB for digital mod/demod with constellation diagram:

53
(a) (b)
Figure 10.7
10.3 Popular Symbol Alphabets
QAM : quadrature amplitude modulation

PAM : pulse amplitude modulation
PSK : "phase shift keying"

Figure 10.8
Note that: "QPSK" = 4-QAM = 4-PSK and "BPSK" = 2-PAM = 2-PSK

When the alphabet entries are spaced by ∆ and picked with equal probability, the symbol variance
2
σa2 = E{|a [n] | } obeys:
alphabet
2
M -QAM M -PAM M -PSK
σa 2
2 2
∆ ∆ ∆2

6 M2 − 1 12 M2 − 1 4sin2 (π/M )
Table 10.1
10.4 Decision Regions
A reasonable way to infer the transmitted symbol a [n] from the received sample y [n] is to decide that a [n]
was the alphabet element nearest to y [n].
Nearest-element decision making is equivalent to using decision regions whose boundaries are equidistant
from the two nearest alphabet elements:
Figure 10.9

55
When a [n] = a, the symbol error rate (SER) equals the probability that y [n] lies outside the decision
region corresponding to alphabet member a. Writing y [n] = a [n] + e [n], we represent the cumulative eect
of noise and ISI by the error e [n]. Usually we model e [n] as a Gaussian random variable with mean 0 and
variance σe 2 .
10.5 Symbol Error Rate (SER) for M -PAM
Let's rst consider an M -PAM alphabet, where a [n] ∈ R. Since the decision regions show that Im{y [n]} is
not useful, we'll consider only the real parts of y [n] and e [n].
When a [n] = a, we have
y [n] = a + e [n], implying that y [n] is Gaussian with mean a and variance σ e 2 ,
abbreviated as N a, σe2 . This is illustrated below for the case of 4-PAM:
Figure 10.10
Formally, we say that py[n]|a[n] (y|a), the probability density function (pdf ) of y [n] conditioned on a [n] =
a, obeys
!
2
1 (y − a)
py[n]|a[n] (y|a) = p exp − . (10.1)
2πσe2 2σe2
| {z }
N (a,σe2 )
Basically, py[n]|a[n] (y|a) tells us how likely it is that y [n] = y given that a [n] = a.
Consider rst the case where a is an interior (not an edge) element of the symbol alphabet. Given
∆ ∆
that a [n] = a, we make an error when y [n] < a − 2 or when y [n] > a + 2 . To nd the probability of the
latter error event, i.e.,
∆
P r{y [n] > a + | a [n] = a}, (10.2)
2
y ∈ a+ ∆

we integrate py[n]|a[n] (y|a) over
2 ,∞ :
!
∞ ∞ 2
(y − a)
Z Z
1
py[n]|a[n] (y|a)dy = exp − dy. (10.3)
2σe2
p
2
2πσe
a+ ∆
2
| {z } a+ ∆
2
N (a,σe2 )
The integral represents the shaded area below:

Figure 10.11
This integral is often solved via
!
∞ 2
(y − µ) x−µ
Z
1
√ exp − dy = Q , (10.4)
x 2πσ 2 2σ 2 σ
| {z }
N (µ,σ 2 )
using the Q function:
Figure 10.12
While the Q function is not represented in MATLAB, it can be calculated using the complementary
error function erf c:

1 x
Q (x) = erf c √ (10.5)
2 2

57
In any case, the latter error event occurs with probability

∆ (a+ ∆2 )−a
P r{y [n] > a + 2 | a [n] = a} = Q σe
(10.6)
∆
= Q 2σe .
By symmetry, the former error event probability is also

∆ ∆
P r{y [n] < a − 2 | a [n] = a} = Q 2σe . (10.7)
Since these two events are disjoint, the probability of making a decision error on an interior symbol equals
their sum:

∆ ∆ ∆
Q +Q = 2Q . (10.8)
2σe 2σe 2σe
For edge symbols, we experience half the decision error probability, since there is only one decision boundary
to cross.
Finally, we average over the conditional error probabilities:
P
P r{error} = P r{error|a [n] = a}P r{a [n] = a}
a ∈ alphabet
| {z }
=1/M for all a
(10.9)

∆ 2 ∆
= Q 2σ e
·M + 2Q 2σ e
· M−
M
2

2 M− 1 ∆

= M Q 2σe
,
2
∆

Using σa2 = 12 M2 − 1 , we can nally write
s !
σa2

M −1 3
SERM -PAM =2 Q . (10.10)
M (M − 1) σe2
2
10.6 Symbol Error Rate (SER) for M 2 -QAM
With QAM, we have complex-valued y [n], a [n], e [n]. We'll assume that Re{e [n]} and Im{e [n]} are uncor-
related and equal variance. To calculate SER, we can re-use the PAM approach with a few modications:
1. integration is done on the complex plane,

σe2
2. σ e 2 -variance e [n]⇒
2 -variance Re{e [n]} & Im{e [n]},
2
3. M -QAM has 4 corner points, 4 (M − 2) edge points, and M 2 − 4M + 4 interior points,
4. calculate P r{error|a [n] = a} via 1−P r{correct|a [n] = a}, since the regions of integration are simpler:
Figure 10.13

After a bit of algebra, we nd
h q 2
i2
M −1
3 σa
SERM 2 -QAM = 1− 1−2 Q . (10.11)
M (M 2 −1) σe2
10.7 Bit Error Rate (BER) and Gray Coding
With an M -ary alphabet, there are log 2 M bits per symbol, so 1 symbol error could cause up to log 2 M bit
errors.
Gray coding is a clever way of mapping bits to symbols so that neighboring symbols dier by only a
single bit. Since the vast majority of errors occur when y [n] falls into a neighboring decision region, Gray
coding yields BER ≈ SER.
Figure 10.14

INDEX 59
Index of Keywords and Terms
Keywords are listed by the section with that keyword (page numbers are in parentheses). Keywords
do not necessarily appear in the text of the page. They are merely associated with that section. Ex.
apples, 1.1 (1) Terms are referenced by the page they appear on. Ex. apples, 1
A alphabet, 2(3) erfc, 10(49)

AM, 5(15) expectation, 6(25)
amplitude modulation, 5(15) eye diagram, 10(49)
analog, 9(43)
analog communication, 1(1)
F ltering, 3(5), 4(13), 6(25), 8(33)
r2, 3(5)
anti-alias ltering, 9(43)
rls, 3(5)
ATSC, 5(15)
rpm, 3(5)
attenuation, 4(13)
FM, 5(15)
autocorrelation, 6(25)
B
Fourier transform, 3(5)
bandlimited, 9(43) fractional sampling, 9(43)
bandpass equivalent, 7(27) frequency domain, 3(5)
bandwidth, 3(5) frequeny modulation, 5(15)
baseband, 2(3)
BER, 10(49)
G Gray coding, 10(49)
bit error rate, 10(49)

bits, 2(3)
H HPF, 3(5)
BPF, 3(5) I in-phase, 5(15)
C carrier frequency, 2(3)

inter-symbol interference, 8(33)
ISI, 8(33)
carrier tone, 5(15)
Carson's rule, 5(15) L large carrier, 5(15)
Cauchy-Schwarz, 8(33) linear time invariant, 3(5)
causality, 3(5) linear-phase, 3(5)
coder/mapper, 2(3) linearity, 3(5)
complex-baseband, 7(27) LPF, 3(5)
conditional probability, 10(49) LTI, 3(5)
conjugate symmetry, 3(5)
constellation diagram, 10(49) M matched lter, 8(33)
convolution, 3(5) Matlab, 3(5)
D
message, 1(1)
DAC, 9(43) modulation, 5(15), 8(33)
decision regions, 10(49) modulator, 2(3)
demodulation, 5(15), 8(33) multipath propagation, 4(13)
digital, 9(43)
digital communication, 1(1) N noise, 4(13), 6(25)
digital-to-analog, 9(43) non-causal, 3(5)
Dirac delta, 3(5) NTSC, 5(15)
discriminator, 5(15) Nyquist criterion, 8(33)
downsampling, 9(43)
P PAM, 10(49)
E envelope detection, 5(15) passband, 2(3)

path delays, 4(13)

60 INDEX
phase-shift keying, 10(49) sinc reconstruction, 9(43)

pilot tone, 5(15) SNR, 4(13), 8(33)
power spectral density, 6(25) square-root raised cos, 8(33)
power spectrum, 6(25) SRRC, 8(33)
PSD, 6(25) suppressed carrier, 5(15)
PSK, 10(49) symbol alphabet, 10(49)
pulse shaper, 2(3) symbol error rate, 10(49)
pulse shaping, 8(33) symbols, 2(3)
Q Q function, 10(49) T thermal noise, 4(13)

QAM, 5(15), 10(49) time domain, 3(5)
quadrature, 5(15)
U
R
uncorrelated, 6(25)
raised-cosine, 8(33)
random process, 6(25)
V vestigial sideband, 5(15)
VSB, 5(15)
random signal, 6(25)
S SER, 10(49)
W white noise, 6(25), 7(27)
wide-sense stationary, 6(25)
sifting, 3(5)
wideband, 7(27)
signal-to-noise ratio, 4(13), 8(33)
sinc pulse shaping, 9(43)
Z zero-mean, 6(25)

ATTRIBUTIONS 61
Attributions
Collection: Introduction to Analog and Digital Communications

Edited by: Phil Schniter
URL: http://cnx.org/content/col10968/1.2/
License: http://creativecommons.org/licenses/by/3.0/
Module: "Introduction to an "Introduction to Analog and Digital Communication""

Used here as: "Introduction"
By: Phil Schniter
URL: http://cnx.org/content/m31807/1.4/
Pages: 1-2
Copyright: Phil Schniter
Module: "Preview of Comm System Components"

By: Phil Schniter
Pages: 3-4
Module: "Preliminaries to an "Introduction to Analog and Digital Communication""

Used here as: "Preliminaries"
By: Phil Schniter
Pages: 5-11
Module: "The Communications Channel"

By: Phil Schniter
Pages: 13-14
Module: "Analog Communication"

By: Phil Schniter
Pages: 15-24
Module: "Random Signals and Noise"

By: Phil Schniter
Pages: 25-26

62 ATTRIBUTIONS
Module: "Complex-Baseband Equivalent Channel"

By: Phil Schniter
Pages: 27-32
Module: "Digital Communication"

By: Phil Schniter
Pages: 33-41
Module: "Discrete-Time Implementation of Digital Communication"

By: Phil Schniter
Pages: 43-47
Module: "Error Analysis of Digital Communications"

By: Phil Schniter
Pages: 49-58

Introduction to Analog and Digital Communications
These modules comprise a one-quarter introductory course on analog and digital communication, with an
emphasis on digital communication, and include many Matlab examples. The contents are summarized
as follows. First, a review of preliminary material is given (including the Fourier transform, Dirac delta,
LTI systems, and Matlab lter design). Next, the communications channel, describing the eects of signal
propagation, is described. Several forms of analog communication are then detailed: AM, QAM, VSB, FM.
An introduction to power spectra is then presented, followed by the complex-baseband equivalent channel
respresentation, for use in the sequel. Digital communication is described next, starting with Nyquist pulse-
shaping, matched ltering, and discrete-time implementation/modeling. After eye diagrams and constellation
diagrams are detailed, expressions for symbol error rate (using the Q function) are derived.
About Connexions
Since 1999, Connexions has been pioneering a global system where anyone can create course materials and
make them fully accessible and easily reusable free of charge. We are a Web-based authoring, teaching and
learning environment open to anyone interested in education, including students, teachers, professors and
lifelong learners. We connect ideas and facilitate educational communities.
Connexions's modular, interactive courses are in use worldwide by universities, community colleges, K-12
schools, distance learners, and lifelong learners. Connexions materials are in many languages, including
English, Spanish, Chinese, Japanese, Italian, Vietnamese, French, Portuguese, and Thai. Connexions is part
of an exciting new information distribution system that allows for Print on Demand Books. Connexions
has partnered with innovative on-demand publisher QOOP to accelerate the delivery of printed course
materials and textbooks into classrooms worldwide at lower prices than traditional academic publishers.

Introduction To Analog and Digital Communications 2.2

Uploaded by

Copyright:

Available Formats

Introduction To Analog and Digital Communications 2.2

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

Introduction To Analog and Digital Communications 2.2

Uploaded by

Copyright:

Available Formats

Introduction to Analog and Digital

Rice University, Houston, Texas

Available for free at Connexions <http://cnx.org/content/col10968/1.2>

Goal: Transmit a message from one location to another.

twisted pair wire (e.g., telephoneA )

coaxial cable (e.g., TVA,D , dataD )

ber optic cable (e.g., ethernetD )

EM waves (e.g., cellular phonesA,D , WiFiD , TVA,D )

water waves (e.g., underwater networkA,D )

where A = analog and D = digital.

Available for free at Connexions <http://cnx.org/content/col10968/1.2>

1.1 Analog Communication

• Perfect recovery is impossible in the presence of noise!

1.2 Digital Communication

Available for free at Connexions <http://cnx.org/content/col10968/1.2>

Preview of Comm System Components

• Translates baseband analog message to passband:

where fc is the carrier frequency.

system transmission band λ/10

UHF (TV) 0.33 GHz 101 cm

cellular 824960 MHz 3 cm

WiFi 2.4 GHz 1 cm

1 This content is available online at <http://cnx.org/content/m31817/1.3/>.

Available for free at Connexions <http://cnx.org/content/col10968/1.2>

then ASCII text would be transmitted via

letter ASCII code symbol sequence

• Converts symbol sequence into a continuous waveform.

Ex: Say symbol sequence is [1, 3, −1, 1, 3]. Then

Available for free at Connexions <http://cnx.org/content/col10968/1.2>

3.1 Fourier Transform (FT)

• Linearity: F{c1 w1 (t) + c2 w2 (t)} = c1 W1 (f ) + c2 W2 (f ).

3.3 Dirac Delta (or continuous impulse) δ (·)

1 This content is available online at <http://cnx.org/content/m31815/1.3/>.

Available for free at Connexions <http://cnx.org/content/col10968/1.2>

that's often used to kick a system and see how it responds.

3. Freq-domain impulse δ (f ) corresponds to a DC waveform:

3.4 Frequency-Domain Representation of Sinusoids

Notice from the sifting property that

Thus, Euler's equations

cos (2πfo t) = 1 j2πfo t

and the Fourier transform pair ej2πfo t ↔ δ (f − fo ) imply that

Available for free at Connexions <http://cnx.org/content/col10968/1.2>

3.5 Frequency Domain via MATLAB

Available for free at Connexions <http://cnx.org/content/col10968/1.2>

3.6 Linear Time-Invariant (LTI) Systems

Available for free at Connexions <http://cnx.org/content/col10968/1.2>

• Freq-domain: Multiplication with freq response H (f )

3.7 Linear Filtering

Freq-domain illustration of LPF, BPF, and HPF:

3.8 Lowpass Filters

Available for free at Connexions <http://cnx.org/content/col10968/1.2>

Ideal LPF with group-delay to :

A causal linear-phase LPF with group-delay to :

but MATLAB can give better causal linear-phase LPFs...

Available for free at Connexions <http://cnx.org/content/col10968/1.2>

Available for free at Connexions <http://cnx.org/content/col10968/1.2>

Available for free at Connexions <http://cnx.org/content/col10968/1.2>

The Communications Channel

The eects of signal propagation are usually modeled as:

ber optic cable (e.g., ethernetD )

• Translates baseband analog message to passband:

where fc is the carrier frequency.

UHF (TV) 0.33 GHz 101 cm

cellular 824960 MHz 3 cm

3.3 Dirac Delta (or continuous impulse) δ (·)

that's often used to kick a system and see how it responds.

The eects of signal propagation are usually modeled as:

• electronic circuitry (thermal/shot noise or quantization noise); usually broadband,

SNR can be improved with appropriate ltering at receiver:

The signal may propagate along paths with dierent lengths:

Since dierent path lengths imply dierent path delays:

5.1 AM with suppressed carrier

When the receiver oscillator has {freq,phase} oset {γ, φ}:

5.2 AM with pilot tone or carrier tone

The passband VSB lter is a BPF C (f ) where

which implies its inside rollo is symmetric around f = fc :

Motivation: ltering at baseband is usually much cheaper than ltering at passband.

D1 : narrowband FM ,

• Message signal ltered to freq interval [30,15k] Hz.