Ch7.

1: Baseband Communications & Noise

Information source
and input transducer

Source Coding

Channel Coding

Modulator

Questions to be answered:
System Model: AWGN Channel
White Gaussian Noise: A Random Process

Ch7.1: Binary
Communications

Channel

Suboptimal Receiver: Integrate-and-Dump

Performance Evaluation: How Good is the
Receiver?

Information sink
and output transducer

Elec3100 Chapter 7.1

Source Decoding

Channel Decoding

Demodulator
(Matched Filter)
1

Ch7.1: Baseband Communications & Noise

System Model
White Gaussian Noise
Suboptimal Receiver

Performance Evaluation
BER
A General Case

Elec3100 Chapter 7.1

Digital Communications
In digital communication systems, the signals
take discrete values to represent binary data.
For example, -A, A can be used to represent
logical levels, 0 and 1, respectively.

The above modulation signaling is called binary

digital communications since there are only two
bits.

Elec3100 Chapter 7.1

Baseband Digital Data Transmission

The system model for baseband digital data
transmission is

AWGN

1
n t : PSD N 0
2

y t

st
Transmitter

Elec3100 Chapter 7.1

(+A,-A)

Receiver

Binary signal

Baseband Digital Data Transmission

Example of a digital signal and transmitted waveform is

st

Example of a noise-corrupted received signal is

Elec3100 Chapter 7.1

Ch7.1: Baseband Communications & Noise

System Model
White Gaussian Noise
Suboptimal Receiver

Performance Evaluation
BER
A General Case

Elec3100 Chapter 7.1

Types of Noise
In this course, we assume internal noise sources are
much more significant than external ones and deal with
these only.
Noise internal to a communications system arises as a
result of random motion of charge carriers within the
devices (transistors, resistors, diodes, etc)
composing the system.

Internal noise is classified as thermal, shot,

flicker or others.
Elec3100 Chapter 7.1

Thermal Noise
Thermal noise arises from the random motion of charges
in conducting medium (such as resistors) and is the most
significant noise source we need to consider in
ELEC3100.
Essentially, by connecting a very sensitive oscilloscope
across a resistor we can observe thermal noise.
A typical noise output would look like:

Elec3100 Chapter 7.1

Nyquist Theorem
The signal is completely random and cannot be
predicted. It has an average value of zero but has
associated with it a certain mean square voltage.
The mean square voltage associated with the thermal
noise can be found from the Nyquist Theorem:

v 4kTBR
2
n

where

T = Temperature of the resistor in Kelvin [K] = [C] + 273.15

k = Boltzmanns constant 1.38 x 10-23 joule/K
B = Bandwidth of interest

R = Resistance
Elec3100 Chapter 7.1

Noise can be modeled as a Random Process

Definition: A random process maps a probability space S
to a set of functions, (, ).
It assigns to every outcome S a time function (, ) for
where is a discrete or continuous index set.
If is discrete (e.g. integer valued), (, ) is a discrete-time
random process.
If is continuous, (, ) is a continuous-time random
process.
For a fixed t, (, ) is a random variable.
Basically, we can understand a random process as a
sequence of random variables.

Elec3100 Chapter 7.1

10

Random Process: Example

Suppose that is selected at random from S = [0,1] and
consider X (t , ) cos(t ) for t

0
13

randomness

23

Elec3100 Chapter 7.1

1
time

11

Characterization of A RP:
Mean and Variance Functions

Mean

mX (t ) E[ X (t )] xf X (t ) ( x)dx

Variance

Note that the mean and variance may be functions of

time. However, since the randomness comes from , we
treat t as a constant in the above calculations.

Elec3100 Chapter 7.1

12

Characterization of A RP:
Autocorrelation and Autocovariance
o Autocorrelation

RX (t1 , t2 ) E[ X (t1 ) X (t2 )] xyf X ( t1 ), X ( t2 ) ( x, y )dxdy

o Autocovariance

C X (t1 , t 2 ) E[( X (t1 ) m X (t1 ))( X (t 2 ) m X (t 2 )]

RX (t1 , t 2 ) m X (t1 )m X (t 2 )
o Correlation coefficient

X (t1 , t2 )

Elec3100 Chapter 7.1

C X (t1 , t2 )
C X (t1 , t1 ) C X (t2 , t2 )

13

Wide Sense Stationary

Definition: A process () is wide sense stationary
(WSS) if and only if its mean is constant and its
autocorrelation function (1 , 2 ) (or autocovariance
function) depends only upon the time difference 1 2 .
mX (t ) m for all t

C X (t1 , t2 ) C X (t1 t2 ) for all t1 , t2

Suppose () is WSS with autocorrelation () where

t = 1 2 .

(0) is the average power of the process, E[X(t)2].

() is an even function of t.
|RX(t)| RX(0)
(the autocorrelation function is maximum at the origin)

Elec3100 Chapter 7.1

14

Wiener-Khinchine Theorem
Wiener-Kinchine theorem states that the autocorrelation
function and the power spectral density of a stationary
random process are Fourier transform pairs.

R X (t ) S X ( f )
Sample functions (, ) of stationary random processes are
power signals.
The power spectral density of a stationary random process x(t)
is defined as
| (, )|2
= lim

where , =

Elec3100 Chapter 7.1

/2
2
(,

/2

15

Gaussian Process
The additive noise in a communications system can be
modeled as a Gaussian process (by the central limit
theorem)
- at a particular time t, the noise signal amplitude will be
Gaussian distributed.
- we further assume that the Gaussian process is stationary and
has zero mean:

(ti ) E X(t i ) 0

i 1,2,, n

and its autocorrelation is

No
R X (t ) E X (ti ) X (ti t )
(t ) i 1,2,, n

2
Elec3100 Chapter 7.1

16

White Noise
If the power spectrum density is white, that is, all frequency
components have equal power.
N0
N0
S
(
f
)

RX (t )
(t )
X
2
2

FT

/2

() is zero except for t = 0 implies:

, = [( )( )]=0 for
( ) and ( ) are uncorrelated
0 = = 2 () = 2
Noise has infinite power

Elec3100 Chapter 7.1

17

Why White noise?

Elec3100 Chapter 7.1

18

Ch7.1: Baseband Communications & Noise

System Model
White Gaussian Noise
Suboptimal Receiver

Performance Evaluation
BER
A General Case

Elec3100 Chapter 7.1

19

Receiver Structure
One of the significant differences between analog and
digital communications systems is that for digital systems,
the probability of error is used as a measure of
performance where as in analog systems SNR is used.
We have modeled the AWGN. The next question is how to
build a receiver to obtain a good performance.
A possible receiver structure (integrate-and-dump) for
detecting the digital transmitted signals is shown below

y t

Elec3100 Chapter 7.1

t 0 T
t0

dt

V Threshold >0 choose +A

device
<0 choose -A

t t0 T

20

Integrate-and-Dump
Not necessarily optimum in all situations.
The integrator averages out the noise received so that the
output waveform will look like

Elec3100 Chapter 7.1

21

Integrate-and-Dump
We know that the noise at the input to the receiver is
AWGN.
We can expect the output from the integrator to have a
Gaussian noise distribution.
Putting these ideas into a mathematical framework we get

st

y t

Gaussian distribution

dt
T

Decision

N0
n t ~ SN f
2

Elec3100 Chapter 7.1

AWGN

Linear System
22

Integrate-and-Dump
A
st
A

0t T

if "1" transmitted

0t T

if " 0" transmitted

st ntdt
T

AT N

AT N
with
Elec3100 Chapter 7.1

if "1" is sent
if " 0" is sent
N is Gaussian distributed

ntdt
T

Linear combination of Gaussian

23

Ch7.1: Baseband Communications & Noise

System Model
White Gaussian Noise
Suboptimal Receiver

Performance Evaluation
BER
A General Case

Elec3100 Chapter 7.1

24

Key Figure of Merit

The probability of receiving a bit in error
for digital systems is an important measure
of performance.

Digital communications relies heavily on

these error calculations - VIP

Elec3100 Chapter 7.1

25

Problem
Assume the decision threshold is set at 0

Pe P 0 received 1 sent P 1 sent P 1 received 0P 0 sent

Pe P(V 01)P 1 P V 0 0P 0

Pe P(E 1)P 1 P E 0P 0

Total probability theorem

Conditional Error Probability

Elec3100 Chapter 7.1

Prior Probability

26

Error Probability Computation

To compute Pe, we need to compute
P(E|0) and P(E|1).
We have

st ntdt
T

AT N

AT N

if "1" is sent
if " 0" is sent

V is Gaussian distributed with variance 2 .

Elec3100 Chapter 7.1

27

Conditional Distribution
The key to estimating the error probability is to find
out the distribution of the received signal.
This in turn relies on the distribution of the noise.
The noise mean can be calculated as

E N E

Elec3100 Chapter 7.1

T
0

n t dt

E ntdt 0
T

28

Noise Variance

2
Var N E N E

Sn(f)

No/2

AWGN
R(t - n)
No/2

t-n

Elec3100 Chapter 7.1

T
0

n t dt

E n t n v dtdv
Rn t v dtdv
N0
t v dtdv
2

N0
0
dv
2
N 0T

2
2
T

29

Conditional Probabilities
fV v1

E V 1 AT
2

f
v1
~
N
AT,

V
2

VarV 1

fV v 0

AT

Shifted positive due to

The positive pulse +A

E V 0 AT
2

f
v
0
~
N
AT,

V
2
Var V 0

-AT

Elec3100 Chapter 7.1

n
30

Conditional Probabilities
fV v1

PE |1 P 0 received |1 sent
PE |1 P V 0 1 sent

fV v1dv

Threshold = 0

AT

PE | 0 P 1 received | 0 sent

PE | 0 P V 0 0 sent

-AT

n
Threshold = 0

fV v 0dv

31

Conditional Error Probability

Thus, we know that the output noise from the
integrator will have the following Gaussian
distribution

N ~ N 0, 2

f n n
Now,

n 2
exp 2
2
2
2
1

PE 1 P V 01 sent

PE 0 P V 00 sent

0
0

Elec3100 Chapter 7.1

fV v 0dv
fV v1dv
32

Total Error Probability

Noise distribution
is the same at 1
and 0

Assumptions:
1. Symmetry,

P E 0 P E 1

2. 0 and 1 are equally likely, then

1
P 0 P 1

Pe P E 0 P E 1

We therefore have

1
Pe P(E 1)P 1 P E 0P 0 P E 0 P E 0 P E 0
2

Pe P E 0

fV v 0dv

1
e
2

vAT 2
2 2

dv
33

Evaluation Relies on Q-Function

x

Let

Pe
where

v AT

AT

Elec3100 Chapter 7.1

VIP Transformation

AT
1 x22
e dx Q

2

Q(x)

dx

dv

1
2

t2
2
x

e dt

Q(.) function

34

Ch7.1: Baseband Communications & Noise

System Model
White Gaussian Noise
Suboptimal Receiver

Performance Evaluation
BER
A General Case

Elec3100 Chapter 7.1

35

A General Case
Next, note that one can represent st as s0 t (0
sent) or s1 t (1 sent).

s1 t

s0 t

A
t

T
t

-A

bit 0

bit 1

E1 0 s t dt A T
T

2
1

E 0 0 s t dt A T
T

2
0

Energy
Elec3100 Chapter 7.1

36

A General Result
Eb Energy/bit E0 P"0" sent E1 P"1" sent
1
E0 E1
2

But

E1 E 0 A T.
2

We can use these energy calculations to find a more

general result for the error probabilities. That is,

where

AT A 2T 2

Pe Q Q
2

N 0T

2
2

Elec3100 Chapter 7.1

37

SNR vs. Eb/No

A 2T 2 A 2T

Therefore, Pe Q

Q
2
2

T
Signal Energy to
2E
b
Noise Power Spectral
Q

Density Ratio
N 0
Relation with SNR?
E
1
erfc b
2
N 0
Error Function

where

with

erfcu 1 erf u

erf u

t 2

Complementary
Error Function

dt

38

Bit Error Rate (BER)

A graph of Pe for baseband signaling is
1.0

Actual

10-1
Pe

10-2
10-3

10-4

10log10Z

-10

where
Elec3100 Chapter 7.1

Pe Q

10

1
2Z erfc
2

Eb
Z
N0
39

Example 7.1
A baseband digital Tx system sends A valued
rectangular pulses through a channel at a rate of
1Mbps with amplitude 1V when the noise PSD is 10-7
W/Hz.

2
Q
2E
/N

Q
2A
T /N 0
b
0
Answer:

T 1/1000000 10

Q 2 106 /(2 107 ) Q( 10) Q(3.16)

u 2 / 2

e
Q(u)
u 2
Q(3.16) 0.00085

Probability of Error, Pe
40

Example 7.2
Digital data is to be transmitted through a baseband system
with N0=10-7 W/Hz and the received signal amplitude
A=20mV.
(a) If 1000 bits per second (bps) are transmitted what is the error
probability? Ans. Pe = 2.58 . 10-3.
(b) If 10000 bps are transmitted, to what value must A be adjusted
in order to attain the same error probability as in part a)?
Ans. A = 63.2 . 10-3 V = 63.2mV.

Elec3100 Chapter 7.1

41

BER vs. Data Rate

In the last example, for a fixed amplitude, error probability
increases as bit rate increases.
This can be understood by looking at the error probability
expression

A 2T 2 A 2T 2E

Q
b

Pe Q

2
2
N o

T
Increase in bit rate smaller bit period T
higher noise power

lower signal energy

2A

Pe Q
No

T
Elec3100 Chapter 7.1

Are they happening at the same time?

42