Вы находитесь на странице: 1из 14

2/17/2012

DEPARTMENT OF ELECTRICAL &ELECTRONICS ENGINEERING

DIGITAL COMMUNICATION
Spring 2010
Yrd. Do. Dr. Burak Kelleci

OUTLINE
Probability of Error due to Noise
| Intersymbol Interference
|

2/17/2012

PROBABILITY OF ERROR DUE TO NOISE


Since the matched filter is the optimum detector
of a known pulse in additive noise, the
probability of error rate due to the noise can be
calculated.
| Lets consider polar NRZ signaling, where
symbols 1 and 0 are represented by positive and
negative pulses by equal amplitude.
| The received signal in additive noise is
|

+ A + w(t ) symbol 1 was sent


x(t ) =
A + w(t ) symbol 0 was sent

PROBABILITY OF ERROR DUE TO NOISE


Lets assume that the receiver knows the starting
and ending times of each transmitted pulse.
| The receiver samples the output of the matched
filter at t=Tb and decides whether the
transmitted symbol is 0 or 1 from the received
noisy signal x(t)
|

2/17/2012

PROBABILITY OF ERROR DUE TO NOISE


|

In decision device, the sampled value is compared


with threshold value .
If the sampled signal is above the receiver assumes
that the transmitted symbol is 1.
y If the sampled signal is below the receiver assumes
that the transmitted symbol is 0.
y If the sampled signal is equal the receiver makes a
guess so that the outcome does not alter the average
probability
b bili off error.
y

There are two possible kinds of errors


y
y

Symbol 1 is chosen when actually 0 was transmitted.


Symbol 0 is chosen when actually 1 was transmitted.

To determine the probability of error, these two


situations should be analyzed separately.

PROBABILITY OF ERROR DUE TO NOISE


|

Suppose that symbol 0 was sent, then the


received signal is
x(t ) = A + w(t ) 0 t Tb

The signal at the matched filter output sampled


at t=Tb becomes
Tb

y = x(t )dt
0

1
= A +
Tb
|

Tb

w(t )dt
0

which represents the random variable Y

2/17/2012

PROBABILITY OF ERROR DUE TO NOISE


|

Since the additive noise is Gaussian distributed,


the variance of the random variable Y is

Y2 = E (Y + A)2

T T

1 b b
= E w(t )w(u )dtdu
Tb 0 0

1
=
Tb
=

1
Tb

Tb Tb

E[w(t )w(u )]dtdu


0 0

Tb Tb

R (t , u )dtdu
W

0 0

RW(t,u) is the autocorrelation function of the


white noise w(t)

PROBABILITY OF ERROR DUE TO NOISE


|

Since the power spectral density of the white


noise is N0/2, RW(t,u) is

RW (t , u ) =
|

N0
(t u )
2

The variance of Y becomes


1
=
Tb
2
Y

Tb Tb

0 0

N0
(t u )dtdu
2

N0
2Tb

2/17/2012

PROBABILITY OF ERROR DUE TO NOISE


|

The probability density function of the random


variable X which has a Gaussian distribution is

f X (x ) =
|

1
2 X

( x X )2

2 X2

Therefore the probability density function of the


random variable Y. given that symbol 0 has sent

f Y ( y | 0) =

N 0 / Tb

( y + A )2
N 0 / Tb

PROBABILITY OF ERROR DUE TO NOISE


|

The probability density function is plotted below/


Pe00 denotes the conditional probability of error,
given that symbol 0 has sent.

The shaded area from to infinity corresponds to


the range of values that are assumed as symbol 1.
| In the absence of noise the decision is always
symbol 0 for the sampled value of A.
|

2/17/2012

PROBABILITY OF ERROR DUE TO NOISE


|

The probability of error on the condition that


symbol 0 has sent is

Pe0 = P( y > | symbol 0 was sent )

= f Y ( y | 0)dy

N 0 / Tb

( y + A )2
N 0 / Tb

dy

PROBABILITY OF ERROR DUE TO NOISE


The value of he threshold can be assigned with
the probabilities of symbols 0 and 1 denoted by p0
and p1.
| It is clear that the sum of p0 and p1 must be 1.
| If we assume that symbols 0 and 1 are occurred
with equal probabilities (p0=p1=0.5), it reasonable
to assume the threshold halfway between +A
(symbol 1) and A
A (symbol 0).
0)
|

=0

2/17/2012

PROBABILITY OF ERROR DUE TO NOISE


|

The conditional probability of error when symbol


0 was sent becomes

Pe 0 =
|

N 0 / Tb

( y + A )2
N 0 / Tb

dy

Lets define a new variable z

y+A

z=

N 0 / 2Tb

PROBABILITY OF ERROR DUE TO NOISE


|

Lets reformulate Pe0


Pe 0 =

1
2

z2
2

dz

2 Eb / N 0

where Eb is the transmitted signal energy per bit


Eb = A 2Tb

2/17/2012

THE Q-FUNCTION
The Q-function is used by communication
engineers to determine the area under the tails of
the Gaussian distribution.
| Error function is the twice of the area under a
normalized Gaussian from 0 to u. The
complementary error function is defined as one
minus the error function

2
z2
erfc(u ) =
e dz
|

The Q-function and complementary error


function is related to each other
1
u
Q(u ) = erfc

2
2

PROBABILITY OF ERROR DUE TO NOISE


|

The conditional probability of error Pe0 can be


written in terms of Q
Q-function.
function.
2 Eb
Pe 0 = Q
N
0

For the conditional probability of error Pe1, the


mean is +A
A

f Y ( y | 1) =

N 0 / Tb

( y A )2
N 0 / Tb

2/17/2012

PROBABILITY OF ERROR DUE TO NOISE


|

The probability of error Pe1 is area from - to .


Pe1 = P( y > | symbol 1 was sent )

f ( y | 1)dy
Y

=
|

N 0 / Tb

( y A )2
N 0 / Tb

dy

Setting the threshold to =0 and putting


yA
z=
N 0 / 2Tb
results in Pe1=Pe0

PROBABILITY OF ERROR DUE TO NOISE


|

The average probability of symbol error Pe is

Pe = p0 Pe 0 + p1 Pe1
|

Since Pe0=Pe1 and p0=p1=0.5


Pe = Pe 0 = Pe1
2 Eb
Pe = Q
N
0

The average probability of symbol error with


binary signaling only depends on Eb/N0, the ratio
of the transmitted signal energy per bit to the
noise spectral density.

2/17/2012

PROBABILITY OF ERROR DUE TO NOISE


10

-1

Matlab Code:
-2

clear all;
close all;

Probability of Error P

10

10

10

-3

EbN0_dB=0:1e-3:10;
EbN0=10.^(EbN0_dB/10);
Pe=0.5*erfc(sqrt(2*EbN0)/sqrt(2));

-4

figure(1);
semilogy(EbN0 dB Pe 'linewidth'
semilogy(EbN0_dB,Pe,
linewidth ,2);
2);
10

10

-5

xlabel('E_b/N_0 (dB)');
ylabel('Probability of Error P_e');
grid;
-6

10

E /N (dB)
b

This waterfall curve makes the digital systems superior compared to analog
systems if Eb/N0 is above few dBs.

INTERSYMBOL INTERFERENCE
Intersymbol Interference (ISI) happens when the
channel has a frequency dependent amplitude
spectrum.
| The simplest case is band-limited channel.
|

For example, a brick-wall band-limited channel


passes all frequencies |f|<W without distortion and
blocks all frequencies |f|>W.

Consider a polar signaling where

+ 1 symbol 1 was sent


ak =
1 symbol 0 was sent

10

2/17/2012

INTERSYMBOL INTERFERENCE
|

Once the impulse modulator output is passed


through the transmit filter with the impulse
response of the pulse g(t), the transmitted signal
becomes

s (t ) = a k g (t kTb )
k

s(t) passes through the channel with impulse


response of h(t) added white Gaussian noise and
received by the receiver.
| The receiver passes the received signal through
the receive filter, which is the matched filter for
the optimum noise performance.
|

INTERSYMBOL INTERFERENCE
|

The output of the receive filter is

y (t ) = a k p(t kTb ) + n(t )


k

sampled at t=Tb and the decision device decides


the bits.
| The receive filter output is

y (t i ) = a k p[(i k )Tb ] + n(t i )


k =

= ai +

a p[(i k )T ] + n(t )

k =
k i

11

2/17/2012

INTERSYMBOL INTERFERENCE
The first term is the contribution of the ith
transmitted bit.
| The second term represents the residual effect of
all other transmitted bits on the decoding of ith
bit.
| This residual effect due to the occurrence of
pulses before and after the sampling instant ti is
called intersymbol interference (ISI).
(ISI)
| The last term n(ti) represents the noise sample at
the time ti
| When the signal-to-noise ratio is high, the system
is limited by ISI rather than noise.
|

THE TELEPHONE CHANNEL


|

The frequency response of a telephone channel is


shown below.

12

2/17/2012

THE TELEPHONE CHANNEL


|

The properties of this channel


The pass-band of the channel cuts of rapidly above
3.5KHz. Therefore, we should use a line code with a
narrow spectrum, so we can maximize the data rate.
y The channel does not pass DC. It is preferable to use
a line code that has no DC component such as bipolar
signaling or Manchester code.
y

These two requirements are contradictory.


The polar NRZ has narrow spectrum but has DC
y The Manchester has no DC, but requires higher
bandwidth.
y

THE TELEPHONE CHANNEL


|

For the data rate 1600bps, the signals are shown below.

a is NRZ line code and b is Manchester line code.


| Since NRZ requires DC, the signal drifts to zero volt
especially when there is long strings of the symbols of
the same polarity.
| Manchester code has no DC drift but has distortion.
|

13

2/17/2012

THE TELEPHONE CHANNEL


For the data rate 3200bps, the DC drift in NRZ is
less evident but there is still considerable signal
distortion.
| The Manchester code has significant spectral
components outside the band limits of this
channel.
|

14

Вам также может понравиться