Digital Communications

Lecture # 11
Performance of Communication Systems
Corrupted by noise
(Section 7-1 thru 7-4)

1
 Primary Considerations
• In the design of a communication system we
focus on two aspects:
– Bandwidth Efficiency.
– Energy Efficiency.

2
 Bandwidth Efficiency

• So far, we have focused on the amount

of bandwidth that would be required to
send a given bandpass signal
• A system which sends a lot of data in a
small bandwidth is said to have a good
Bandwidth Efficiency

3
 Energy Efficiency
• Now we consider how the communication
system performs in noise
• A system which makes few errors when the
transmitter power is small is said to have good
Energy Efficiency
• It is possible to trade bandwidth efficiency for
energy efficiency in system design

4
 Binary Communication System Model:
Transmitter

• Binary signal is generated by transmitter

s1 t , 0  t  T , for a binary 1

st   
s2 t , 0  t  T , for a binary 0
• Where s1(t) and s2(t) might have different:
• Phases
• Amplitudes
• frequencies
5
 Binary Communication System Model:
Received Signal
• Received Signal   
r t  s t n t
• AWGN
• Additive White Gaussian Noise (AWGN)
Pn  f   N 0 2 [w/Hz]
• Equal noise power at all frequencies

Pn  f   N0 2
Power spectral
Density
(flat => “white”)

• Samples uncorrelated in time domain:

Autocorrelation
Rn      
N0 Function
2 (uncorrelated)

• Good approx. to receiver noise in many cases

6
 Binary Communication System Model:
baseband Received Signal

• After demodulation and matched filtering we


have a baseband signal: r t  s t  n
0 0  0 t 
• Baseband noise signal: n0 t 
• Baseband signal:

s01t   01
r t  m  1
s0 t    r0 t   
7
s02 t  r02 t  m  0
 Binary Communication System Model:
Sampled Received Signal
• After sampling, the “decision variable” or “test
statistic” is used for decisions
r0 t0   s0 t0   n0 t0  r0  s0  n0

• Sampled noise n0

• Sampled signal r01 m 1

r0  
8
r02 m0
 Digital Receiver Performance
• Bit Error Rate (BER):
– Number of errors Ne occurring in a certain interval
t, divided by the number of pulses Nt (0’s&1’s)
transmitted over the same period.
Ne Ne
BER  
N t Rbt
– Error rate depends on signal to noise ratio.

9
 Digital Receiver Performance
• Bit Error Rate (BER):
– Computation:
Amplitude

f r0 s2 
f r0 s2 
VT
Variation
in 1 level  f r0 s1 dr0


Decision
Threshold

 f r s dr
Variation

f r0 s1 
in 0 level
10
0 2 0
V
 Probability of Error
• We are interested in the probability of making
a wrong decision.

11
 Probability of Error
• We are interested in the probability of making
a wrong decision.

fr|m VT
fr(r0|s2) fr(r0|s1)

r0
s02 VT s01
12
 Probability of Error
   f r s dr
VT

• Let: Pr error s1 sent  0 1 0



   f r s dr

Pr error s2 sent  0 2 0
VT

• Average probability of error:

   
Pe  Perror s1 sent P s1 sent  P error s2 sent P s2 sent 
• If either signal is equally likely:

13
  
Pe  0.5P error s1 sent  0.5P error s2 sent 
 Two Basic Questions:
• What is the conditional pdfs for r01 and r02
– For a linear receiver, AWGN at the input will result
in AWGN at the output.
– Not valid for nonlinear receivers (e.g. Envelope
detectors = non coherent).
• Where do we put the optimum threshold to
minimize the error probability.
– Differentiate Pe and equate to zero.

14
 PDF of the “decision variable”

• The noise n(t) is a Gaussian random process

• After filtering, n0(t) is a Gaussian random
process
• After sampling, n0 is a Gaussian random
variable
• Mean of n0 : En0   0

• Variance of n0 :::  
E n 
2
0
2
0
15
 PDF of the “decision variable”
r0  s0  n0
• Since the signal is deterministic, s01 and s02
are constants

Er0 m  1  s01

Er0 m  0  s02

•When 1 or 0 is transmitted, the pdf

16
is simply shifted by s01 or s02.
 PDF of the “decision variable”
• If m =1 is sent:

 r0  s01 2

f r0 s1  
1 2 02
e
2  0
• If m =0 is sent:

 r0  s02 2

f r0 s2  
1 2 02
e
2  0
17
 Finding the optimum threshold
VT 
 r0  s01 2  
 r0  s02 2
1 1 1 1
Pe   e 2 02
dr0   e 2 02
dr0
2  2  0 2 VT 2  0
VT 
 r0  s01 2
dPe 1 d 1

2 02
 e dr0 
dVT 2 dVT  2  0
 
 r0  s02 2 •Use Leibniz’s
1 d 1

2 02
e dr0 Rule
2 dVT VT 2  0

18
 Finding the optimum threshold


VT  s0 1 
2

dPe 1 1 2 02
 e
dVT 2 2  0

VT  s0 2 
2

1 1 2 02
 e 0
2 2  0


VT  s01 2

VT  s02 2
2 02
e 2 02
 VT  s01   VT  s02 
2 2
e
s01  s02
19
VT 
2
 Problem Statement
• Our goal is to
• Find an expression for probability of error
• Choose a value of VT to minimize error
probability
• Now we must work with random
signals

20
 Calculate Pe (AWGN)
• Let    r0  s0   0 in the first integral

• Let   r0  s0   0 in the second integral

• We now have
VT 
 r0  s01 2  
 r0  s02 2
1 1 1 1
Pe   e 2 02
dr0   e 2 02
dr0
2  2  0 2 VT 2  0
• Or  2  2
1 1  1 1 
Pe  
2 V s   2
e d 
2
2 V  2
e 2 d
T 01 0 T  s02   0
1   VT  s01  1  VT  s02 
Pe  Q   Q 
21 2  0  2  0 
 General Result of Binary Signaling
• If we let
s01  s02
VT 
2
• Then we have:

 s01  s02   s01  s02 2 

Pe  Q   Q 
 2 0   4 2

 0 
22
 Interpretation of this result
• The term s01  s02  corresponds to signal power
2

• Increasing signal power decreases bit error rate

• The term  0 represents noise power
2

• Increasing noise power increases bit error rate

• The argument of the Q-function may be thought
of as a signal to noise ratio
• Usually this is expressed in terms of Eb N 0 ,
where Eb is the average energy per bit of
transmitted data
• The Q-function decreases very rapidly with its argument

23
 General Result of Binary Signaling
• To maximize the Q-function argument
The matched filter needs to be matched to the
difference signal s01(t )  s02 (t )2

Let N 0 2 is the PSD of the noise at the receiver

input, and Ed is the difference signal energy at
the receiver input, where
T
Ed   s1 t   s2 t  dt
2

24 0
 General Result of Binary Signaling
• From equation 6-161(also from last lecture)
for the matched filter output
S 2 Es
  
 N out N 0
• For binary signaling corrupted by AWGN,
matched filter reception, and by using the
optimum threshold setting, the bit error rate
is
 Ed 
Pe  Q 

 2N 0 
25
 Application of General Result

• For a specific signal set and modulation type:

• Determine s1 t  and s2 t 
T T
Eb   S1 t dt   S 2 t dt
1 2 1 2
• Determine Eb
20 20
T
• Express E  s t   s t 2 dt in terms of E
d  10
2 b

• Express Pe in terms of Eb and N 0

26
 Most important Results
(baseband)
• Unipolar Signaling

s1 t   A 0 s2 t   0 0
T T

A2T
Eb  Ed  A2T  2Eb
2

 Eb 
Pe  Q 

27
 N 0 
 Most important Results
(baseband)
• Polar Signaling

s1 t   A 0 s2 t    A 0
T T

Eb  A2T Ed  4 A2T  4Eb

 2 Eb 
Pe  Q 

28
 N0 
 Most important Results
(Bandpass)
• Binary Phase shift key (BPSK)

s1 t   A coswct   c  0 s2 t    A coswct   c  0
T T

A2T
Eb  Ed  2 A2T  4Eb
2

 2 Eb 
Pe  Q 

29
 N0 
 Most important Results
(Bandpass)
• On-Off key (OOK)

s1 t   A coswct   c  0 s2 t   0 0
T T

A2T A2T
Eb  Ed   2 Eb
4 2

 Eb 
Pe  Q 

30
 N 0 
 Error Probabilities for M-ary Modulation
• We present some simple results without
derivation
• Different quantities: symbol error rate, bit
error rate
• M-ary PSK:

Pe  Q 2 log 2 M Eb N0 sin M  
• Energy efficiency gets worse as M increases
• Bandwidth efficiency improves as M
increases
• QAM: Results are similar to M-ary PSK
31
 Error Probabilities for M-ary
Modulation
• M-ary FSK:

Pe  M  1Q  log 2 M Eb N0 

• Energy efficiency improves as M
increases
• Bandwidth efficiency gets worse as M
increases

32
 Comparison of different Modulation
schemes
• See figures in the text 504
• See the Table in the text (p505)

33
 Non Coherent Detection of Bandpass
Binary Modulation (7.4)
• Circuits for non coherent detection is
relatively simple.
– Non coherent OOK is the most popular signaling in
Fiber optics.
• Derivation of BER equations is more difficult.
• Will consider:
– Non coherent OOK.
– Non coherent FSK.

34
 Non Coherent Detection of Bandpass
Binary Modulation (7.4)

Signal
+ Binary
noise r (t )  s(t )  n(t ) ro (t ) output
Threshold device
Envelope Sample ~
m ~
BPF m
Detector & Hold
ro (to ) VT

 A cos(wot   )  n(t ) 0  t  T binary 1 sent

r (t )  
n(t ) 0  t  T binary 0 sent

n(t )  x(t ) cos(wot   )  y(t ) sin( wot   ) (Section 4.1)

35
 Non Coherent Detection of Bandpass
Binary Modulation (7.4)
  A  x(t )  cos(wot   )  y(t ) sin( wot   ) 0  t  T binary 1 sent
r (t )  
 x(t ) cos( wot   )  y(t ) sin( wot   ) 0t T binary 0 sent

How to get r0(t) from r(t) ?

V 
1 T 1
Pe   f (r0 / s1 )dr0   f (r0 / s2 )dr0
2  2 VT

Rician Rayleigh
 r  r0  A2   r A   r  r0 2
2 2 2

 02 e 2 I 0  0 2  r  0  0
e 2
r0  0
f (r0 / s1 )   f ( r / s )  
  
0 2
0 2
0 
0 Otherwise
 Otherwise
36
 Non Coherent Detection of Bandpass
Binary Modulation (7.4)
Rician Rayleigh
 r  r0  A2   r A  r  2
2 2 2
r0

 2e
0 2 
I  0
 r  0  0
e 2
r0  0
f (r0 / s1 )   f ( r / s )  
  
0 2 0 2
0 2
0 
0 Otherwise
 Otherwise

37
 Non Coherent Detection of Bandpass
Binary Modulation (7.4)
VT r
0 A
2 2
  r02
1 r0   r A 1 r 
Pe   2
I 0  0 2 dr0   02 e 2 2
2
e dr0
2 0 2
  2 VT 

• Notes:
– We are considering the equiprobable case.
– Integrals involving Bessel functions cannot be
obtained in closed form.
– We will make simplifying assumptions and try to
get Pe in closed form. A
 1
38

 Non Coherent Detection of Bandpass
Binary Modulation (7.4)
VT r
0 A
2 2
  r02
1 r0   r A 1 r 
Pe   2
I 0  0 2 dr0   02 e 2 2
2
e dr0
2 0 2
  2 VT 

A A
 1  Opt.Threshold close to
 2

r
0 A
2 2
  r02
 r0 A 
A/ 2
1 r0  1 r0  2 2
Pe   2
I 0  2 dr0   2 e
2
e dr0
  2 A/ 2 
2
39
2 0
 Non Coherent Detection of Bandpass
Binary Modulation (7.4)
r
0 A
2 2
  r02
 r0 A 
A/ 2
1 r0  1 r0  2 2
Pe   2
I 0  2 dr0   2 e
2
e dr0
  2 A/ 2 
2
2 0

ez
 1  I 0 z  
A r0 A
, let z  2
 2z 

 
r0 A
A/ 2 r02  A2 2  r02
1 r0  e 1 r0  2 2
Pe   2
dr0   2 e
2
e dr0
2 2
r0 A 2 A/ 2 
0
2
2
A/ 2

 r0  A 2 A2
1 r0 1  8 2
40
Pe 
2 
0
2 A 2
e 2 2
dr0  e
2
 Non Coherent Detection of Bandpass
Binary Modulation (7.4)
r
0 A
2 2
  r02
 r0 A 
A/ 2
1 r0  1 r0  2 2
Pe   2
I 0  2 dr0   2 e
2
e dr0
  2 A/ 2 
2
2 0

A/ 2 A/ 2
A r0

 1  Integrand negligible except near A   1 &
A 
0
 


 r0  A 2
1
A/ 2
1  1
A2
 2 Q( z)  e  z2 / 2
/ 2z 2 for z  1
Pe 
2 
 2 2
e 2 2
dr0  e
2
8

A2  A2 A2 A2 Eb 1

1  A  1  2 1  2 1  2 1  1 .
Pe  Q  e   e Pe  e  e
8 8 8 8 2 N 0 2TB p
e
2  2  2 2
A 2 2 2

Bb is the equivalent
N0
Where Eb  A T / 4 and   ( )(2 Bb )  N 0 Bb
2 2 bandwidth of the
41 2 bandpass filter
 Non Coherent Detection of Bandpass
Binary Modulation (7.4)
Non Coherent FSK
Signal
+
noise r (t )  s(t )  n(t ) Binary
vU (t ) output
BPF Envelope
f1,Bp Detector ro (t ) ~
m
 Threshold device
Sample ~
S & Hold
m

BPF Envelope
 ro (to ) VT

f2,Bp Detector
vL (t )

42
f (r0 / s2 )  f (r0 / s1 )
 Non Coherent Detection of Bandpass
Binary Modulation (7.4)
0 
1 1
Pe   f (r0 / s1 )dr0   f (r0 / s2 )dr0
2  20

Pe   f (r0 / s2 )dr0
0 
1 1
Pe   f (r0 / s2 )dr0   f (r0 / s2 )dr0
2  20
0
0 

Pe  P(vU  vL / s2 )
1 1
Pe    f (r0 / s2 )dr0   f (r0 / s2 )dr0
2 20
 
1 1
Pe   f (r0 / s2 )dr0   f (r0 / s2 )dr0
20 20

43
 Non Coherent Detection of Bandpass
Binary Modulation (7.4)

Pe  P(vU  vL / s2 )
Pe  P (vU  vL / s2 , vL )
vL

Pe   P(vU  vL / s2 ) f (vL / s2 )dvL
0

  
Pe     f (vU / s2 )dvU  f (vL / s2 )dvL
0 
vL 
44
 Non Coherent Detection of Bandpass
Binary Modulation (7.4)
  
Pe     f (vU / s2 )dvU  f (vL / s2 )dvL
0 
v L 
  v  2
 vU2 v v 2
 A 2

 L
 vL A 
Pe     U2 e 2 dvU  L2 e 2
I 0  2 dvL
2

0 
Lv
 

  
A2 Eb  1 
  
1  4 2 1 N 0  2 B pTb 
Pe  e  e
2 2
45
 Comparison to Coherent Detection

46