ECE 6640

Digital Communications

Dr. Bradley J. Bazuin

Assistant Professor
Department of Electrical and Computer Engineering
College of Engineering and Applied Sciences
 Chapter 3
3. Baseband Demodulation/Detection.
1. Signals and Noise.
2. Detection of Binary Signals in Gaussian Noise.
3. Intersymbol Interference.
4. Equalization..

ECE 6640 2
 Sklar’s Communications System

Notes and figures are based on or taken from materials in the course textbook:
ECE 6640 Bernard Sklar, Digital Communications, Fundamentals and Applications, 3
Prentice Hall PTR, Second Edition, 2001.
 Chapter Goals

• Detection of Binary Signals plus Gaussian Noise

• The decision process
• Define Intersymbol Interference
• Error Performance Degradation
• Equalization Techniques

ECE 6640 4
 Demodulation and Detection

• Focus on Symbols, Samples, and Detection

• In the presence of Gaussian Noise and Channel Effects
ECE 6640 5
 Demodulation and detection

• Major sources of errors:

– Signal Path Loss
• Friis equation relates the received signal power to the transmitted
power, antenna gains, and distance that the signal travels in free space
– Thermal noise (AWGN)
• disturbs the signal in an additive fashion (Additive)
• has flat spectral density for all frequencies of interest (White)
• is modeled by Gaussian random process (Gaussian Noise)
– Inter-Symbol Interference (ISI)
• Due to the filtering effect of transmitter, channel and receiver,
symbols are “smeared”.
• Time spreading effects cause symbols to “overlap”
– Total symbol “length” may easily be 3+ symbol periods!

6
 Receiver Block Diagram

• Receive the transmitted symbol plus noise

– Symbol filter by channel
• Frequency down-conversion to baseband
– Receiver filtering and equalization (if needed) applied
• Symbol filter
– Matched filtering with Nyquist shaping for ISI
– Optimize the pre-detected signal prior to sampling
• Optimal Time Sampling
ECE 6640 – Peak filter response time 7
 Review Slides from ECE5640

• Chapter 9: Noise
• Chapter 10: Noise in analog modulated signals
• Chapter 11: Baseband Digital Transmission

ECE 6640 8
 Noise Approximation
• Uniform Noise Spectral Density
– Resistor description (Thevenin Model)

G vv f   2  R    T
• Available Power from the “noise source”
– Source output power into a matched load

v sout 
R
vs Psout
v 
 sout
2 2
v  1 v
 s    s
2

2R R  2  R 4R

G vv f  2  R    T   T N 0
G ss f     
4R 4R 2 2
N0
R ss     
2 9
 System Noise
• Since the noise power spectrum is uniform, a systems
noise power is the product of the noise power and the
integral of the filter power.

N0
SNN f   Hf   SN 0 N 0 f    Hf 
2 2

 
N0
R NN 0    Hf   df  N 0   Hf   df
2 2

2  0

10
 Noise Equivalent Bandwidth
• If we want the total noise power after the filter, we can
integrate the PSD for all frequencies or use the Filtered
noise autocorrelation function at zero.
– Both of these approaches may be difficult
– Could we great a more simple “noise equivalent bandwidth for
filters” that is rectangular?

11
 Noise Equivalent Bandwidth
 
N N
R NN 0  0   Hf   df  0  2   Hf   df
2 2

2  2 0

• When filtering, it is convenient to think of band-limited

noise, where the filter is a rect function with bandwidth
BEQN

 
H rect _ mod el  f   GainDC _ Power  f 
 rect Gain DC _ Power  H0 
2
 2 B 
 EQN 
 
H  f   df   H  f rect _ mod el  df  GainDC _ Power  BEQN  H 0   BEQN
2

2 2

0 0


 Hf   df
2

BEQN  0

H0
2
12
 Noise Equivalent Bandwidth
• Low pass filter
coherent _ gain  H0  Gain DC _ Power  H0 
2

 Hf   df
2

B EQN  0

H0 
2

N0
PN  R NN 0    2  H0   BEQN  H0   N 0  BEQN
2 2

• For a unity gain filter

– assumed when computing receiver input noise power

B EQN   Hf   df
2

N0
PN  R NN 0    2  BEQN  N 0  BEQN
2 13
Model of Received Signal
with Noise

© 2010 The McGraw-Hill Companies

 Signal Plus Noise
Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display.

Analog baseband transmission system with noise: Figure 9.4-2

 At  
Pr eDt     cos2  f c  t  t   nt   ht 
x t   At   cos2  f c  t  t   L 

At  At 
xR t    cos2  f c  t  t  st    cos2  f c  t  t   nt 
L L

• Additive Gaussian White Noise 15

 Signal-to-Noise Ratio
• Comparing the desired signal power to the undesired noise
power.
y t   xc t   nt 
• To compare signal and noise power, we must assume a
filtering operations

Linear Nonlinear Atten-

Filtering Distortion uation

xc t 

nt 
Noise
Transmitting Receiving y t   xc t   nt 
Antenna Antenna

RF Communication Channel
16
 Signal-to-Noise Ratio

y D t   xR t   nt  ht 

• Equivalent receiver input signal and noise (ER)

y R t   xER t   nER t 
• Equivalent destination signal and noise (D) or pre-
demodulation (PreD)

yPr eD t   xPr eD t   nPr eD t  17

 Signal-to-Noise Ratio

• Equivalent receiver input SNR (ER)

SNRR 

E xER t 
2

 
E xER t 
2

 SR

E nER t 
2

N 0  BEQN N 0  BEQN
 can be used to
• Equivalent destination SNR represent receiver
noise figure

SNRR 

E xPr eD t 
2


SD

SD
contributions


E nPr eD t 
2

N D   N 0  BEQN 18
 Increase in SNR with filtering
• If a filter matched to the input signal is applied, the noise
power would be reduced to the smallest equivalent noise
bandwidth that is allowed.
– Filter to minimize noise power
– Importance of the IF filter in a super-het receiver!

• Front-end filtering goals – a dilemna

– Minimize signal power loss (wider bandwidth)
– Minimize filter equivalent noise bandwidth
(narrower bandwidths)
– A trade-off must be made!

19
Typical Transmission Requirements
Signal Type Freq. Range SNR (dB)
Intelligible Voice 500 Hz to 2 kHz 5-10
Telephone Quality 200 Hz to 3.2 kHz 25-35
AM Broadcast Audio 100 Hz to 5 kHz 40-50
High-fidelity Audio 20 Hz to 20 kHz 55-65
Video 60 Hz to 4.2 MHz 45-55
Spectrum Analyzer 100 kHz-1.8 GHz 65-75

20
 CW Communication with Noise
Model of a CW communication system with noise: Figure 10.1-1

At 
x t   At   cos2  f c  t  t  vt    cos2  f c  t  t   n t 
L

A t   At  
x c t    cos2  f c  t  t  PreDt     cos2  f c  t   t   nt   hR t 
L  L 
21
Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display.
 Signal and Noise Power
• What are the signal and noise powers at the receiver?
T
1 2
Ps  lim   x t   x t   dt

T  T
T
2


Pn  E n t 
2

• What is the receiver input power
T T

   
2
1 1 2
Pv  lim   E vt   vt   dt  lim   E x c t   n t   x c t   n t   dt
 
T  T T  T
T T
2 2

22
 Receiver Signal plus Noise Power
• What is the receiver input power
T T

   
2
1 1 2
Pv  lim   E vt   vt   dt  lim   E x c t   n t   x c t   n t   dt
 
T  T T  T
T T
2 2
T
1 2
 
Pv  lim   E x c t   2  x c t   n t   n t   dt
T  T
2 2

T
2
T
1 2
  
Pv  lim   x c t   2  x c t   En t   E n t   dt
T  T
2 2

T
2

 T T T

1 
 
2 2 2

Pv  lim    xc t   dt   2  xc t   0  dt   E nt   dt 
2 2
T  T
T 2 T
2
T
2

T
1 2
 
Pv  lim   x c t   dt  E n t   Ps  Pn
T  T
2 2

T
2
23
 Signal-to-Noise Ratio (SNR)
• The SNR is a measure of the signal power to the noise
power at a point in the receiver.
– Typically described in dB
– The above computation was performed at the input
Ps
SNR 
Pn

• Matlab SNR example: SNR_AM_Example.m

– Pre-D “AM” SNR based in filter Beqn
• Effective BEQN RF due to sampling spectrum (B=Fis/2)
• AM signal power based on carrier plus signal

24
 Noise Equivalent Bandwidth
• Since the noise power spectrum is uniform, a systems
average noise power is the product of the noise power and
the integral of the filter power.

N0
SNN f   Hf   SN 0 N 0 f    Hf 
2 2

 
N0
R NN 0     Hf   df  N 0   Hf   df
2 2

2  0

25
 Noise Equivalent Bandwidth
 
N N
R NN 0   0   Hf   df  0  2   Hf   df
2 2

2  2 0

• When filtering, it is convenient to think of band-limited

noise, where the filter is a rect function with bandwidth
BEQN

 
H rect _ mod el  f   GainDC _ Power Gain DC _ Power  H0 
f
 rect   2
 2 B 
 EQN 
 

 Hf   df   Hf rect _ mod el  df  Gain DC _ Power  BEQN  H0  BEQN

2 2 2

0 0


 H  f   df
2

BEQN  0

H 0 
2
26
 Noise Equivalent Bandwidth
• Low pass filter
coherent _ gain  H0  Gain DC _ Power  H0 
2

 Hf   df
2

BEQN  0

H0 
2

PN  RNN 0   2  H 0  BEQN  H 0   N 0  BEQN

N0 2 2

• For a unity gain filter

– assumed when computing receiver input noise power

B EQN   Hf   df
2

N0
PN  R NN 0   2  BEQN  N 0  BEQN
2 27
 Filtering
• What happens if the receiver input is filtered?
v f t   x c t   n t  h1 t   h 2 t 
v f t   x c t   h 1 t   h 2 t   n t   h 1 t   h 2 t 

• What effect does the filter have on the signal?

– None or slight band edge de-emphasis, if and only if the filter is
“wider” than the signal bandwidth
– Now you know why a 3dB bandwidth isn’t that useful,
(3dB1/2 power point)!

Ps
SNR 
Pn
28
 Filtering
• What effect does the filter have on the noise?
– Normally you would expect for two filters
Ps
PN _ Post _ Filter1  N 0  BEQN _ Filter1 SNRPost _ Filter1 
N 0  BEQN _ Filter1

PN _ post _ filter 2  N 0  BEQN _ Filter 2 Ps

SNRPost _ Filter 2 
N 0  BEQN _ Filter 2

– Assume that the filters follow each other and that the first filter is
narrower than the second filter

PN _ Post _ Filter 2  N 0  min BEQN _ Filter1 , BEQN _ Filter 2 

 N 0  BEQN _ Filter1

Ps
SNRPost _ Filter 2 
N 0  BEQN _ Filter1 29
 Filters Provide SNR “Gain”
• If filter 2 Beq < filter 1 Beq:
Ps  BEQN _ Filter1 
SNR Post _ Filter 2   SNR Post _ Filter1   
N 0  BEQN _ Filter 2 B 
 EQN _ Filter 2 

• You expect the IF filter to be smaller than the front-end RF

or “pre-filtering” performed
 B EQN _ Filter1 
Gain Filter   
B 
 EQN _ Filter 2 
– Think about kTB at different bandwidths and you will derive the
same “gain”
– In typical receivers, the IF filter sets the
Pre-Demodulation Bandwidth
30
 Bandpass Noise Processing
cos2  f c  t 
Band Pass LowPass
Filter Filter

Bandpass filter bandwidth

may not be centered on fc
• an alpha offset
BT N0 BT
2

 fc fc
 f c    BT f c    BT

• What happens after mixing and lowpass filtering?

– assume LPF passes the entire baseband. 31
 Quadrature Noise (1)
• Question: Is Quadrature noise a different power than “real”
baseband noise
• Noise in a quadrature process
nt   ni t   cos2  fc  t   nq t   sin2  fc  t 
• Noise power is related as

 2
  2
 
E n t   E n i t   E n q t  
2
 N0
2

• What about ?

  
E nt   E ni t   cos2  fc  t     nq t   sin2  fc  t   
2 2


32
 Quadrature Noise (2)
• Noise in a quadrature process

  
E n t   E n i t   cos2  f c  t    n q t   sin 2  f c  t  
2 2

n t 2  cos 2 2  f  t   
 
 
i c

E n t   E  2  n i t   n q t   cos2  f c  t    sin 2  f c  t  

2

 
 n q t   sin 2  f c  t  
2 2


   2 1  2 1 
E nt   E ni t     cos2  2 f c  t  2     nq t     cos2  2 f c  t  2   
2

2  2 


   2 1

2
 

2 1

2
  
E nt   E ni t     E cos2  2 f c  t  2     E nq t     E cos2  2 f c  t  2   
2

 
E nt  
2 1
2
 
1

 E ni t    E nq t   0
2

2
2 N
2

33
 Mixing Noise (1)
• Think of the two noise bands as
1. The band of interest
2. The image band

nt   cos2  f LO  t   hIF t   n1i t   cos2   f LO  f IF   t   n1q t   sin 2   f LO  f IF   t  cos2  f LO  t   hIF t 
 n2i t   cos2   f LO  f IF   t   n2 q t   sin 2   f LO  f IF   t  cos2  f LO  t   hIF t 

1 n1i t   cos2  f IF  t   n1q t   sin 2   f IF   t  

nt   cos2  f LO  t   hIF t      hIF t 
2 n1i t   cos2  2 f LO  f IF   t   n1q t   sin 2  2 f LO  f IF   t 

1 n2i t   cos2   f IF   t   n2 q t   sin 2   f IF   t  

    hIF t 
2 n2i t   cos2  2 f LO  f IF   t   n2 q t   sin 2  2 f LO  f IF   t 

 n1i t   cos2  f IF  t   n1q t   sin 2  f IF  t  hIF t 

1
nt   cos2  f LO  t   hIF t  
2
  n2i t   cos2  f IF  t   n2 q t   sin 2  f IF  t  hIF t 
1
2 34
 Mixing Noise (2)
• Defining the equivalent IF noise
 n1i t   cos2  f IF  t   n1q t   sin 2  f IF  t  hIF t 
1
nt   cos2  f LO  t   hIF t  
2
  n2i t   cos2  f IF  t   n2 q t   sin 2  f IF  t  hIF t 
1
2
 1 1  

 2 1i  n t    n2i t    cos 2  f IF  t   
 2  
nt   cos2  f LO  t   hIF t      hIF t 
  1  n t   1  n t   sin 2  f  t  
 2 1i 2
2i

IF



E ni t  
2
 1
2
 1
 
 E n1i t    E n2i t   0
2

2
2 N
2
 
E nq t  
2
 1
2
 
1

 E n1q t    E n2 q t   0
2

2
2 N
2

• But this is the same as quadrature noise
nt   hIF t   ni t   cos2  fIF  t   nq t   sin2  fIF  t  hIF t 

• Mixing doesn’t change the noise power 35

 Mixing Noise to Baseband
• What if we split bandpass noise into two
distinct noise bands, BT/2 above and below
B
the carrier/IF? for W  T
2
nt   n1i t   cos2   f c  f B   t    n1q t   sin 2   f c  f B   t   h1 t 
 n2i t   cos2   f c  f B   t    n2 q t   sin 2   f c  f B   t   h2 t 

• Noise power is related as

 
E nt 
2
carrier 
N0
2
 2  BT  N 0  BT  2  N 0 W


E n1 t 
2

AboveC

 E n2 t 
2

BelowC
N0
2
  2 
BT
2
 N 0 
BT
2
 N 0 W

  
E nt 
2
  
 E n1 t   E n2 t   N 0  BT  2  N 0 W
2 2

• Noise bands get added … 36

 Mixing Noise to Baseband
Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display.
BT N0 BT
(a) General case;
2
(b) symmetric-sideband case;
 fc
 f c    BT
fc
f c    BT (c) suppressed-sideband case: Figure 10.1-3

Not Desired 0    0.5

Mix to Baseband   0.5

W=½ BT
N0BT=2N0W

Mix to IF
  0 or   1
BT
N0BT 37
 Why did we do these derivations?
• The past derivations were all about mixing and filtering.
– Quadrature noise is the noise that gets mixed to the intermediate. The
bandwidth and noise power do not change

 
E nt  
2 N0
2
 2  BT  N 0  BT  2  N 0  W

– Quadrature noise is the noise that gets mixed to baseband. The

bandwidth is halved and the noise power is doubled the LPF
bandwidth standard noise power.

BT
W  
E nt   N 0  BT  2  N 0 W
2

38
 Complex Noise
• Noise in a complex process
nt   ni t   j  nq t 
• Noise power is related as

  
E n t   E n i t   j  n q t  n i t   j  n q t 
2
 H

E n t    E n t   j  n t  n t   j  n t 

2
i q i q

E n t    E n t   j  n t   n t   j  n t   n t   n t  
2 2 2
i i q q i q

     
E n t   E n i t   E n q t 
2 2 2

  E n t    E n t   
2 N 2
E n t 
2 N 0
 0 i q
4
2
MATLAB : n  randn ( m , n )
39
MATLAB : n  randn ( m , n )  i  randn ( m , n )  sqrt 2 
 Noise Envelope and Phase (1)
• Noise as a magnitude and phase
n t   A n t   cos2  f c  t   n t 

n i  A n  cos n  n q  A n  sin  n 

• The magnitude is a Rayleigh distribution

– Mean and moment
A  An 2 
p A A n   n  exp    u A n 

NR  2  NR 
n

EA n  
  NR
2
  2
E An  2  NR

40
 Noise Envelope and Phase (2)
• Probability of An exceeding a value “a”

P An  a   exp  a 
2

 2  N R 
• Phase Distribution
  NR
1 E  An  
p  n   for 0   n  2 2
2
  2
E An  2  N R
• Noise Power

    
E nt   E An t   E cos2  f c  t   n t 
2 2 2

E nt    2 N     N
2 1 N0
R R
2 2

41
 Noise Characteristics
• The noise power does not change based on the
representation, the center frequency, or due to mixing.

• The noise power will change when the bandwidth is

further limited in some way!

42
 CW Communication with Noise
Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display.

Model of a CW communication system with noise: Figure 10.1-1

At 
x t   At   cos2  f c  t  t  vt    cos2  f c  t  t   n t 
L

A t   At  
x c t    cos2  f c  t  t  Pr eDt     cos2  f c  t   t   nt   hR t 
L  L 
43
 Chapter 11

• Baseband Transmission of PAM Symbols

• PAM Symbol Autocorrelation and Power Spectral Density
• Symbol Detection in Noise

ECE 6640 44
 Digital Pulse-Amplitude Modulation
(PAM)
• Also referred to as pulse-code modulation (PCM)
• The amplitude of pulse take on discrete number of
waveforms and/or levels within a pulse period T.
x t    a k  pt  kT 
k

• p(t) takes on many different forms, a rect for example

1 0tT
pt   
0 else

x mT     a k  pmT    kT   a m  p  a m , for 0    T

k

45
 Digital Signaling Rate
• For symbols of period T,
the symbol rate is 1/T=R

• The rate may be in bits-per-second when bits are sent.

A bps rate is usually computed and defined.

• The rate may be in symbols-per-second when symbols are

sent. When there are a defined number of bits-per-symbol,
the rate may be defined in bits-per-second.
– If parity or other non-data bits are sent, the messaging rate and the
signaling rate may differ.

46
 Transmission
Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display.

(a) Baseband transmission system (b) signal-plus-noise waveform: Figure 11.1-2

y t    a k  ~
p t  t d  kT   n t 
k

47
 Transmission
y t    a k  ~
p t  t d  kT   n t 
k

• The digital signal is time delayed

td
• The pulse is “filtered” and/or distorted by the channel
p t   fn pt   hc t 
~

• Recovering or Regenerating the signal may not be trivial

– Signal plus inter-symbol interference (ISI) plus noise

yˆ mT  t d   am   ak  ~
p mT  kT   nmT  t d 
k m

48
 ABC Binary PAM formats

(a) Unipolar RZ & NRZ

(b) Polar RZ & NRZ

(c) Bipolar NRZ

(d) Split-phase Manchester

(e) Polar quaternary NRZ

49
Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display.
 PAM Power Spectral Density:
Polar NRZ

 t  Td  k  Tb 
vt    a k  rect  
k    Tb 
1  
Ea n   0, E a n   2
2

pTd   ,
Tb
0  Td  Tb
 
E a j  a k  0, for j  k

 
Rvv   E vt   vt       1  ,
 2
Tb    Tb
 Tb 
S vv w  E vt   vt     2  Tb  sinc 2  f  Tb 

• For a zero mean, polar NRZ of amplitude +/- A and

symbol duration Tb
S vv w  A2  Tb  sinc 2  f  Tb 
  2
E an   2  A 2

A2 f 
 sinc 2  
rb  rb  50
 PAM Power Spectral Density:
Arbitrary Pulse

 t  Td  k  D  pTd  
1
vt    a  p   , 0  Td  D
k  
k
 D  D

 
E an   ma , E an   a  ma
2 2 2


1
S vv  f    P f    Ra n   exp j  2  f  D 
2

D n  

 a 2  ma 2 , n0 1
Ra n    2 Tb  D, rb 
D
ma , n0
rb is symbol rate
• Using Poisson’s sum formula
2 2 2
a  ma    n   n
S vv  f    P f       P    f  
2

D  D  n    D   D

S vv  f    a  rb  P f   ma  rb    Pn  rb    f  n  rb 
2 2 2 2

n   51
 Power spectrum of
Unipolar, binary RZ signal
Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display.
 
Figure 11.1-5  t 
pt   rect   rect2  rb  t 
 Tb
2
A2  f  A2  n
2

S vv  f    sinc     sinc    f  n  rb   2 
16  rb  2  rb  16 n    2
1  f 
P f    sinc 
2  rb 2
 b r

E an   , E an 
A
2
2 A2
2
 
 2 A2
 a  ma  ,n  0
2

2
Ra n    2
m 2  A , n0
 a 4

52
 Power spectrum of
Unipolar, binary RZ signal
2 2
A2  f  A2  n
S vv  f    sinc     sinc    f  n  rb 
16  rb  2  rb  16 n    2
Unipolar Binary RZ
0.07

• For rb=2
0.06

0.05

0.04

0.03

0.02

0.01

0
-8 -6 -4 -2 0 2 4 6 8
freq (f)

Plots from PSD_PCM.m 53

 Power spectrum of
Unipolar, binary NRZ signal
2
A2  f  A2   t 
S vv  f    sinc     sincn    f  n  rb  pt   rect   rectrb  t 
2

4  rb  rb  4 n    Tb 
2
2
 f  A2
S vv  f     f 
A f
 sinc   P f  
1
 sinc 
4  rb  b
r 4 rb  rb 
Unipolar Binary NRZ

 
0.25
A2
E an   , E an 
A 2

0.2
2 2
• For rb=2  2 A2
 a  ma  ,n  0
0.15 2

2
Ra n    2
0.1
m 2  A , n0
 a 4
0.05

0
-8 -6 -4 -2 0 2 4 6 8
freq (f)
54
 Power spectrum of
Polar, binary RZ signal (+/- A/2)
 
 t 
pt   rect   rect2  rb  t 
2
A2  f 
S vv  f    sinc 
 Tb 
16  rb  2  rb   2 
1  f 
• For rb=2 P f    sinc 
0.035
Polar Binary RZ 2  rb  2  rb 

 
0.03

E an   0, E an  A
2 2

0.025 4
0.02

0.015  2  m 2  A2 , n  0
Ra n    4
a a

0.01
ma2  0, n0
0.005

0
-8 -6 -4 -2 0 2 4 6 8
freq (f)
55
 Power spectrum of
Polar, binary NRZ signal (+/- A/2)
2
f 
2
 t 
S vv  f  
A
 sinc  pt   rect   rectrb  t 
4  rb  rb   Tb 

1 f
• For rb=2 P f    sinc 
0.14
Polar Binary NRZ rb  rb 
0.12

0.1
 
E an   0, E an  A
2 2

0.08

 2  m 2  A2 , n  0
Ra n    4
0.06 a a

0.04 ma2  0, n0

0.02

0
-8 -6 -4 -2 0 2 4 6 8
freq (f)
56
Spectral Attributes of PCM
If Bandwidth W=1/T,
then WT=1

Note that WT=0.5 or a

bandwidth equal to ½ the
symbol rate can be used!

57
 Baseband Binary Receiver
Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display.

Figure 11.2-1

 
yt    a k  pt  kT  h t   n in t   h t 
k 
y t k   ak  n t k 

• Synchronous Time sampling of maximum filter output

58
Regeneration of a unipolar signal
Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display.

(a) signal plus noise (b) S/H output (c) comparator output: Figure 11.2-2

59
 Unipolar NRZ Binary Error
Probability
• Hypothesis Testing using a voltage threshold
– Hypothesis 0
• The conditional probability distribution expected if a 0 was sent
pY  yk | H 0   pY ak  n t k  | ak  0  pY n t k 
pY  yk | H 0   p N  yk 
– Hypothesis 1
• The conditional probability distribution expected if a 1 was sent

pY  yk | H1   pY ak  n t k  | ak  A  pY A  n t k 
pY  yk | H1   p N  yk - A 

60
 Decision Threshold and
Error Probabilities
Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display.

Conditional PDFs Figure 11.2-3

y t k   ak  n t k 

V
Pe1  PY  V    p Y y | H1   dy


Pe 0  PY  V    p Y y | H 0   dy
V

• Use Hypothesis to establish a decision rule

– Use threshold to determine the probability of correctly and
incorrectly detecting the input binary value
61
 Average Error Probability
• Using the two error conditions:
– Detect 1 when 0 sent
– Detect 0 when 1 sent
Perror  PH 0   Pe 0  PH1   Pe1

• Selecting an Optimal Threshold

PH   p V | H   PH   p V
0 Y opt 0 1 Y opt | H1 

• For equally likely binary values

1
Perror   Pe 0  Pe1 
1 2
PH 0   PH1  
pY Vopt | H 0   pY Vopt | H1 
2

62
Threshold regions for conditional PDFs
Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display.

Figure 11.2-4

1 A
P H 0   P H1   Vopt 
2 2
63
 For AWGN
• The pdf is Gaussian
1  y2 
p Y y | H 0   p N y    exp  
2 
2   2
 2 

for 1

 2 
Q x     exp    d
2 x  2

V  A 
Pe 0  PY  V    p N y   dy  Q   Q 
V     2   
V
AV  A 
Pe1  PY  V    p N y  A   dy  Q   Q 
    2   
for 

1 A
P H 0   P H 1   Vopt 
2 2

 A V   A 
Pe1  Q   Q   Pe 2 64
    2  
 Modification for
Polar NRZ Signals (+/- A/2)
• Hypothesis Testing using a voltage threshold
– Hypothesis 0
• The conditional probability distribution expected if a 0 was
sent
 A  A 
pY  yk | H 0   pY  ak  n t k  | ak     pY    n t k 
 2  2 
 A
pY  yk | H 0   p N  yk  
 2
– Hypothesis 1
• The conditional probability distribution expected if a 1 was
sent
 A A 
pY  yk | H1   pY  ak  n t k  | ak    pY   n t k 
 2 2 
 A
pY  yk | H1   p N  yk - 
 2
A A
Vopt    0 65
2 2
 Modification for
Polar NRZ Signals (+/- A/2)
• Determining the error probability
  A V 
 A   Q A 
Pe 0  PY  V    p N  y    dy  Q 2  
 2     2   
V
 

 A V 
 A   Q A 
V
Pe1  PY  V    p N  y    dy  Q 2  
 2     2   

 
• Notice that the error is the same as Unipolar NRZ
– The distance between the expected signal values is the
same
– The “distance” between the expected values determines
the error …
66
 Modification for
Bipolar NRZ Signals (+/- A)
• Hypothesis Testing using a voltage threshold
– Hypothesis 0
• The conditional probability distribution expected if a 0 was
sent
pY  yk | H 0   pY ak  n t k  | ak   A  pY  A  n t k 

pY  yk | H 0   p N  yk  A

– Hypothesis 1
• The conditional probability distribution expected if a 1 was
sent
pY  yk | H1   pY ak  n t k  | ak  A  pY  A  n t k 

pY  yk | H1   p N  yk - A 

Vopt  A  A  0 67
Modification for Bipolar NRZ Signals
• Determining the error probability

AV A
Pe 0  PY  V    p N y  A   dy  Q   Q 
V    
V
AV A
Pe1  PY  V    p N y  A   dy  Q   Q 
    

• Notice that the error has been reduced

– The distance between the expected signal values may
be twice as large as the unipolar case (using +/- A)

68
 Relationship to signal power
• Defining the average received signal power
1
– Unipolar NRZ S R   A2 , 0, A
2
  T2 
 A A  1 
1 S R  E  lim   xc t   dt  
2
– Polar NRZ S R   A2 ,  2 , 2 
  T T 2 
T 
4 

– Bipolar NRZ S R  A2 ,  A, A
• In terms of SNR
1  S 
2 2  N  for Unipolar
 A  A 2
  R
   
 2  4  N R  S 
  for Polar
 N  R
2
A A 2  S 
      for Bipolar

  N R  N  R 69
 Probability of error
• The probability of detecting a transmitted symbol
correctly is dependent upon the received signal-to-
noise ratio …. assuming PH   PH  0
1
1
2
– Unipolar NRZ (orthogonal)
 A   1 S   A 
2
A2 1 S
Pe  Q   Q   
      
 2    2  N R   2   4  NR 2  N R

– Polar NRZ (antipodal)

 A   S   A 
2
A2 S
Pe  Q   Q        
 2     N R   2   4  N R  N R

– Bipolar NRZ (antipodal)

 A  S   A
2
A2  S 
Pe  Q   Q        
    N R    N R  N R
  70
 Power Ratio vs. Bit Energy
• For continuous time signals, power is a normal
way to describe the signal.
• For a discrete symbol, the “power” is 0 but the
energy is non-zero
– Therefore, we would like to describe symbols in terms
of energy not power
• For digital transmissions how to we go from
power to energy?
– Power is energy per time, but we know the time
duration of a bit. Noise has a bandwidth.
1 S SR
S R  Eb  N R  N 0 W    ?
Tb  N R N R 71
 SNR to Eb/No
• For the Signal to Noise Ratio
– SNR relates the average signal power and average noise
power (Tb is bit period, W is filter bandwidth)
1
Eb 
S Tb  E b  1  Eb  R b
  
 
     
 N  N 0  W  N 0  Tb  W  N 0  W

– Eb/No relates the energy per bit to the noise energy

(equal to S/N times a time-bandwidth product)
 Eb   S  W  S 
          Tb  W 
 N0   N  R b  N 
72
 Relationship to Eb/No

• Defining the energy per bit to noise power ratio

for a time-bandwidth product of
Rb 1
W  Tb   Tb 
2 2
2
 A  A2 1 S E 
– Unipolar          b 
 2  4  N R 2  N R  N0 
2
 A  A2 S  2  Eb 
– Polar         
 2  4  N R  N R  N0 

2
– Bipolar A A2  S   2  Eb 
        
 N R  N R  N0 

73
Relationship to Bit Error Probability

• Defining the binary bit error probability

for a time-bandwidth product, assuming
1
PH 0   PH1  
2

A  Eb 
– Unipolar (orthogonal) Perror  Q   Q 

 2   N0 
A  2  Eb 
– Polar (antipodal) Perror  Q   Q 

 2   N0 
A  2  Eb 
Perror  Q   Q 
– Bipolar (antipodal)   N0



74
 Bit Error Rate Plot
Classical Bit Error Rates
0.5
Orthogonal
0.45 Antipodal

0.4

0.35 EbNo=(0:10000)'/1000;

0.3
Bit Error Rate

% Q(x)=0.5*erfc(x/sqrt(2))
0.25
Ortho=0.5*erfc(sqrt(EbNo)/sqrt(2));
0.2 Antipodal=0.5*erfc(sqrt(2*EbNo)/sqrt(2));

0.15
semilogx(EbNo,[Ortho Antipodal])
0.1 ylabel('Bit Error Rate')
xlabel('Eb/No')
0.05 title('Classical Bit Error Rates')
legend('Orthogonal','Antipodal')
0
-3 -2 -1 0 1
10 10 10 10 10
Eb/No

75
BER Performance, Classical Curves
log-log plot
0
Classical Bit Error Rates
10
Orthogonal
-1 Antipodal
10

-2
10
Bit Error Rate

-3
10

-4
10

-5
10

-6
10

-7
10
-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Eb/No

76
 Antipodal and Orthogonal Signals

• Antipodal
– Distance is twice “signal voltage”
– Only works for one-dimensional signals
1
T
 1 for i  j
d  2  Eb zij    si t   s j t   dt  
E 0  1 for i  j

• Orthogonal
– Orthogonal symbol set
– Works for 2 to N dimensional signals
1
T
1 for i  j
d  2  Eb z ij    s i t   s j t   dt  
E 0 0 for i  j
77
 M-ary Signals

• Symbol represents k bits at a time M  2k

– Symbol selected based on k bits
– M waveforms may be transmitted
• Allow for the tradeoff of error probability for
bandwidth efficiency
• Orthogonality of k-bit symbols
– Number of bits that agree=Number of bits that disagree

 sumb   
K N
i
 b   sum b ik  b kj
j
k k
1 for i  j
z ij  k 1 k 1

K 0 for i  j
78
 Example 11.2-1
• Unipolar computer network with
Rb  106 bps N 0  4 10 20 W / Hz  194dB / Hz
– Desired BER is one bit per hour
Pe  1  3 10 10
3600  Rb
• Solve for the signal energy
 A A
Perror  Q   3 10 10 From p. 790  6.2
2
 2 
2
 A  A2 1 S 
      R  for Unipolar
 2  4  NR 2  NR 

2  R 
2  6.2   N R  2  6.2   N 0  b   S R
2

 2

 
S R  2  38.44  4 10 20  0.5 106 W  1.54 10 12W 79
 Exercise 11.2-1 (1)
• Unipolar system with equally likely digits and
SNR = 50
• Calculate the error probabilities when the
threshold is set to V=0.4 x A
2
 A  1 S 
     R  for Unipolar
 2  2  NR 

 A 1
   2  50  10

  2

 A V 
V
Pe1  PY  V    N
p  y  A   dy  Q  
  

V 
Pe 0  PY  V    p N  y   dy  Q 
V  80
 Exercise 11.2-1
• Calculate the error probabilities when the
threshold is set to V=0.4 x A PH   PH  
1
0 1
V  2
Pe 0  PY  V   Q   Q0.4 10

Pe 0  Q4.0  3.5 10 5 V=0.5 x A
 A V 
Pe1  PY  V   Q   Q0.6 10 Pe 0  Pe1  Q0.5 10  Q5.0
  
Pe1  Q6.0  1.5 10 9
Pe 0  Pe1  Perror  3.5 10 7
1
Perror   Pe 0  Pe1 
2

Perror   3.5 10 5  1.5 10 9   1.75 10 5

1
2
81
 Gaussian Distribution
The Gaussian probability density function (pdf)
The Gaussian or Normal probability density function is defined as:

f X x  
1
 exp
 
 x X 2 
, for    x  
2    2 
 2  

where X is the mean and  is the variance

The Gaussian Probability Distribution Function (PDF)

 
Gaussian PDF and pdf
x
1  v X 2  1
FX x  
 2  
 exp

 2 
2
  dv


0.9
v  
0.8

The PDF can not be represented in a closed form solution! 0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
-8 -6 -4 -2 0 2 4 6 8

ECE 6640 82
 Gaussian Distribution
The Gaussian Probability Distribution Function is
x
1 
 v X 2  
FX x  
 2  
 exp

 2 
2
  dv


v  

The PDF can not be represented in a closed form solution!

The PDF is tabulated for a zero mean, unit variance pdf.

For these values, it is often described as “normalized” and is defined as
x
1  u2 
x  
2

 exp 
 2 

  du

u  

The distribution function is then defined as

x X 
FX  x    

  

When using Appendix D, the negative values in x are derived as

  x   1   x 

ECE 6640 83
 Q Function
Another defined function that is related to the Gaussian (and used) is the Q-function.:

Qx  
1


 u2 
exp 
 2 
  du Q Function Table p. 858
2  
ux

The Q-function is the complement of the normal function, :

Q x   1    x 

Therefore not that:

Q x   1  Q x 

x X 
FX  x   1  Q 

  

84
 Using MATLAB
Another way to find values for the Gaussian
The error function

 exp u  du
x
2
erf  x    2

u 0

1   x 
Q x    1  erf  
2   2 

1   x X  1 1  xX 
FX x   1   1  erf      erf 
  2 


2   2   2 2  

The error function (Y = ERF(X)) is built-in to MATLAB. .

From MATLAB:
ERF Error function.
Y = ERF(X) is the error function for each element of X. X must be
real. The error function is defined as:

erf(x) = 2/sqrt(pi) * integral from 0 to x of exp(-t^2) dt.

See also erfc, erfcx, erfinv.

Reference page in Help browser

doc erf

ECE 6640 85
 Using MATLAB (2)
The complementary error function
erfcx   1  erf x 

1  x 
Q x    erfc 
2  2

The error function (Y = ERFC(X)) is built-in to MATLAB. .

From MATLAB:
ERFC Complementary error function.
Y = ERFC(X) is the complementary error function for each element
of X. X must be real. The complementary error function is
defined as:

erfc(x) = 2/sqrt(pi) * integral from x to inf of exp(-t^2) dt.

= 1 - erf(x).

Class support for input X:

float: double, single

See also erf, erfcx, erfinv.

Reference page in Help browser

doc erfc

ECE 6640 86
 Qfn and Qfninv
• These function are now in the Misc_Matlab zip file on the
web site

function [Qout]=Qfn(x)
% Qfn(x) = 0.5 * erfc(x/sqrt(2));
Qout = 0.5 * erfc(x/sqrt(2));

function [x]=Qfninv(Pe)
% For Qfn(x) = 0.5 * erfc(x/sqrt(2));
% The inverse can be found as
x=sqrt(2)*erfcinv(2*Pe);

87
 Properties of Matched Filter
• See ECE3800 Notes
– Review from Chapter 9
• Wikipedia
– http://en.wikipedia.org/wiki/Matched_filter
– “The matched filter is the optimal linear filter for maximizing the
signal to noise ratio (SNR) in the presence of additive stochastic
noise.”

88
 Signal to Noise Ratio Definition

• The filtered response becomes




so t   no t   h   s t     nt     d
0

• The SNR
PSignal    Es t  
E so t 2 2

E n t   N  B
SNRout   o
PNoise 2
o o EQ

   
2
  
 0

E  h   st     d  
 
  
SNRout 

1

N o   ht 2  dt
2
0
ECE 6640 89
 Optimized Matched Filter

• The matched filter is

h   K  s T     u  

• The resulting SNR

 T  
2

E   K  s T     s t     d  
 0   K 2  Rss t  T  2  Rss 0 
2
SNRout   
T
1 2
1
 
N o    K  s T    dt 2
N o   K  Rss 0  No
2 0 2

• Or for symbol energy

2  Es
SNRout
No

   S f 

Rss 0   E s t  
2 2
 df


ECE 6640 90
 Autocorrelation versus Integration

• For a matched filter, the integral of the matched filter and

the autocorrelation of the signal are approximately
equivalent
– For signals of limited time duration it can be exact!
• The condition is related to
 T
Rss t  T    K  s T     s t     d
0

Rss 0   Rss t  T 
t T

• But notice that “integrating” and filtering may produce

very different results except at t=T.
– the integral is monotonically increasing
ECE 6640
– the filter is not! 91
 Correlation Receivers

• The concept that a matched filter is performing an

autocorrelation has resulted in the concept of the
Correlation Receiver.
• The symbol of interest is auto-correlated.
• All other symbols are cross-correlated.

• It is assumed that the two outputs are different enough to

allow symbol detection
– Noise power is present based on the “equivalent noise bandwidth”
of the “matched” symbol receivers.

ECE 6640 92
 Corr. vs Int. for an RF/IF Envelope

ECE 6640 93
 Question?

• Are there some better shapes than others for signals and
their matched filters?
– Easy to generate.
– Frequency band limited.
– Finite time duration.
– Minimize inter-symbol interference.

• Start with rects or pulses

• More advanced use raised cosine or Nyquist filter,
or for matched transmit and receive filters
use “square root Nyquist filters”.

ECE 6640 94
Defining a Shape or Filter for Pulses
Section 3.3
y t    ak  pt  t d  k  T 
k

• We want to minimize or zero inter-symbol interference

(ISI)
1 t 0
pt   
0 t  T ,2T , 

• We want a frequency band limited filter

P f   0 B f
r r 1
where B   , with 0    and B  BT r
2 2 T
– Allowable signal rates with  as the excess bandwidth

r  2  B  , for B  r  2  B
95
 Defining a Filter for Pulses
• Possible solutions
1 t 0 pt   p t   sincr  t 
pt   
0 t  T ,2T , 

P f   0 B f  p t   P  f   0 for   f


p 0   P  f  df
 1


• Therefore we select
1
1  f  r
P f   P  f     rect  T
r  r 

These are considered the Nyquist conditions for the filter

96
 Cosine Spectral Shaping
Raised cosine pulse.
(a) Waveform
From Chap 2
(b) Derivatives Raised cosine
(c) Amplitude pulse
spectrum
Figure 2.5-7

• A candidate filter is (with

with  as the excess BW)

  2  f   f 
P  f    cos   rect 
4  4   2   

97
 Convolving
• Raised Cosine Convolution with Bandlimited Spectrum
1  f 
P f   P  f     rect 
r  r 
1 r
r f  
2

1  2   r
P f     cos 2 
r r
  f        f  
r  4   2  2 2
 r
0 f  
 2
r r
where B   , with 0    and B  BT
2 2

• Transforming to the time domain filter

cos2    t 
pt    sincr  t 
1  4    t 
2
98
Nyquist/Raised Cosine Pulse Shaping
http://en.wikipedia.org/wiki/Raised-cosine_filter

 ABC

r
2
1
r
T

GNU FDL:Oli Filth, Raised Cosine Filter Response , en.wikipedia.org, 3 November 2005, Oli Filth

 ABC

r
2
1
r
T

GNU FDL:Oli Filth, Raised Cosine Filter, Impulse Response, en.wikipedia.org, 3 November 2005, Oli Filth 99
 Nyquist Filter (discrete raised cosine)
 n 
cos     
cos2    t   M  n 
pt    sincr  t  pn    sinc  , for k  M  n  k  M
1  4    t 
2 2
 n  M 
1   2   
 M
% function hnyq=nyquistfilt(alpha,M)
2    r
% or
r
0  0  1 % function hnyq=nyquistfilt(alpha,fsymbol,fsample,k)
2 %
% alpha roll-off
cos    r  t 
pt    sincr  t  % fsample rate
1  2    r  t 
2
% fsymbol rate
% M = fsample/fsymbol (an integer value)
% k is 1/2 the number of symbols in the filter
n fs % The filter length is euqal to 2*ceil(k*M)+1
t r
fs M %
% A discrete time cosine taperd Nyquist filter
n % Based on frederic harris, Multirate Signal Processing for
r t  Communications
M
% Prentice-Hall, PTR, 2004. p. 89 100
 MATLAB Raised Cosine Filters (1)
• Rcosine (obsolete)
– [NUM, DEN] = RCOSINE(Fd, Fs, ‘fir’, R)
– FIR raised cosine filter to filter a digital signal with the digital
transfer sampling frequency Fd. The filter sampling frequency is
Fs. Fs/Fd must be a positive integer. R specifies the rolloff factor
which is a real number in the range [0, 1].
• rcosfir (obsolete)
– B = RCOSFIR(R, N_T, RATE, T)
– Raised cosine FIR filter. T is the input signal sampling period, in
seconds. RATE is the oversampling rate for the filter (or the
number of output samples per input sample). The rolloff factor, R,
determines the width of the transition band. N_T is a scalar or a
vector of length 2. If N_T is specified as a scalar, then the filter
length is 2 * N_T + 1 input samples.
101
 MATLAB Raised Cosine Filters (2)
• firrcos (also obsolete)
– B=firrcos(N,Fc,DF,Fs)
– Returns an order N low pass linear phase FIR filter with a raised
cosine transition band. The filter has cutoff frequency Fc,
sampling frequency Fs and transition bandwidth DF (all in Hz).
– The order of the filter, N, must be even.
– Fc +/- DF/2 must be in the range [0,Fs/2]
– The coefficients of B are normalized so that the nominal passband
gain is always equal to one.
– B=firrcos(N,Fc,R,Fs,'rolloff') interprets the third argument, R, as
the rolloff factor instead of as a transition bandwidth.
– R must be in the range [0,1]

102
MATLAB Raised Cosine Filters (3)
• firrcos
– B = rcosdesign(BETA, SPAN, SPS)
– Returns square root raised cosine FIR filter coefficients, B, with a
rolloff factor of BETA. The filter is truncated to SPAN symbols
and each symbol is represented by SPS samples. rcosdesign
designs a symmetric filter. Therefore, the filter order, which is
SPS*SPAN, must be even. The filter energy is one.
– Beta [0,1]
– SPS number os samples per symbol
– SPAN length of filter in number of symbols

103
 Textbook Waveform Energy

• Waveform Energy
T
Ei   si t   dt
2

• Matched Filter
t
z t   r t   ht    r    ht     d


ht   u t   s * T  t 
t
z t    s   s * T  t     d
0
T T T
z T    s   s T  T     d   s   s    d   s    d
* * 2

0 0 0

ECE 6640 Correlation 104

 Optimum binary detection
Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display.

(a) parallel matched filters (b) correlation detector: Figure 14.2-3

105
 Symbols and Matched Filters
y (t )  si (t ) h opt (t )
si (t ) h opt (t ) A2
A A
T T

T t T t 0 T 2T t

y (t )  si (t ) h opt (t )
si (t ) h opt (t ) A2
A A
T T

T/2 T t T/2 T t 0 T/2 T 3T/22T t


2
A A A
2
T T
2006-01-31 Lecture 3 106
 Optimized Error Performance

• Maximize the “distance”, Ed, between the sampled values

that are used for detection.
– The distance is based on the correlation of the symbols
T
Ed   s1 t   s2 t   dt
2

T T T
Ed   s1 t   dt  2   si t   s j t   dt   s2 t   dt
2 * 2

0 0 0

T T
Eb   si t   dt   Eb   si t   s j * t   dt
2

0 0

Ed  Eb  2   Eb  Eb  Eb  1   
ECE 6640 107
 Optimized Error Performance

• To maximize the distance:  = -1 Ed  2  Eb

– Symbols are said to be antipodal
– Examples: +/- 1 Symbols (Bipolar), BPSK
– Not always achievable
• Useful performance:  = 0 Ed  Eb
– Symbols are said to be orthogonal
– Examples: On-Off Keying (ASK), FSK, “independent symbols”

ECE 6640 108

 Antipodal and Orthogonal Signals

• Antipodal
– Distance is twice “signal voltage”
– Only works for one-dimensional signals
1
T
 1 for i  j
d  2  Eb zij    si t   s j t   dt  
E 0  1 for i  j
• Orthogonal
– Orthogonal symbol set
– Works for 2 to N dimensional signals

1
T
1 for i  j
d  2  Eb z ij    s i t   s j t   dt  
E 0 0 for i  j

109
Relationship to Bit Error Probability

• Defining the binary bit error probability

for a time-bandwidth product PH   PH  
0 1
1
2

– Orthogonal  Eb 
Perror  Q 

 N0 

– Antipodal  2  Eb 
Perror  Q 

 N0 

110
 Bit Error Rate Plot-Linear BER
Classical Bit Error Rates
0.5
Orthogonal
0.45 Antipodal

0.4

0.35 EbNo=(0:10000)'/1000;

0.3
Bit Error Rate

% Q(x)=0.5*erfc(x/sqrt(2))
0.25
Ortho=0.5*erfc(sqrt(EbNo)/sqrt(2));
0.2 Antipodal=0.5*erfc(sqrt(2*EbNo)/sqrt(2));

0.15
semilogx(EbNo,[Ortho Antipodal])
0.1 ylabel('Bit Error Rate')
xlabel('Eb/No')
0.05 title('Classical Bit Error Rates')
legend('Orthogonal','Antipodal')
0
-3 -2 -1 0 1
10 10 10 10 10
Eb/No

ECE 6640 111

 BER Performance Fig. 3.14
0
Classical Bit Error Rates
10
Orthogonal
-1 Antipodal
10

 Eb 
 Q 
-2
10 Perror 
 N0 
Bit Error Rate

-3
10

-4
10

 2  Eb 
-5
10
Perror  Q 

10
-6
 N0 
-7
10
-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Eb/No

ECE 6640 112

 Equalization

• An advanced topic that I skip at this time.

• Narrowband communications may go through a magnitude

and phase change due to “the channel”
• Wideband communications likely experiences channel
effects that may be non-linear across the signal frequency
band. To correctly detect the information, an inverse
channel filter or “equalizer” is used.
– The channel is usually not predictable. Therefore, the equalizer
must “learn” from the transmitted signal how to correct for channel
effects. There are “adaptive” algorithms that can do this and are
taught in a more advanced course.
ECE 6640 113