ECE 6640

Digital Communications

Dr. Bradley J. Bazuin

Assistant Professor
Department of Electrical and Computer Engineering
College of Engineering and Applied Sciences
 Chapter 9
9. Modulation and Coding Trade-Offs.
1. Goals of the Communications System Designer.
2. Error Probability Plane.
3. Nyquist Minimum Bandwidth.
4. Shannon-Hartley Capacity Theorem.
5. Bandwidth Efficiency Plane.
6. Modulation and Coding Trade-Offs.
7. Defining, Designing, and Evaluating Systems.
8. Bandwidth-Efficient Modulations.
9. Modulation and Coding for Bandlimited Channels.
10. Trellis-Coded Modulation.

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.
 System Level Tradeoffs

• The Nyquist theoretical minimum bandwidth requirements

• The Shannon-Hartley capacity theorem
– The Shannon limit
• Government regulatory involvement
– frequency allocation, bandwidth limitations
• Technology limitations
– physically realizable components using current technology
• Other system requirements
– For satellite: orbits and energy limitations

ECE 6640 4
 Error Probability Plane

• Error probability performance curves

– define acceptable BER
– determine required Eb/No
• We would prefer equivalent bandwidth performance
curves
– allows system level tradeoffs
– trade-off Eb/No for modulation type at fixed BER
– trade off BER vs modulation type at fixed Eb/No
– show range of expected BER as Eb/No varies

ECE 6640 5
 BER vs Eb/No Curves

ECE 6640 6
 Nyquist Minimum Bandwidth

• Nyquist showed that the theoretical minimum bandwidth

needed for baseband transmission of Rs symbols per
second without ISI is Rs/2 Hz.
– A theoretical minimum constraint on bandwidth required.
– Referred to as 2 symbols/sec/Hz
– Typical systems and filters are 10%-40% wider
– More likely 1.8 to ¼ symbols/s/Hz.
• Rs in terms of M symbol modulation
R R
R  k  Rs Rs  
k log 2 M

ECE 6640 7
 Example 9.1: Digital Schemes

• Orthogonal Signaling
– expect improvement in BER as k or M increases
• Non-orthogonal signaling
– expect a decrease in BER as k or M increases

a) Does error-performance improve or degrade with

increasing M, for M-ary signaling?
b) The choices available almost always involve a tread-off.
If error performance improves, what price must we pay?
c) If error-performance degrades, what benefit is exhibited?

ECE 6640 8
 Example 9.1

• Expected trade-offs
• M-FSK
– as M increases, the required transmission bandwidth increases for
minimum frequency spacing.
– to maintain a constant bit rate, the symbol transmission rate
decreases with increasing M
• M-PSK
– while there is degradation as M increases, the symbol transmission
rate may be decreased as M increases
– M-PSK systems plot equal-bandwidth curves, as the bit
transmission rate increases.

ECE 6640 9
 Shannon-Hartley Capacity Theorem

• The capacity relation in

AWGN can be stated as
 S
C  W  log 2 1  
 N

– where S is the signal power,

N the noise power, and W the
bandwidth
– the value is defined in bits per
second

ECE 6640 10
 Shannon-Hartley Capacity Theorem

• The normalized channel

bandwidth vs. SNR may
also be plotted

C  S
 log 2 1  
W  N

1
W   S 
 log 2 1  
C   N 

ECE 6640 11
 S-H Equivalent Equations

• Rearranging and defining the noise power and signal power

C  S C  E R 
 log 2 1    log 2 1  b  b 
W  N W  N0 W 

• For
C
Eb S 1 Rb Eb Rb
  2  1
W

N0 N 1 W N0 W

• Letting C = Rb

Eb W  W 
C C
E C
2  1 b 
W
   2  1
N0 W N0 C  

C  Eb C 
 log 
2 1   
ECE 6640 W  N0 W  12
 Shannon Capacity Theorem

• There is a limiting case as C/W  0

– let Eb  C 
x  
N0  W 
C  E C Eb N W  E C
 log 2 1  b   1   0    log 2 1  b  
W  N0 W  N0  Eb C   N0 W 

E 1
1  b   log 2 1  x  1
Eb
N0

 log 2 1  x  x
1

N0 x

1  lim
x 0
Eb
N0
 1

 log 2 1  x  x  b  log 2 e
E
N0

Eb 1
  0.693  1.6dB
N 0 log 2 e
ECE 6640 13
 Shannon Limit
Eb 1
  0.693  1.59dB
N 0 log 2 e

• As C/W  0 or W/C∞
• In practice, it is not possible to
reach the bound.
• Provides an improvement bound
for encoding and decoding.
• For example: raw BPSK requires
approximately 9.6 dB Eb/No to
achieve a BER of 10-5 which
suggests that up to an 11.2 dB
improvement is possible.
– Turbu Codes can achieve ~ 10 dB.

ECE 6640 14
 Entropy

• To compute communication capacity, a metric for the

message content of a system is also important.
• Entropy is defined as the average amount of information
per source output.
• It is expressed by: n
H   pi  log 2  pi 
i 1
– where pi is the probability of the ith output and
the sum of all pi is 1.
• For a binary system, entropy can be expressed as:

H   p  log 2  p   1  p   log 2 1  p 

ECE 6640 15
 Entropy for a Binary System
• The entropy is based on the
probability, p, of an event.
• This can also be looked at as
the randomness of successive
events or how correlated
individual events are.
• Note that maximum entropy is
achieved when the probability
is 50%
– A sample provides no information
about a succeeding sample.

ECE 6640 16
 Example 9.2 English Language

• The English language is highly redundant.

– The probability of the next letter in a word is not equally likely for
all possible characters.
– Determine the Entropy based on the letter probabilities
– p=0.10 for the letters a, e, o, t n
– p=0.07 for the letters h, I, n, r, s H   pi  log 2  pi 
i 1
– p=0.02 for the letters c, d, f, l, m, p, u, y
– p=0.01 for the letters b, g, j, k, q, v, w, x, z

H  4  0.1  log 2 0.1  5  0.07  log 2 0.07  1  1 

H  26    log 2    log 2 26
 8  0.02  log 2 0.02   9  0.01  log 2 0.01  26  26 
 4.17 bits/char  4.70 bits/char

English Language Equal Probability

ECE 6640 17
 Equivocation

• A term used by Shannon to account for the uncertainty in a

received signal. It is defined as the conditional entropy of
the message X (transmitted source message), given Y (the
received signal).
H  X | Y    P X , Y   log 2 P X | Y 
X ,Y
– based on conditional probability
H  X | Y    PY    P X | Y   log 2 P X | Y 
Y X

ECE 6640 18
 Equivocation Example

• Consider a binary sequence, X, where the bits are equally

likely. Assume that the channel produces on error in a
received sequence of 100 bit (Pb=0.01).
H  X | Y    P X , Y   log 2 P X | Y 
X ,Y

H  X | Y   1  Pb   log 2 1  Pb   Pb  log 2 Pb 

H  X | Y   0.99  log 2 0.99  0.01  log 2 0.01

H  X | Y   0.081

• Interpretation: the channel introduces 0.081 bit/received

symbol of uncertainty.

ECE 6640 19
 Effective Transmission Rate

• Using the equivocation computation, the effective

transmission rate of the channel can be computed as
H eff  H  X   H  X | Y 

– based on the previous example, the binary system would have an

effective transmission rate (in terms of bit/received symbol) of

H eff  1  0.081  0.919

– for a communication system with R = 1000 bits/sec,

the effective transmission rate would become
Reff  R  H eff  1000  0.919  919

ECE 6640 20
 Pb vs Eb/No Curves

• It appears that Pb approaches 0.5 as Eb/No decreases …

but the Shannon limits is Eb/No=-1.6 dB.
Is this a contradiction or not?
• Shannon refers to received information bits based on
equivocations.

ECE 6640 21
 Deriving an Effective Eb/No

• As an example, take Eb/No=-10 dB for coherent BPSK


PB  Q 2  Eb N 0 
PB  Q0.447   0.33
H  X | Y   1  0.33  log 2 1  0.33  0.33  log 2 0.33  0.915

H eff  1  0.915  0.085

– from this form an effective Eb/No

 Eb  E N 0.1
   b 0   1.176  0.7dB
 N 0  eff H eff 0.085

– Thus, he effective Eb/No is well above the Shannon limit, -1.6dB

ECE 6640 22
 Bandwidth-Efficieny Plane

• Using Shannon-Hartley Capacity, the “normalized”

channel bandwidth versus Eb/No for different symbol
schemes can be compared.
– Typically performed for a defined bit-error probability and under
optimal symbol detection assumptions.
– Let R=C, then
R  E R
 log 2 1  b  
W  N0 W 
– The bounds and appropriate values for MPSK, MFSK and MQAM
symbol schemes are shown on Fig. 9.6

ECE 6640 23
 Figure 9.6:
Bandwidth-Efficiency Plane
• Factors of note:
– MPSK and QAM nominally
maintain the same bandwidth
will increasing the bits per
symbol and required Eb/No
– MFSK uses an increasing
bandwidth as the bits per
symbol increases while the
Eb/No is decreasing
– BPSK and QPSK have the
same Eb/No but different bits
per symbol

ECE 6640 24
 Bit and Symbol Rate Considerations

• For MPSK
R  k  Rs  log 2 M   Rs log 2 M   Rs
 log 2 M 
R
1 
WIF   Rs WIF Rs
Ts
– R/W increases with M

• For MFSK
R  k  Rs  log 2 M   Rs R log 2 M   Rs log 2 M 
M  
WIF   M  Rs WIF M  Rs M
Ts
– R/W decreases with M

ECE 6640 25
 Bandwidth versus Power

• For a bandwidth-limited system

– spectral efficiency is important
– expect that signal power may be increases to offset the limitation
– study the bandwidth-efficient plane
– PSK allows for fixed bandwidths
• For a power-limited system
– a defined transmission power limit has been established
– expect that signal bandwidth may increase to offset the limit
– study the bit-error probability planes
– FSK allows for limited spectral power

ECE 6640 26
 Digital Comm. System Engineering

• Defining, designing, and evaluating communication

systems.
• Comparing MPSK and MFSK (table 9.1)
MPSK Non‐Coherent MFSK
M k R Rs min W R/W Eb/No (dB) min W R/W Eb/No (dB)
bits/sec sym/sec (Hz) Pb=1e‐5 (Hz) Pb=1e‐5
2 1 9600 9600 9600 1 9.6 19200 0.5 13.4
4 2 9600 4800 4800 2 9.6 19200 0.5 10.6
8 3 9600 3200 3200 3 13.0 25600 0.375 9.1
16 4 9600 2400 2400 4 17.5 38400 0.25 8.1
32 5 9600 1920 1920 5 22.4 61440 0.15625 7.4

ECE 6640 27
 System Example #1:
Bandwidth Limited
• W = 4000 Hz, Pr/No=53 dB-Hz, R=9600 bps, PB=1e-5

• Equations needed for the computations (assuming M-PSK)

Pr E E
 b  R  s  Rs
N0 N0 N0

 log 2 M 
Es Eb

N0 N0

   
PE M   2  Q 2  s  sin 
E

 N0 M 

PE M 
PB 
ECE 6640 log 2 M  28
 System Example #1:
Bandwidth Limited
• W = 4000 Hz, Pr/No=53 dB-Hz, R=9600 bps, PB=1e-5

 log 2 M 
Pr E E Es Eb
 b  R  s  Rs 
W 4000 Hz
N0 N0 N0 N0 N0
Pr/No 53 dB‐Hz

PE M 
R 9600 bps
    PB 
PE M   2  Q 2  s  sin 
Pb 1.00E‐05 BER E
 log 2 M 
Pr/No 199526.23 Hz  N0 M 

Eb/No 20.78
Eb/No 13.18 dB

M‐PSK Rs lin dB sqrt(2*Es/No) sin(pi/M) x Q(x)=Pe Pb

2 9600 sym/s Es/No 20.78 13.18 6.45 1.00 6.45 1.14E‐10 1.14E‐10
4 4800 sym/s Es/No 41.57 16.19 9.12 0.71 6.45 1.14E‐10 5.69E‐11
8 3200 sym/s Es/No 62.35 17.95 11.17 0.38 4.27 1.92E‐05 6.42E‐06
16 2400 sym/s Es/No 83.14 19.20 12.89 0.20 2.52 1.19E‐02 2.97E‐03

ECE 6640 29
 System Example #2:
Power Limited
• W = 45 kHz, Pr/No=48 dB-Hz, R=9600 bps, PB=1e-5

• Equations needed for the computations (assuming M-FSK)

Pr E E
 b  R  s  Rs
N0 N0 N0

 log 2 M 
Es Eb

N0 N0

M 1  1 E 
PE M    exp   s 
2  2 N0 

2 k 1
PB  PE M   k
2 1
ECE 6640 30
 System Example #2:
Power Limited
• W = 45 kHz, Pr/No=48 dB-Hz, R=9600 bps, PB=1e-5

Pr E E
 log 2 M 
 b  R  s  Rs Es Eb

W 45000 Hz N0 N0 N0 N0 N0
Pr/No 48 dB‐Hz
R 9600 bps
Pb 1.00E‐05 BER
M 1  1 Es  2 k 1
Pr/No 63095.73 Hz  
PE M   exp    PB  PE M   k
2  2 N0  2 1
Eb/No 6.57
Eb/No 8.18 dB

M‐FSK k Rs Ws lin dB exp(‐Es/No/2) PE Pb

2 1 9600 sym/s 19200 Hz Es/No 6.57 8.18 0.04 1.87E‐02 1.87E‐02
4 2 4800 sym/s 19200 Hz Es/No 13.14 11.19 0.00 2.10E‐03 1.40E‐03
8 3 3200 sym/s 25600 Hz Es/No 19.72 12.95 0.00 1.83E‐04 1.05E‐04
16 4 2400 sym/s 38400 Hz Es/No 26.29 14.20 0.00 1.47E‐05 7.82E‐06
32 5 1920 sym/s 61440 Hz Es/No 32.86 15.17 0.00 1.13E‐06 5.85E‐07

ECE 6640 31
 Coded System Example

• When the previous methods do not produce a valid

implementation, encoding and decoding will be required.
– Monitor the effect of code rates on symbols/sec and bandwidths

ECE 6640 32
 System Example #3:
Encode-Decode
• W = 4000 Hz, Pr/No=53 dB-Hz, R=9600 bps, PB=1e-9
• Starting with the previous 8-PSK system, we need
additional coding gain
 R  log 2 M   Rs
n
Pr E E E Rc 
 b  R  c  Rc  s  Rs k
N0 N0 N0 N0

E k
 log 2 M   b     log 2 M 
E s Ec

N0 N0 N0  n 

   
PE M   2  Q 2  s  sin 
E

 N0 M 
1 n n j
PB    j     Pc  1  Pc 
n j

PE M  n j t 1  j 
PC 
ECE 6640 log 2 M  33
 Solution is Steps
• Step 1: Compute the Es/No
E k
 log 2 M   b     log 2 M 
E s Ec
Pr E E E
 b  R  c  Rc  s  Rs 
N0 N0 N0 N0 N0 N0 N0  n 

• Step 2: Compute the codeword symbol error rate PE(M)

   
PE M   2  Q 2  s  sin 
E

 N0 M 
• Step 3: Compute the codeword-bit-error rate
PE M 
PC 
log 2 M 
• Step 4: Compute the decoded bit error probability
1 n n j
PB    j     Pc  1  Pc 
n j

n j t 1  j 

ECE 6640 34
 Excel Computations

• An excel spreadsheet can be used for all of the examples.

• see results for Example #3

• Alternate Approach
– the coding gain formula can be used.
E   
G in dB    b  in dB    Eb  in dB 
 N 0 uncoded  N 0  coded

G in dB   16  13.2  2.8

– an encoding scheme that meets the bandwidth requirement and has

2.8 dB or more coding gain is sufficient for solving this problem.
ECE 6640 35
 Bandwidth Efficient Modulations

• Modern communication is hungry for bandwidth,

demanding an every increasing communications capacity
within the fixed frequ3ency bands available,
• Additional requirements to allow for non-linear
amplification put a premium on using signals that are
minimally effected by AM to PM conversion, limiting the
amplitude variations of the signal (desiring a constant
modulus).

ECE 6640 36
 QPSK and Offset QPSK

• Conventional QPSK uses

consecutive bits received to
determine I-Q pairs for transmission.
• Offset QPSK also uses the bits, but
directs them to the I and Q ports as
they arrive in time (next slide)

ECE 6640 37
 QPSK versus Offset QPSK

• OQPSK makes 90
degree phase transitions
• 180 degrees phase
changes may result in
significant amplitude
variation

ECE 6640 38
 Minimum Shift Keying (MSK)

• Avoiding discontinuous phase transitions of the signal

– maintain a constant amplitude
– use a form of continuous-phase FSK
– also a modified form of OQPSK

  d  
st   cos 2     f 0  k   t  xk , k  T  t  k  1  T
  4 T  
  k  
xk  mod  xk 1   d k 1  d k ,2   
 2  

ECE 6640 39
 MSK Quadrature Representation

• Expanding the cosine term cos(a+b)

  t 
st   ak  cos   cos2    f 0  t 
 2 T 
  t 
 bk  sin   sin 2    f 0  t 
 2 T 
ak  cosxk   1
bk  d k  cos xk   1

  k  
xk  mod  xk 1   d k 1  d k ,2   
 2  

– the similarity to OQPSK is based on the

amplitude weighted quadrature
structure of this formulation
ECE 6640 40
 Bandwidth Comparison:
BPSK, QPSK & OQPSK, & MSK

ECE 6640 41
 Modulation and Coding for
Bandlimited Channels
• Research Areas (as of 2001 copyright):
– Optimum signal constellation boundaries (choosing a closely
packed signal subset from any regular array or lattice of candidate
points)
– Higher density lattice structures (adding improvement to the signal
subset choice by starting with the densest possible lattice for the
space)
– Trellis-coded modulation (combined modulation and coding
techniques for obtaining coding gain for bandlimited channels).
• Ungerboeck Partitioning

ECE 6640 42
 Evolution of Telephone
Modem Standards (1)
• Telephone modems have dealt with the limited power and bandwidth
problem for a considerable time.
• Progress was made at different times for both leased-lines and dial-line
services.

ECE 6640 43
 Evolution of Telephone
Modem Standards (2)
• Home modem standards
– Mostly replaced by telephony DSL or cable TV access

ECE 6640 44
 Signal Constellation Boundaries
• Various QAM constellations that
have been investigated.
– optimal packing of points with
maximum separation
– reduce maximum amplitude
– optimize PE(M)

ECE 6640 45
 Trellis-Coded Modulation (TCM)

• Developing combined modulation and coding schemes

• Use a redundant nonbinary modulation in combination
with a finite-state machine based encoding process.
– FSM could be similar to convolutional encoding
– A multi-level/phase modulation scheme

• The concept, when performing MATLAB simulations of

encoded bit streams using MPSK or QAM symbols, is
there an optimal combination?
– if you know the symbols being used, could one convolutional code
leading to an appropriate trellis decoding perform better than
another?
ECE 6640 46
 TCM Encoding

• Ungerboeck, G., "Channel coding with multilevel/phase

signals," Information Theory, IEEE Transactions on,
vol.28, no.1, pp.55,67, Jan 1982.
• Initial paper describing trellis coded, soft decision
encoding and modulation technique for communications.

ECE 6640 47