Telecommunications Engineering

Dr. David Tay

Room BG434
x 2529
[email protected]

DT
Bandpass Modulation / Demodulation
• Digital modulation transforms digital symbols into wave-
form compatible with channel.

• Baseband modulation: waveforms are pulse shapes like

the raised cosine or sinc.

• Bandpass modulation: use baseband waveform to mod-

ulate a sinusoidal carrier so that spectrum is shifted to
high-frequency (RF).

Purpose:

1. Reduce the size of antenna which depends on wavelength

(frequency).

2. Slot different users into different frequency band - fre-

quency division multiplexing.

3. Minimize the effect of interference - spread spectrum.

4. Some channel require high frequency such as the iono-

sphere.

Bandpass modulation techniques:

• The process involve varying the amplitude, frequency or

phase of an RF carrier in accordance with the informa-
tion to be transmitted.

DT
 • General form of a carrier wave:

s(t) = A(t) cos[ω0 t + θ(t)]

where A(t) is the amplitude, ω0 is the frequency and θ(t)

is the phase.

• Types of detection

1. Coherent detection - uses the phase reference infor-

mation.
2. Noncoherent detection - does not utilize phase infor-
mation.

Noncoherent system are easier to implement but gener-

ally do not perform as well as coherent ones.

Phasor representation of carrier wave

ejω0 t = cos(ω0 t) + j sin(ω0 t)

1. cos(ω0 t) is the in-phase (I) component.

2. sin(ω0 t) is the quadrature (Q) component.

DT
Graphical representation:

Common digital modulation format

PSK, FSK, ASK and hybrid ASK/PSK.
In general M = 2k waveform types are used to carry k bits
in each waveform.

DT
DT
Phase Shift Keying (PSK)

• Developed during the early days of deep-space program

and widely used in both military and commercial comms
systems.

• Analytic expression for PSK


2E
si (t) = cos[ω0 t + φi (t)]
T
for 0 ≤ t ≤ T and i = 1, . . . , M . E is the energy per bit.

• Phase term given by

2π
φi (t) = i
M
for i = 1, . . . , M encodes the information.

Simplest case M = 2

• Binary PSK (BPSK) exhibits abrupt phase changes at

the symbol transitions.

• Antipodal signal set with opposing vectors in the signal

space (constellation).

DT
Frequency Shift Keying (FSK)


2E
si (t) = cos[ωi t + φ]
T
for 0 ≤ t ≤ T and i = 1, . . . , M .

• Frequency term ωi has M discrete values which encodes

the information.

• Phase term φ is an arbitrary constant.

• In general, the change from one tone (frequency) to an-

other tone can be quite abrupt.

• In practice ωi are chosen so that the signals are orthog-

onal to each other
 T
si (t)sj (t)dt = 0 if i = j
0

• This orthogonality criterion imposes restrictions on the

choice of frequencies - see later.

• Only orthogonal sets can be characterized by perpendic-

ular axes in the signal space.

DT
Amplitude Shift Keying (ASK)


2Ei (t)
si (t) = cos[ω0 t + φ]
T
for 0 ≤ t ≤ T and i = 1, . . . , M .

Amplitude term 2ETi (t) have M discrete values and encodes
the information while ω0 and φ are constants.

Amplitude / Phase Keying (APK)

Combine ASK and PSK to give APK


2Ei (t)
si (t) = cos[ω0 t + φi (t)]
T
When the set of M symbols in the signal space are arranged
in a rectangular constellation the signalling is referred to as
QAM (Quadrature Amplitude Modulation).

DT
Example of QAM

Detection of signals in Gaussian noise

• Result from previous chapter on baseband system can
be used for bandpass system by virtue of the equivalence
theorem: frequency shifting does not affect the mathe-
matics of linear signal processing.

DT
 • Performance of most DCS can be studied and analyzed
by an equivalent baseband model.

Decision regions
• Use concepts from vector space representation of signals:
consider 2-D signal space.

DT
• Binary signalling (M = 2) with either signal s1 or s2
transmitted.

• Due to noise, received signal is either s1 + n or s2 + n.

Noise vector n has Gaussian distribution.

• If the received signal is given by r, the detector must

decide which of the signals (s1 or s2 ) was actually trans-
mitted.

• Minimum error decision achieved by choosing the signal

that will give the minimum distance

d(r, si ) = ||r − si ||

where ||x|| is called the norm or magnitude of x.

• This requires the signal space to be partitioned into two

decision regions:

1. If r is in region 1, choose s1 .
2. If r is in region 2, choose s2 .

• The boundary between the regions is the perpendicular

bisector of the line joining the tips of the vectors s1 and
s2 .

• This principle of decision regions extends to higher di-

mension and more general signalling where M > 2.

DT
Correlation receiver

Steps involved for detection of bandpass digital modulated

signal is similar to the baseband equivalent:

1. Matched filtering of received signal to maximize SNR.

This step can be equivalently implemented by using a
correlator. Basically this step transforms the received
waveform r(t) into a vector (point) r in the signal space.

2. Decision making as to which symbol was transmitted.

Correlation receiver comprising M correlators.

DT
Each correlator output is given by
 T
zi (T ) = r(t)si (t)dt
0

for i = 1, . . . , M .
Decision stage chooses si (t) with the largest zi (T ), i.e. best
match.
Alternative implementation of correlation receiver:

• It was shown that any signal set {s1 (t), s2 (t), . . . , sM (t)}

DT
 can be expressed in terms of some set of basis functions
{Ψ1 (t), Ψ2 (t), . . . , ΨN (t)} where N ≤ M .

• Using basis functions to form the reference signals:

• The decision stage will be different: the choice of si (t) is

made according to the best match of the coefficient aij
with the set of outputs {zi (T )}. The coefficient aij are
from
N

si (t) = aij Ψi (t)
j=1

DT
 • The correlator implementation using the basis functions
{Ψi (t)} is cost effective if M is larger than N .

• If the signal set {si (t)} is orthogonal however then M =

N and both implementation are essentially the same.

Binary signalling detection similar to the baseband case. De-

cision threshold

• Decide H1 (s1 was transmitted) if z(T ) > γ0 .

• Decide H2 (s2 was transmitted) if z(T ) < γ0 .

Coherent detection of BPSK


2E
s1 (t) = cos[ω0 t]
T
for 0 ≤ t ≤ T .
 
2E 2E
s2 (t) = cos[ω0 t + π] = − cos[ω0 t]
T T
for 0 ≤ t ≤ T .
Using the basis function

2
Ψ1 (t) = cos[ω0 t]
T

DT
for 0 ≤ t ≤ T , we have
√
s1 (t) = a11 Ψ1 (t) = EΨ1 (t)
√
s2 (t) = a21 Ψ1 (t) = − EΨ1 (t)

Correlator output:
√
zi (T ) = ± E + n0

Choose signal with the largest value of zi (T ).

Coherent detection of MPSK

Consider the 4-symbol PSK, i.e. M = 4, k = 2.

DT
Signal space diagram:

DT
For a general M -symbol PSK
  
2E 2π(i − 1)
si (t) = cos ω0 t −
T M

for 0 ≤ t ≤ T , i = 1, . . . , M .
Orthonormal basis functions

2
Ψ1 (t) = cos(ω0 t) (I component)
T

2
Ψ2 (t) = sin(ω0 t) (Q component)
T
Signal waveform in terms of basis functions

si (t) = ai1 Ψ1 (t) + ai2 Ψ2 (t)

 
√ 2π(i − 1)
= E cos Ψ1 (t)
M
 
√ 2π(i − 1)
+ E sin Ψ2 (t)
M

For M = 4, the four decision regions are as shown in figure.

• In theory, the detection can be achieve with M correla-

tors, one for each of the waveform si (t).

• In practice only 2 correlators are needed by using the

orthonormal basis functions.

DT
Received signal can be expressed as

2E
r(t) = (cos φi cos(ω0 t) + sin φi sin(ω0 t)) + n(t)
T
2π(i−1)
where φi = M and n(t) is AWGN.
Upper correlator gives
 T
X= r(t)Ψ1 (t)dt
0

Lower correlator gives

 T
Y = r(t)Ψ2 (t)dt
0

DT
Angle  
Y
φ̂ = tan−1
X
is computed.
Decision stage choose waveform si (t) that gives the smallest
difference in phase angle, i.e. min |φi − φ̂|.

Coherent detection of FSK


2E
si (t) = cos[ωi t]
T
for 0 ≤ t ≤ T , i = 1, . . . , M .

• (ωi+1 −ωi ) is typically assumed to be an integral multiple

of Tπ , i.e. k Tπ .

• The signals si (t) will then be orthogonal to each other

and an orthonormal basis set is given by

2
Ψj (t) = cos(ωj t)
T
for j = 1, . . . , N (= M ).

PROBLEM WITH COHERENT DETECTION: Need to be

able to generate a reference carrier in the receiver that is in
synchronization with the received signal carrier.

DT
Non-coherent detection of differential PSK
(DPSK)

• The term ’differential’ can mean different things in dif-

ferent context.

• Differential encoding means data is encoded differen-

tially; the presence of binary one or zero is manifested
by the symbol’s similarity or difference when compared
with the preceeding symbol.

• Differentially coherent detection of DPSK actually is non-

coherent as a reference phase is not required in the re-
ceived carrier.

• DPSK can also be coherently detected.

Transmitted waveform is

2E
si (t) = cos[ω0 t + θi (t)]
T
i = 1, . . . , M .
Received signal is

2E
r(t) = cos[ω0 t + θi (t) + α] + n(t)
T
where α is the arbitrary unknown phase shift and n(t) is
AWGN.

DT
The phase θi (t) contains the data but the absolute value is
not important.
Relative value determines the data
2πi
θi (t) = θi−1 (t) +
M

• The carrier phase of the previous signalling interval can

be used as a reference phase for demodulation.

• In the detector, the phase of the locally generated sinu-

soidal signal is not important, i.e. synchronization not
needed.

• The detector measures the angle between the currently

received signal vector and the previously received signal
vector:

DT
 • Problem with DPSK is that errors tend to propagate due
to the correlation between signalling waveforms.

Binary DPSK - see book for details.

Non-coherent detection of orthogonal FSK

Orthogonal
 T
cos(ωi t) cos(ωj t)dt = 0
0

for i = j.

• Cannot rely on phase information.

• Use information about energy

DT
DT
 • Suppose symbol 1 (cos(ω1 t)) transmitted. Received sig-
nal is
r(t) = cos(ω1 t + φ) + n(t)
where φ is the unknown phase and n(t) is the noise.

• Using trigonometric expansion, this can be expressed as

r(t) = cos(φ) cos(ω1 t) − sin(φ) sin(ω1 t) + n(t)

where

1. The first term is the I component and is captured by the

I channel. Energy K cos2 φ.
2. The second term is the Q component and is captured by
the Q channel. Energy K sin2 φ.

• Combine the energy of both components before decision

making.

• The combined energy K cos2 φ + K sin2 φ = K is inde-

pendent of the unknown phase φ.

• The energy detected for the lower branch will be small

due to the orthogonality condition with other symbol
waveform, e.g. cos(ω1 t) is orthogonal to cos(ω2 t). The
energy is due primarily due to noise.

Alternative implementation uses a bank bandpass filters - see

book for details.

DT
Tone spacing and bandwidth of orthogonal
FSK

• A property that ensures orthogonality between two tones

(frequencies) in FSK signalling is that the frequency sep-
aration is a multiple of 1/T Hz

k
f1 − f2 =
T
where k is an integer (see book for proof).

• Minimum frequency spacing is therefore 1/T (when k =

1).

• Bandwidth of FSK is related to the tone (frequency)

spacing. An estimate of the bandwidth is given by:

M
Bandwidth of MFSK =
T
(see book for justification).

• In FSK bandwidth increases with symbol size (M ) but

in PSK bandwidth is independent of symbol size.

Complex envelope: efficient way to implement PSK modula-

tion/demodulation by using two amplitude modulators - see
book for details.

DT
Error performance for binary signalling schemes

Coherently detected BPSK:

• Two types of error probability

1. PE : symbol error.
2. PB : bit error.

• In general PB = PE if M > 2.

• BPSK is a form of antipodal signalling, i.e.

√
s1 (t) = EΨ1 (t)
√
s2 (t) = − EΨ1 (t)

• Therefore performance is similar to that for baseband

system using matched filter detection

2Eb
PB = PE = Q
N0

Coherently Detected DBPSK:


 
2Eb 2Eb
PB = 2Q 1−Q
N0 N0

which is a little bit larger than PB for coherent detection of

BPSK.

DT
Coherently Detected Orthogonal BFSK:

Eb
PB = Q
N0

using the result from previous chapter on binary orthogonal

signalling.
Non-coherently Detected Orthogonal BFSK: see book for deriva-
tion  
1 −Eb
PB = exp
2 2N0
assuming envelope detectors are used.
Non-coherently Detected DBPSK: see book for derivation
 
1 −Eb
PB = exp
2 N0

DT
Plot of error performance.

For PB = 10−4 , approximately 4dB between the best (coher-

ent PSK) and worst (non-coherent orthogonal FSK).

DT
Example: A BPSK system consist of the following waveforms

s1 (t) = A cos ω0 t s2 (t) = −A cos ω0 t

with A = 10 mV and operating with a bit-rate of 1 Mbits/s.

The noise PSD N0 = 10−11 W/Hz. Determine the error
probability.
Solution:
1 1
Rb = Rs = →T = = 10−6
T R
1 2 1
Eb = A T = (10−2 )2 × 10−6 = 5 × 10−11
2 2

2Eb
= 3.16
N0

2Eb
PB = Q = Q(3.16) = 3 × 10−4
N0

Example: If the main criterion is bit error probability, which

of the following two modulation schemes would be selected
for an AWGN channel?

1. Binary noncoherent orthogonal FSK with Eb /N0 = 13

dB.

2. Binary coherent PSK with Eb /N0 = 8 dB.

What if the criterion was transmission bandwidth?

Solution:

DT
 1. Eb /N0 = 13 dB = 19.95. Noncoherent BFSK.
   
1 Eb 1 19.95
PB = exp − = exp − = 2.32 × 10−5
2 2N0 2 2

2. Eb /N0 = 8 dB = 6.31. Coherent BPSK.


2Eb √
PB = Q = Q( 2 × 6.31)
N0
1
= Q(3.55) ≈ √ exp(−12.62/2) = 2.04 × 10−4
2π × 12.62
Choose non-coherent BFSK.

Symbol error performance for M > 2

• In most cases no exact analytical formula exist.

• The derivation of the approximate formulas are usually

quite complicated and found in research papers.

• In some situations only through computer simulation can

the performance be determined.

1. MPSK: Assuming large energy-to-noise ratios

• With coherent detection

2Es π
PE (M ) ≈ 2Q sin
N0 M

DT
 where Es = Eb log2 M is the energy per symbol. Per-
formance curves in Fig. 4.35.
• Differentially coherent detection (actually non-coherent)
of MDPSK

2Es π
PE (M ) ≈ 2Q sin √
N0 2M

2. MFSK with orthogonal signalling

• Coherent detection

Es
PE (M ) ≤ (M − 1)Q
N0

• Non-coherent detection
 
1 Es
PE (M ) = exp −
M N0
M  
E s
. (−1)j Bin(M, j) exp
j=2
jN0

where Bin(M, j) is the binomial formula

M!
Bin(M, j) =
j!(M − j)!

DT
Relationship between PB and PE
1. Orthogonal signalling
PB 2(k−1) M/2
= k =
PE 2 −1 M −1
As k → ∞;
PB 1
→
PE 2
Note that PB < PE for M > 2 because a symbol error
does not mean all bit it carries have error.

2. MPSK with Gray encoding

PE
PB ≈
log2 M
assuming PE is small.

DT
Example:

1. A 16-ary PSK system is operating with symbol error

probability PE = 10−5 and Gray code assignment from
symbols to bits. Determine the approximate bit error
probability.

2. Repeat part 1 for a 16-ary orthogoal FSK system.

Solution:
1.

PE 10−5
PB ≈ = = 2.5 × 10−6
log2 M 4

2.
M/2 8
PB = PE = × 10−6 = 5.3 × 10−6
M −1 15

Example: If the main criterion is bit error probability, which

of the following two modulation schemes would be selected
for an AWGN channel?

1. Coherent 8-ary orthogonal FSK with Eb /N0 = 8 dB.

2. Coherent 8-ary PSK (using Gray code assignment) with

Eb /N0 = 13 dB.

DT
What is the ratio of bandwidths between scheme 1 and scheme
2?
Solution:

1. M = 8, k = 3.
Eb /N0 = 108/10 = 6.31
ES /N0 = kEb /N0 = 3 × 6.31 = 18.93

Es √
PE (M ) ≤ (M − 1)Q = 7Q( 18.93)
N0

7 4.352
≈ √ exp − = 4.98 × 10−5
4.35 2π 2
M/2 4
PB = PE = × 4.98 × 10−5 = 2.85 × 10−5
M −1 7
2.
Eb /N0 = 1013/10 = 20
ES /N0 = kEb /N0 = 3 × 20 = 60

2 Es  π 
PE (M ) ≈ 2Q sin =
N0 M
√
2Q( 120 sin(π/8)) = 2Q(4.192)
 2

2 4.192
≈ √ exp − = 2.9 × 10−5
4.192 2π 2
PE
PB ≈ = 2.9 × 10−5 /3 = 9.7 × 10−6
log2 M

DT
Choose coherent 8-ary PSK.
Exercise to try yourself from texbook: problems 4.1,
4.2 (page 237), 4.7 (page 238), 4.13, 4.14, 4.15 (page
240). Solution will be provided at a later date on LMS.

DT