1

Digital Modulation for Wireless Communication
Requirements for modulation method
• Compact spectrum (many bits/sec/Hz)
• Immune to interference
1. Cochannel interference
2. Adjacent channel interference
• Robust against fading (frequency dispersion) and
multipath (time dispersion)
• Simple implementation of transmitter and receiver
• Low-power consumption of transmit power amplifiers
For instance: "constant envelope signals" allow the use
of nonlinear Class C amplifiers
2
Spectral Efficiency of Cellular Network
Two aspects: Reuse distance R
u
determines
• propagation distances of interfering signals, thus the
C/I ratio, and
• the cluster size, thus the number of channels needed
C/I ratio γ is proportional to R
u
β
where β path loss exponent
For a certain required protection ratio z, one must take
C/I ∝ R
u
β
> z so R
u
> z
1/β
Number of channels C : proportional to R
u
2
Hence, the number of channels needed is proportional to z
2/β
3
Minimize Bandwidth of Cellular System
The system designer can choose the modulation method.
• This affects the transmit bandwidth B and
• the required reuse distance R
u
Typically, one can trade transmit bandwidthB for good
immunity to interference and fading.
Optimum spectrum efficiency:
• Minimize B z
2/β
Extreme choices:
• DS-CDMA: large B, small z, very dense reuse
• Analog AMPS: small B, but large reuse distance
4
Detection of Digital Signals
Theory of digital communication is most detailed for channels
• with Additive White Gaussian Noise (AWGN)
• no interference
• Linear Time-Invariance
• Non-dispersive
One of the most important results is
the “matched filter receiver”.
However, wireless communication suffers
• Time varying channels
• Frequency disperisive channels
• Interference from other users
5
Matched Filter Receiver
• Matched Filter:
Multiply received signal by reference signal waveform
• This is equivalent to filtering with time-inverse impulse
response
• Decision Variable is "Sufficient Statistic"
• Interpretation:
• weigh strong components heavily
• ignore weak components
Matched Filter is optimum if
• Noise is AWGN (presence of interference may require
a more complex receiver structure)
• Channel is non-selective, Linear Time-Invariant (LTI)
• Exact time reference is available
6
One Shot Transmission over Dispersive Channel
• How to apply Matched Filter concept for LTI AGWN
channel with dispersion?
H(f)
Wireless Channel
H’(f)
Locally Stored Copy of Message Alphabet
Locally Generated
Estimate of Channel
Integrate
Message
Figure: Optimum One-Shot Receiver
• Match the filter to expected receive waveform
• Hence, receiver needs to know channel behavior
• For CDMA this leads to the rake receiver
• Transmission of multiple signals: Intersymbol
Interference occurs.
further investigation is needed, e.g. treat total bit stream
as one composite symbol, correlate received signal with
all possible sequences
7
Transmission over slowly fading, non-
dispersive channel
Assumptions
• No ISI
• Perfect Synchronization
• Fading affects only received energy per bit
Amplitude is Rician distributed, phase is known
Definitions
• Instantaneous BER:
BER experienced at a particular location
• Local-mean BER:
BER averaged over multipath fading fading
• Fast fading and slow fading:
fast/slow compared to symbol duration
fast/slow compared to block duration
8
Effect of Fading on BER
For Very Slow Fading
• Fading increases the BER: with some probabilityE
b
is
small, even for large average E
b.
• Waterfall behavior vanishes:
only a slow decrease of BER with improving C/N
• Most bit errors occur during deep fades, in bursts
• The burst length depends on the speed of antenna
• Stationary users do not experience burst errors,
but receive constant power.
Then the BER depends highly on location
• Coding and interleaving may help
• Local-mean BER does not say so much in practice
Other Effects
• Rapid (Doppler) channel changes causes bit errors
• Doppler affects accuracy of phase reference
• Channel amplitude changes may also cause errors in
QAM
• ISI due to delay spread
9
COHERENT BPSK
• Binary Phase Shift Keying:
For 0 and 1: carrier has opposite phase
Instantaneous Bit Error Rate
• For coherent detection
• Non-selective Linear Time-Invariant (LTI) channels
• Additive White Gaussian Noise ( AWGN)
b
1
2
b
0
P (error) =
E
N
. erfc
where
• E
b
is the energy per bit
E
b
= p
0
T
b
with T
b
the bit duration, and
• N
0
is the one-sided spectral power density of the AWGN.
E
b
/ N
0
is the signal-to-noise ratio γ.
• erfc: complementary error function: erfc(k) = 2Q(√2 k).
10
COHERENT BPSK
Local-mean BER
• Slowly Rayleigh-fading channel
• Exponentially distributed power
• Local-mean: Take average over the PDF of received
signal power. For Rayleigh fading:
P =
1
p
-
p
p
1
2
erfc -
pT
N
dp
b
0
b
0


¹
'
¹
¹
'
¹
¹
'
¹
¹
'
¹
exp
so
P =
1
2
1
2
pT
pT N
b
b
b

+
0
• Asymptotically
b
0
b
P

N
4 pT
=
1
4

γ
Inversely proportional to signal-to-noise ratio, i.e., only a
slow improvement in BER with increasing C/N
11
Area-mean BER
Bit error rate averaged over amplitude distribution of due to
fading and shadowing
Curve c-f: 12 dB shadowing
Curve b-e: 6 dB shadowing
Curve a-d: no shadowing
Wanted signal: Rician fading, K=4 (a,b,c) K=16 (d,e,f)
Interference: Rayleigh fading
12
Average BER for fading channel
Figure: Area-mean BER versus C/I ratio.
C/I ratio can be converted into the required reuse distance
(using a 40 log d path loss model) and into a cluster size.
Curve
a Rayleigh fading, no shadowing
b Rayleigh fading, 6 dB shadowing
c Rayleigh fading, 12 dB shadowing
13
Fast Fading
There are different definitions of fast fading.
1) Packet Duration >> Coherence Time of Channel
2) Bit Duration >> Coherence Time of Channel
We use definition 1 and assume that
• Bit Duration << Coherence time of Channel
• Amplitude and phase statistically independent from bit to
bit
• Receiver stays in lock
Packet Success Probability
( )
[ ] [ ]
Pr
i
m=0
M
L- m
b
m
b
s
=
L
m
1- P P

|
.

`
,

where
L packet length in bits
M bit error correcting code
14
Slow Fading
Amplitude and phase constant for duration of a packet.
⇔ During one packet time,
motion of the mobile << wave length
Probability of not more than M bit errors in block of L bits
( )
Pr erfc erfc
i
0
m=0
M
L-m
1
2
0
b
0
m
1
2
0
b
0
s
| p =
L
m
1-
p
T
N
p
T
N

|
.

`
,

¹
'
¹
¹
¹
¹
'
¹
¹
¹

]
]
]
]
¹
'
¹
¹
¹
¹
'
¹
¹
¹

]
]
]
]
Slow Fading:
• First compute conditional Packet success probability;
• Then average over all possible channels
Fast Fading:
• First find average BER;
• Then compute Packet Success Probability
15
Performance comparison
For Slow Fading
• the number of bit errors is not Binomial
• errors are highly correlated (bursts)
• error correction is less effective
Performance
• if no coding is used: slow fading performs better
• if coding is used: BER better for fast fading if C/N is
sufficiently large (cellular telephony)
• with fast fading (independent bit errors), coding can
change to slope of the BER vs. C/N curve.
Increasing C/N has more effect for improving BER
(Consider coding as a kind of diversity)
• better for slow fading if C/N is very small (random
access, collision channels)
16
BER of Binary Phase Shift Keying (continued)
Noncoherent Detection for BPSK
Delay
1 bit time
Integrate
X
• No carrier phase reference needed:
- simplicity of receiver
- Advantageous in fading channel
• Typically 3 dB more sensitive to noise
• Previous bit is used as reference:
Differential encoding
0: no phase reversal
1: phase reversal
• BER analysis is complicated due to inaccurate phase of
reference bit
BER analysis simple for binary DPSK
17
Differential BPSK
For LTI AWGN, the instantaneous BERis
b
1
2
0
b
0
P = -
p
T
N
exp
¹
'
¹
¹
'
¹
Local-mean BER
For a flat, slow Rayleigh-fading channel
P=
N
2 pT +2 N
0
b 0
• Asymptotically 3 dB more vulnerable to AWGN than
coherent BPSK
18
Example:
Noncoherent detection of D-BPSK in a slow
Rayleigh fading channel
Transmit digital words of L bits.
Find the probability of success for a code that can correct up
to M bit errors.
The conditional probability of successful reception for slow
fading is
P exp exp (S| p) =
M
m
1
2
-
pT
N
1-
1
2
-
pT
N
m=0
M
m
b
0
L-m
b
0

|
.

`
,

¹
'
¹
¹
'
¹
|
.

`
,

¹
'
¹
¹
'
¹
|
.

`
,

Averaging over PDF of received power gives the local-mean
success rate.
Closed-form solutions only exist for special cases.
19
Special Case
• no error correction is used (M = 0)
• A slow fading channel
The L-th order binomial can be rewritten as a sum of terms.
P(S) =
1
p
-
p
p
L
i
-
1
2
-
ipT
N
dp
0
i=0
L
i
b
0



¹
'
¹
¹
'
¹
|
.

`
,

|
.

`
,

¹
'
¹
¹
'
¹
exp exp
After interchanging sum and integral, this gives
P(S) =
L
i
(-2 )
N
N +i T p

1-
N
pT
L
i
1
i
-
1
2
i=0
L
-i 0
0 b
0
b i=1
L
i


|
.

`
,


|
.

`
,

|
.

`
,

• Probability of error is inversely proportional to C/N
For slow fading, coding does not change the slope of the BER
curve.
20
DQPSK:
Differential Quadrature Phase Shift Keying
• Used in U.S. and Japanese standards for digital
telephony, because of
• simplicity of implementation,
• good spectrum efficiency.
01
10
11
00
01
00
11 10
Phase of previous bits
Initial phase: λ = 0 λ = π/4
Figure: Signal constellations
Relative phase shift between
successive bit intervals for D - Q PSK
21
QPSK Receivers

Carrier
Recovery
X
X
Phase
Shift Q
Integrator Sample
Symbol
Sync
Integrator Sample
Inphase
Quadrature
F
i
n
d

n
e
a
r
e
s
t

s
y
m
b
o
l
Phase
Shift I
Coherent detection of QPSK signals
Phase shift I: 0, Phase shift Q: π/2
For noncoherent Detection:
• Carrier recovery is replaced by a delay line
Phase shifts
λ = 0 λ = π/4
Phase Shift I 0 π/4
Phase Shift Q π/2 -π/4
22
Differential Quadrature Phase Shift Keying
Instantaneous BER
Pierce (1962): BER for DQPSK,
given instantaneous amplitude ρ of wanted signal
b
1
2
-
2
P
(error| ) = Q(a,b) -
I
(ab)
0
2
a +
2
b
ρ exp
|
.

`
,

where
• Q(a,b) is the Marcum Q-function, i.e.,
- cumulative distribution of Rician pdf
- Prob ( x > b )
Q(a,b) =
b
x -
a
+
x
2

I
(ax) dx
2 2
0


|
.

`
,

exp
with
• I
k
is the k-th order modified Bessel function, and
• a and b are defined as
a = [2 - 2 ] b = [2+ 2 ] γ γ and
23
DIFFERENTIAL QPSK
The local-mean BER for Rayleigh fading,
b
2
2
2
P (error) =
1
2 1+4 +2
2 +[ 2 - 1][1+2 - 1+4 +2 ]
2 - [ 2 - 1][1+2 - 1+4 +2 ] γ γ
γ γ γ γ
γ γ γ γ
24
COMPARISON BPSK, QPSK and FSK
COHERENT BPSK
( )
b
0
1
2
T
i=1
N
i
b 0
P error| =
1
2
p T
+
N
0
2
b
ρ
ρ
erfc

COHERENT FSK
( )
b
0
1
2
T
i=1
N
i
b 0
P error| =
1
2
p T
+2
N
0
2
b
ρ
ρ
erfc

COHERENT QPSK
( )
b
0
1
2
T
i=1
N
i
b 0
P error| =
1
2
p T
+2
N
0
2
b
ρ
ρ
erfc

25
Discussion
• Interference affects BER in a different way than AWGN.
• Increasing T
b
reduces effect of noise
but has no effect on interference
• FSK is fairly robust
• QPSK is more vulnerable to interference than expected
Intuition: take BPSK; double T
b
; insert Q-phase;
reduce amplitude by √2
- interference amplitude stays constant
- amplitude of wanted signal drops 3 dB
26
Pulse Shaping
• Pulse shaping is needed make transmit spectrum
bandlimited


Spectral
Amplitude
Frequency
Point of
Symmetry
• Rectangular spectrum requires SINC-shaped pulses.
Disadvantage: the addition of many randomly polarized,
time-shifted SINC may give large signal peaks.
• Smoother behavior is guaranteed by raised-cosine type
spectra.
• Symmetric spectral roll-off ensures absence of ISI
27
Square Root Nyquist
Wireless
Channel
Transmit
Filter H
Receive
Filter G
Pulse Shaper
Matched Filter
Two requirements for pulse shape:
• Matched Filter
Receive Filter Characteristic = Transmit waveform
• No Intersymbol Interference
Transfer function of all filters should be symmetric.
Take the transfer function such that
• H G is symmetric
• Frequency transfer G = H
28
Quadrature Amplitude Modulation
01
00
11 10
Q
I
I and Q diagram of 4-QAM Signal with noise
• multi-level modulation packs more bits/sec per Hz
• But: QAM is more vulnerable to interference
Larger reuse distance needed: Less Spectrum-
Efficient?
• But: QAM is more vulnerable to AWG noise
• Application:
• Radio relay links
• Cable TV including "Wireless Cable" Microwave
Multi-Point Video Distribution
29
M-ary Phase Shift Keying (PSK)
110
000
100
010
011
001
101
111
M-PSK Constellation, M = 8
• For large M: Theoretically, BER worse than QAM
• Constant amplitude: simple power amplifiers
• Use Gray code
• QPSK is particularly popular (M = 4)
• Special case Continuous Phase- Phase Shift Keying
30
Frequency Shift Keying (FSK)
Oscillator
Freq. 1
Oscillator
Freq. 2
Oscillator
Freq. n
Wireless
Channel
Filter
Freq. 1
Filter
Freq. 1
Filter
Freq. 1
Envelope
Detector
Envelope
Detector
Envelope
Detector
S
e
l
e
c
t

L
a
r
g
e
s
t
FSK communication system
FSK:
• Theoretically sub-optimum, compared to BPSK
• Noncoherent detection
• Simple to implement, robust
• Noncoherent FSK works well on fading channel
• Special case: Continous Phase FSK
31
Coherent Frequency Shift Keying (FSK)
Instantaneous BER
For LTI AWGN, the BER is
b
1
2
b
0
P (error) =
E
2 N
. erfc
Local-mean BER
( )
b
0
b
0
b 0
P
error =
1
2
-
1
2
p T
p T +2 N
• Coherent FSK is 3 dB less immune to noise than BPSK
(FSK is orthogonal, BPSK is antipodal)
32
Non-coherent FSK
Instantaneous BER
For LTI AWGN, the BER is approximately
b
1
2
0
b
0
P (e) = -
p
T
2 N
exp
¹
'
¹
¹
'
¹
Local-mean BER
b
0
0
b 0
P (error) =
N
p T +2 N
33
Continuous Phase FSK (CP FSK)
• CP reduces spectral sidelobes
• Data is embedded in the "phase trajectory"
• Can be interpreted as Phase Shift Keying or as
Frequency Shift Keying

Time
Phase
1 1 0 1 0 0 0

−hπ
−2hπ
2hπ
Phase trajectory for
• Binary Continuous Phase Frequency Shift Keying (blue)
• Filtered CP FSK to reduce spectral sidelobes, MSK
(Red)
34
Minimum Shift Keying (MSK)
• Special case of CP FSK:
phase trajectory is 'filtered'
• Special case of phase modulation
MSK can be interpreted as a form of two time-shifted
binary PSK signals, each with a sinusoidal envelope
• Very compact spectrum
h is chosen as small as possible, but large enough to
ensure orthogonal signal waveforms, h = 0.5
• Robust against interference
• Anti-podal signals: robust against noise
• Constant envelope
• Used in GSM, Hiperlan, ....
35
MSK, QPSK and Offset QPSK
During the transmission of a single bit the MSK signal
smoothly increases or reduces the signal phase by 90°.
In normal QPSK, abrupt phase changes of 90° or 180° can
occur at the end of each symbol transmission (2 bits)
In Offset QPSK, abrupt phase changes of 90° occur but twice
as frequent as in normal QPSK.
36
INTERFERING BPSK SIGNALS
The effect of interference differs significantly from
the effect of AWGN
Wanted BPSK signal with a single interferer.
Effect of interfering BPSK Signal depends on
• Relative amplitude
• Phase offset
• Bit timing offset
37
INTERFERING FADING CARRIERS
Received signal
( )
r(t)
a
2 ( + )t+ +
n
= ( t + )+
f f
0
0
c
i=1
N
i
c i i
I
ρ
π ω φ ρ
φ
cos cos

with bit a
0
= +1. After matched filtering, the decision variable
becomes
v =
a T
+ (
i
t)dt
+ (
i
t)dt +
n
0
0
b
i=1
N
i
0
T
i=1
N
i
0
T
I
b
b
ρ ξ
ω
ζ
ω




cos
sin
where
ξ
i
and ζ
i
are inphase and quadrature component of i-th
interfering signal with
ξ
i
= ρ
i
cos θ
i,
(zero-mean Gaussian)
ζ
i
= ρ
i
sin θ
i
(zero-mean Gaussian)
and Eξ
i
2
= Eζ
i
2
=
i
.
Variance of the interference term is Σ
i
T
b
2
sincf
i
T
b
.
38
Effect of BPSK Interference
• Bit synchronisation offset α
i
, α
i
uniformly distributed
between 0 and 1.
Decision variable for a receiver in perfect lock to the wanted
signal:
{ ¦
ν
ρ
κ
ζ κ α κ α
= + + (1- ) +n .
0
0
i=1
n
i
i i i i
I ∑

This contains components from previous and next bit interfering
signal
Remarks:
• Interference is NOT Gaussian
• Interference cancellation is easier if bit timing is
synchronous
• Interference affects synchronization; it does not "average
out" in the receiver PLL
39
Average BER for synchronization offset
• Bit-timing offset α
i
is uniformly distributed for 0 < α
i
< 1
If phase reversal occurs at the instant (k + α
i
)T
b
,
the variance of the interference sample is p
t
(1 - 2α
i
)
2
.
• If there is a timing offset, interference components are a
little weaker
Assumption
• "0"'s and "1"'s are equiprobable and independent,
• the interference is Rayleigh fading, and
• the receiver locks perfectly to the wanted signal
the bit error probability is
b
0
b
0
b
1
b 0
0
b
0
b
1
b 1
2
0
P
=
1
2
-
1
4
p T
p T + p T + N

-
1
4
p T
p T + p T (1- 2 ) + N
.
α
• We only modeled the effect on the decision variable, not
on the timing recovery
• In practice the effect on receiver synchronization may be
more significant
40
Average BER for Sync. Random Offset
For a Rayleigh-fading wanted signal, the average BER is
b
0
b
0
b
1
b 0
1
4
0
1
0
b
1
b 0
1
b
0
b 0
P
=
1
2
-
1
4
p T
p T + p T + N
+
p
p
p T + p T + N - p T
p T + N
ln
|
.

`
,

• Non-synchronous interference is effectively 1.8 dB
weaker than synchronous interference
41
Many interfering signals
• Relevant case in Spread Spectrum systems
• Many weak signals may not have such a dramatic effect
on receiver synchronization
• CLT: The sum of many weak signals is a Gaussian
distributed random variable.
Variance of interference
• If an interfering signal makes a phase reversal, the
variance of the interference sample is
E[ ( ) ] ζ α
i i
1
p
2 2
1 2
3
− ·
• If no phase reversal occurs, the variance is
E[ ] ζ
i
1
p
2
·
42
Many interfering signals
Local-mean BER
b
0
b
0
b
P
=
1
2
-
1
2
p T
p T +
t
p
b
T +
0
N
2
3
For Spread Spectrum with Random Codes
b
0
b
0
b
2
3C
t
b
P
=
1
2
-
1
2
p T
p T + p T +N
0
where
• C is the spread factor
• Chips of different users have random timing offset
43
Effect of Intersymbol Interference
The impulse response of a discrete-time multipath channel.
h(t) = h T
k=0
k b
(t+k )


δ
Example of (discrete time) impulse response multipath
channel and frequency transfer function.
44
Effect of Intersymbol Interference
Irreducable BER
• Excessively delayed reflections causeintersymbol
interference (ISI).
• If unequalized, this results in a residual BER, even if no
noise is present.
• If perfect synchronization to the first (resolvable) path,
the decision variable becomes
n n 0
k=1
k n-k
= a h + h a ν


For a known signal power p
0
in the first resolvable path, the
BER becomes
b
0
1
2
0
b
k=1
k
b 0
P
(error| p ) =
p
T
p
T
+
N
. erfc


45
Effect of Intersymbol Interference
2. Noise penalty due to channel equalization
`zero forcing equalizer'
• eliminates the effect of intersymbol interference
altogether.
• Disadvantage: it excessively enhances the noise.
Signal power density spectrum S(f)
AWGN channel with frequency transfer function H(f)
Zero Forcing filter 1/H(f)
Received power spectrum
R(f) = S(f) +
N
H(f)H* (f)
0
Thus signal to noise ratio
S
N
=
E T
N
1
| H(f)|
df
b b
0
-1
B
2
T

|
.

`
,

46
Equalizer Approaches
• Best equalization minimizes the signal-to-interference-
plus-noise, but this is not necessarily the best receiver
design:
• Maximum likelihood (ML) detector
Find the bit sequence that was most likely
• Minimize the Mean Square Error(MSE) caused by ISI
and noise
LMS: Least mean square algorithm recursively adjusts
tap weights

Spectral Efficiency of Cellular Network

Two aspects: Reuse distanceRu determines • propagation distances of interfering signals, thus the C/I ratio, and • the cluster size, thus the number of channels needed C/I ratio γ is proportional to Ruβ where β path loss exponent For a certain required protection ratio z, one must take C/I ∝ Ruβ > z so Ru > z1/β Number of channels C : proportional to Ru2 Hence, the number of channels needed is proportional toz2/β

2

Minimize Bandwidth of Cellular System
The system designer can choose the modulation method. • This affects the transmit bandwidthB and • the required reuse distance Ru

Typically, one can trade transmit bandwidthB for good immunity to interference and fading.

Optimum spectrum efficiency: • Minimize B z2/β

Extreme choices:

• •

DS-CDMA: large B, small z, very dense reuse Analog AMPS: small B, but large reuse distance

3

Detection of Digital Signals
Theory of digital communication is most detailed for channels • with Additive White Gaussian Noise (AWGN) • no interference • Linear Time-Invariance • Non-dispersive

One of the most important results is the “matched filter receiver”.

However, wireless communication suffers • Time varying channels • Frequency disperisive channels • Interference from other users

4

Linear Time-Invariant (LTI) Exact time reference is available 5 .Matched Filter Receiver • Matched Filter: Multiply received signal by reference signalwaveform • This is equivalent to filtering with time-inverse impulse response • • Decision Variable is "Sufficient Statistic" Interpretation: • • weigh strong components heavily ignore weak components Matched Filter is optimum if • Noise is AWGN (presence of interference may require a more complex receiver structure) • • Channel is non-selective.

e. further investigation is needed. receiver needs to know channel behavior For CDMA this leads to the rake receiver Transmission of multiple signals:Intersymbol Interference occurs. correlate received signal with all possible sequences 6 . treat total bit stream as one composite symbol.g.One Shot Transmission overDispersive Channel • How to apply Matched Filter concept for LTI AGWN channel with dispersion? Wireless Channel Message H(f) Integrate H’(f) Locally Generated Estimate of Channel Locally Stored Copy of Message Alphabet Figure: Optimum One-Shot Receiver • • • • Match the filter to expected receive waveform Hence.

nondispersive channel Assumptions • • • No ISI Perfect Synchronization Fading affects only received energy per bit Amplitude is Rician distributed. phase is known Definitions • Instantaneous BER: BER experienced at a particular location • Local-mean BER: BER averaged over multipath fading fading • Fast fading and slow fading: fast/slow compared to symbol duration fast/slow compared to block duration 7 .Transmission over slowly fading.

but receive constant power. inbursts The burst length depends on the speed of antenna Stationary users do not experience burst errors. Then the BER depends highly on location • • Coding and interleaving may help Local-mean BER does not say so much in practice Other Effects • • • Rapid (Doppler) channel changes causes bit errors Doppler affects accuracy of phase reference Channel amplitude changes may also cause errors in QAM • ISI due to delay spread 8 . • Waterfall behavior vanishes: only a slow decrease of BER with improving C/N • • • Most bit errors occur during deep fades.Effect of Fading on BER For Very Slow Fading • Fading increases the BER: with some probabilityEb is small. even for large averageEb.

9 . N0 Eb is the energy per bit Eb = p0Tb with Tb the bit duration.COHERENT BPSK • Binary Phase Shift Keying : For 0 and 1: carrier has opposite phase Instantaneous Bit Error Rate • • • For coherent detection Non-selective Linear Time-Invariant (LTI) channels Additive White Gaussian Noise ( AWGN) Pb (error) = where • 1 2 erfc Eb . • erfc: complementary error function:erfc(k) = 2Q(√2 k). and • N0 is the one-sided spectral power density of the AWGN. Eb / N0 is the signal-to-noise ratio γ.

For Rayleigh fading: ∞ Pb = so ∫ 0  p 1 1  pTb exp .e. only a slow improvement in BER with increasing C/N 10 ..COHERENT BPSK Local-mean BER • • • Slowly Rayleigh-fading channel Exponentially distributed power Local-mean: Take average over the PDF of received signal power. i. erfc  dp 2 p  N0   p Pb = 1 1 − 2 2 pTb pTb + N0 • Asymptotically Pb → N0 4 pT b = 1 4γ Inversely proportional to signal-to-noise ratio.

c) K=16 (d.b.f) Interference: Rayleigh fading 11 .Area-mean BER Bit error rate averaged over amplitude distribution of due to fading and shadowing Curve c-f: 12 dB shadowing Curve b-e: 6 dB shadowing Curve a-d: no shadowing Wanted signal: Rician fading. K=4 (a.e.

Curve a Rayleigh fading. no shadowing b Rayleigh fading.Average BER for fading channel Figure: Area-mean BER versus C/I ratio. 12 dB shadowing 12 . 6 dB shadowing c Rayleigh fading. C/I ratio can be converted into the required reuse distance (using a 40 log d path loss model) and into a cluster size.

1) 2) Packet Duration >> Coherence Time of Channel Bit Duration >> Coherence Time of Channel We use definition 1 and assume that • • Bit Duration << Coherence time of Channel Amplitude and phase statistically independent from bit to bit • Receiver stays in lock Packet Success Probability  L Pr(si) = ∑   1 .Fast Fading There are different definitions of fast fading.P b m=0  m M [ ] [P ] L-m b m where L packet length in bits M bit error correcting code 13 .

2 erfc   N 0  m=0  m     M L-m   p0 T b    1  2 erfc   N 0      m Slow Fading: • • First compute conditional Packet success probability. Then average over all possible channels Fast Fading: • • First find average BER.Slow Fading Amplitude and phase constant for duration of a packet. motion of the mobile << wave length Probability of not more than M bit errors in block of L bits  p0 T b   L  1   Pr(si | p0 )= ∑   1 . ⇔ During one packet time. Then compute Packet Success Probability 14 .

Performance comparison For Slow Fading • • • the number of bit errors is not Binomial errors are highly correlated (bursts) error correction is less effective Performance • • if no coding is used: slow fading performs better if coding is used: BER better for fast fading if C/N is sufficiently large (cellular telephony) • with fast fading (independent bit errors). collision channels) 15 . Increasing C/N has more effect for improving BER (Consider coding as a kind of diversity) • better for slow fading if C/N is very small (random access. coding can change to slope of the BER vs. C/N curve.

BER of Binary Phase Shift Keying (continued) Noncoherent Detection for BPSK Delay 1 bit time X Integrate • No carrier phase reference needed: simplicity of receiver Advantageous in fading channel • • Typically 3 dB more sensitive to noise Previous bit is used as reference: Differential encoding 0: 1: no phase reversal phase reversal • BER analysis is complicated due to inaccurate phase of reference bit BER analysis simple for binary DPSK 16 .

the instantaneous BERis Pb = 1 2  p0 T b  exp  N0   Local-mean BER For a flat. slow Rayleigh-fading channel N0 P= 2 pT b + 2 N 0 • Asymptotically 3 dB more vulnerable to AWGN than coherent BPSK 17 .Differential BPSK For LTI AWGN.

18 .Example: Noncoherent detection of D-BPSK in a slow Rayleigh fading channel Transmit digital words of L bits. Closed-form solutions only exist for special cases. The conditional probability of successful reception for slow fading is  M  1  p T b   p T b   1 P(S| p) = ∑    exp1 .exp    N 0   N 0   2 m=0  m  2 M m L-m Averaging over PDF of received power gives the local-mean success rate. Find the probability of success for a code that can correct up to M bit errors.

coding does not change the slope of the BER curve. P(S) = ∫ 0 ∞  p  L  L  1  1  ip T b  exp . 19 . p T b i=1  i i  2  i • Probability of error is inversely proportional to C/N For slow fading.Special Case • • no error correction is used (M = 0) A slow fading channel The L-th order binomial can be rewritten as a sum of terms. this gives  L N0 -i P(S) = ∑   (-2 ) N0+ iTb p i=0  i L N 0 L  L 1  1  → 1∑  . ∑    . exp  dp p  i=0  i  2  p  N0   i After interchanging sum and integral.

01 01 11 00 11 10 00 10 Phase of previous bits Initial phase: λ = 0 λ = π/4 Figure: Signal constellations Relative phase shift between successive bit intervals for D .DQPSK: Differential Quadrature Phase Shift Keying • Used in U. and Japanese standards for digital telephony.S.Q PSK 20 . • good spectrum efficiency. because of • simplicity of implementation.

Phase shift Q: π/2 For noncoherent Detection: • Carrier recovery is replaced by a delay line Phase shifts Phase Shift I Phase Shift Q λ=0 0 π/2 λ = π/4 π/4 -π/4 21 Find nearest symbol Sample X .QPSK Receivers Inphase X Phase Shift I Integrator Carrier Recovery Phase Shift Q Symbol Sync Integrator Sample Quadrature Coherent detection of QPSK signals Phase shift I: 0.

2 ] and b = γ [2 + 2 ] 22 . given instantaneous amplitude ρ of wanted signal Pb (error| ρ ) = Q(a.b) is the Marcum Q-function.b) = ∫ x exp  I 0 (ax) dx  2  b with • Ik is the k-th order modified Bessel function.e. cumulative distribution of Rician pdf Prob ( x > b ) ∞  a2 + x2  Q(a..b) where • 1 2 I 0 (ab) exp    a 2 + b2   2     Q(a. i. and • a and b are defined as a = γ [2 .Differential Quadrature Phase Shift Keying Instantaneous BER Pierce (1962): BER for DQPSK.

1+ 4γ + 2 γ ] 2 2 2 1+ 4γ + 2 γ γ 2 .1][1+ 2γ .[ 2 .1+ 4γ + 2 γ ] 23 . Pb(error) = 1 2 γ 2 + [ 2 .1][1+ 2γ .DIFFERENTIAL QPSK The local-mean BER forRayleigh fading.

COMPARISON BPSK. QPSK and FSK COHERENT BPSK 1 Pb (error| ρ0)= erfc 2 1 ρ2 T 2 0 b ∑ p T +N i b i=1 N 0 COHERENT FSK 1 Pb (error| ρ 0 )= erfc 2 1 ρ2 T 2 0 b ∑ p T +2 N i b i=1 N 0 COHERENT QPSK 1 Pb (error| ρ0)= erfc 2 1 ρ2 T 2 0 b ∑ p T +2 N i b i=1 N 0 24 .

interference amplitude stays constant . • Increasing Tb reduces effect of noise but has no effect on interference • FSK is fairly robust • QPSK is more vulnerable to interference than expected Intuition: take BPSK. double Tb. insert Q-phase. reduce amplitude by √2 .amplitude of wanted signal drops 3 dB 25 .Discussion • Interference affects BER in a different way than AWGN.

time-shifted SINC may give large signal peaks. • Symmetric spectral roll-off ensures absence of ISI 26 . Disadvantage: the addition of many randomly polarized.Pulse Shaping • Pulse shaping is needed make transmit spectrum bandlimited Spectral Amplitude Point of Symmetry Frequency • Rectangular spectrum requires SINC-shaped pulses. • Smoother behavior is guaranteed by raised-cosine type spectra.

Square Root Nyquist Wireless Channel Transmit Filter H Pulse Shaper Receive Filter G Matched Filter Two requirements for pulse shape : • Matched Filter Receive Filter Characteristic = Transmitwaveform • No Intersymbol Interference Transfer function of all filters should be symmetric. Take the transfer function such that • H G is symmetric • Frequency transfer G = H 27 .

Quadrature Amplitude Modulation 01 Q 00 I 11 10 I and Q diagram of 4-QAM Signal with noise • • multi-level modulation packs more bits/sec per Hz But: QAM is more vulnerable to interference Larger reuse distance needed: Less SpectrumEfficient? • • But: QAM is more vulnerable to AWG noise Application: • • Radio relay links Cable TV including "Wireless Cable" Microwave Multi-Point Video Distribution 28 .

M = 8 • • • • • 001 000 100 For large M: Theoretically. BER worse than QAM Constant amplitude: simple power amplifiers Use Gray code QPSK is particularly popular (M = 4) Special case Continuous Phase.M-ary Phase Shift Keying (PSK) 011 010 110 111 101 M-PSK Constellation.Phase Shift Keying 29 .

1 Envelope Detector FSK communication system FSK: • • Theoretically sub-optimum. compared to BPSK Noncoherent detection • • • Simple to implement.Frequency Shift Keying (FSK) Oscillator Freq. 1 Oscillator Freq. 2 Oscillator Freq. 1 Envelope Detector Envelope Detector Select Largest Filter Freq. robust Noncoherent FSK works well on fading channel Special case: Continous Phase FSK 30 . n Wireless Channel Filter Freq. 1 Filter Freq.

the BER is Pb (error) = 1 2 erfc Eb . 2 N0 Local-mean BER Pb (error ) = 1 1 2 2 p0 T b p0 T b + 2 N 0 • Coherent FSK is 3 dB less immune to noise than BPSK (FSK is orthogonal.Coherent Frequency Shift Keying (FSK) Instantaneous BER For LTI AWGN. BPSK is antipodal) 31 .

Non-coherent FSK Instantaneous BER For LTI AWGN. the BER is approximately Pb (e) = 1 2  p0 T b  exp   2 N0  Local-mean BER Pb(error) = N0 p0 T b + 2 N 0 32 .

MSK (Red) 33 .Continuous Phase FSK (CP FSK) • • • CP reduces spectral sidelobes Data is embedded in the "phase trajectory" Can be interpreted as Phase Shift Keying or as Frequency Shift Keying 2hπ hπ −hπ −2hπ Phase Time 1 1 0 0 0 1 0 Phase trajectory for • • Binary Continuous Phase Frequency Shift Keying (blue) Filtered CP FSK to reduce spectralsidelobes.

. each with a sinusoidal envelope • Very compact spectrum h is chosen as small as possible.Minimum Shift Keying (MSK) • Special case of CP FSK: phase trajectory is 'filtered' • Special case of phase modulation MSK can be interpreted as a form of two time-shifted binary PSK signals. Hiperlan. h = 0. but large enough to ensure orthogonal signal waveforms.. 34 ..5 • • • • Robust against interference Anti-podal signals: robust against noise Constant envelope Used in GSM..

MSK. QPSK and Offset QPSK During the transmission of a single bit the MSK signal smoothly increases or reduces the signal phase by 90 °. abrupt phase changes of 90 or 180° can ° occur at the end of each symbol transmission (2 bits) In Offset QPSK. abrupt phase changes of 90 occur but twice ° as frequent as in normal QPSK. In normal QPSK. 35 .

INTERFERING BPSK SIGNALS The effect of interference differs significantly from the effect of AWGN Wanted BPSK signal with a single interferer. Effect of interfering BPSK Signal depends on • Relative amplitude • Phase offset • Bit timing offset 36 .

37 .INTERFERING FADING CARRIERS Received signal r(t) = a ρ cos( ω 0 0 c t + φ )+ ∑ ρ i cos i=1 N (2π ( f + f )t+φ )+ n c i i I with bit a0 = +1. (zero-mean Gaussian) ζi = ρi sin θi (zero-mean Gaussian) and Eξi2 = Eζi2 = i. After matched filtering. the decision variable becomes v = a 0 ρ 0 T b + ∑ ξ i ∫ cos( ω i t)dt i=1 0 N Tb + ∑ ζ i ∫ sin( ω i t)dt + n I i=1 0 N Tb where ξi and ζi are inphase and quadrature component of i-th interfering signal with ξi = ρi cos θi. Variance of the interference term isΣi Tb2 sincfiTb.

αi uniformly distributed between 0 and 1. Decision variable for a receiver in perfect lock to the wanted signal: ν = ρ0 κ 0 + ∑ ζ i κ i α i+κ i→ (1-α i ) + n I .Effect of BPSK Interference • Bit synchronisation offset αi . i=1 n { } This contains components from previous and next bit interfering signal Remarks: • • Interference is NOT Gaussian Interference cancellation is easier if bit timing is synchronous • Interference affects synchronization. it does not "average out" in the receiver PLL 38 .

2 α 1 ) + N 0 • We only modeled the effect on the decision variable. • If there is a timing offset. k the variance of the interference sample is pt(1 .2αi)2. and the receiver locks perfe ctly to the wanted signal 1 1 2 4 p0 T b p0 T b + p1 T b + N 0 the bit error probability is Pb = - 1 4 p0 T b . the interference is Rayleigh fading. interference components are a little weaker Assumption • • • "0"'s and "1"'s are equiprobable and independent. not on the timing recovery In practice the effect on receiver synchronization may be more significant • 39 . 2 p0 T b + p1 T b (1 .Average BER for synchronization offset • Bit-timing offset αi is uniformly distributed for 0 < αi < 1 If phase reversal occurs at the instant ( + αi)Tb.

8 dB weaker than synchronous interference 40 . Random Offset For a Rayleigh-fading wanted signal. the average BER is Pb = 1 1 2 4 p0 T b p0 T b + p1 T b + N 0 + 1 4 p0  p0 T b + p1 T b + N 0 .p1 T b   ln    p1  p0 T b + N 0  • Non-synchronous interference is effectively 1.Average BER for Sync.

Variance of interference • If an interfering signal makes a phase reversal.Many interfering signals • • Relevant case in Spread Spectrum systems Many weak signals may not have such a dramatic effect on receiver synchronization • CLT: The sum of many weak signals is aGaussian distributed random variable. the variance of the interference sample is E[ζi2 (1 − 2αi )2 ] = p1 3 • If no phase reversal occurs. the variance is E[ζi2 ] = p1 41 .

Many interfering signals Local-mean BER Pb = 1 1 2 2 p0 T b p0 T b + 2 pt T b + N 0 3 For Spread Spectrum with Random Codes Pb = 1 1 2 2 p0 T b 2 p0 T b + 3C pt T b + N 0 where • • C is the spread factor Chips of different users have random timing offset 42 .

Effect of Intersymbol Interference The impulse response of a discrete-timemultipath channel. h(t) = ∑ h δ (t+ k T ) k b k=0 ∞ Example of (discrete time) impulse responsemultipath channel and frequency transfer function. 43 .

even if no noise is present. • If perfect synchronization to the first (resolvable) path. 44 .Effect of Intersymbol Interference Irreducable BER • Excessively delayed reflections causeintersymbol interference (ISI). the decision variable becomes ∞ ν n = a n h0 + ∑h k=1 k a n-k For a known signal power p0 in the first resolvable path. • If unequalized. this results in a residual BER. the BER becomes Pb (error| p0 ) = 1 erfc 2 k=1 ∑ pk T b + N 0 ∞ p0 T b .

Signal power density spectrum S(f) AWGN channel with frequency transfer functionH(f) Zero Forcing filter 1/H(f) Received power spectrum R(f) = S(f) + N0 H(f)H * (f) Thus signal to noise ratio  S 1 Eb T b  ∫  =  | H(f)|2 df  N N 0  BT  -1 45 . • Disadvantage: it excessively enhances the noise. Noise penalty due to channel equalization `zero forcing equalizer' • eliminates the effect ofintersymbol interference altogether.Effect of Intersymbol Interference 2.

Equalizer Approaches • Best equalization minimizes the signal-to-interferenceplus-noise. but this is not necessarily the best receiver design: • Maximum likelihood (ML) detector Find the bit sequence that was most likely • Minimize the Mean Square Error(MSE) caused by ISI and noise LMS: Least mean square algorithm recursively adjusts tap weights 46 .

Sign up to vote on this title
UsefulNot useful