Chap6 5to6 11

6.
5 Coherent frequency-shift keying – Error

probability of binary FSK
o Error probability of Binary FSK
© Po-Ning Chen@cm.nctu Chapter 6-101
6.5 Coherent frequency-shift keying – Error

o Error probability of binary FSK
n Based on the decision rule
1
o Comparison between BPSK and BFSK
3 dB difference
6.5 Coherent frequency-shift keying – Power

spectra of binary FSK
o Power spectra of binary FSK
n Assumption: f1 and f2 differ by 1/Tb.
(See Slide 6-99.)
n Under such assumption,
2
o General time-averaged power spectra
n The text derives the (time-averaged) power spectra of

the baseband signal as the sum of the in-phase power
spectra and the quadrature power spectra.
n This may not be “correct” in general (See the next

Slide).

3
The cross-correlation also affects the resultant power spectra.

n Since in-phase and quadrature components are
independent, and one of them is zero-mean (See the
next slides), the technique used in text is applicable to
binary FSK.
4
n Assumption: f1 and f2 differ by 1/Tb.
(See Slide 6-99.)
n Under such assumption,
o Equivalent baseband signal
5
(See Slides 6-35 and 6-37.)
6
(From Slide 6-38.)
1. Decay much faster (as 1/f 4)

2. Rule of Thumb: The smoother the
Figure 6.5 pulse (transition), the faster the drop
of spectral tails.
This should be

o Final note
n If BFSK is not continuous phase (due to f1 and f2 are not
multiple of 1/Tb), then the “rule of thumb” indicates that
the spectral decay should be slower.
n In fact, the rate of spectral decay for non-continuous-
phase FSK will become 1/f 2.
7
6.5 Coherent frequency-shift keying – Memoryless
versus continuous-phase
o Memoryless versus continuous-phase

From the previous slide,
In order to maintain phase-continuity, s(t) must be of the form:
8
From the previous slide,
In order to maintain phase-continuity, s(t) must be of the form:

o So, in order to maintain phase-continuity subject to that fc is
a multiple of 1/Tb, we obtain:
Require “memory”
of all histories.
o Memoryless, hence, requires:
9
o This kind of continuous-phase and memoryless FSK (with h
integer) is called Sunde’s FSK.
o Sunde’s FSK is a special case of the continuous-phase FSK

(CPFSK) family.
o For general CPFSK, the system requires to memorize

Phase tree of CPFSK

10
6.5 Coherent frequency-shift keying – Minimum
shift keying
o Minimum shift keying
n The passband signals respectively for In = -1 and In =
+1 are better to be coherent orthogonal, i.e., we require

shift keying
n h = ½ is the minimum h that satisfies “coherent orthogonality”
condition; hence, it is named minimum shift keying.
11
shift keying

shift keying
From textbook (Table 6.5)
12
Chapter 6-125
13
14
shift keying
For simplicity, assume q (0) = 0.
6.5 Coherent frequency-shift

keying – Minimum shift keying
15
Chapter 6-131
Chapter 6-132
16
Therefore,
By following similar procedure,

probability of MSK
o Error probability of MSK (Decision rule)
17
Figure 6.29
o Error probability of MSK (Quick derivation)
18
o Error probability of MSK (Direct derivation)
n Based on the decision rule

probability of MSK
o Final note on error probability of MSK
n
n In such case, the error rate of MSK becomes the one

indicating in Eq. (6.127) as:
19
6.5 Coherent frequency-
shift keying – Block
diagrams

spectra of MSK
o Power spectra of MSK
n No cross-correlation between gI and gQ (cf. Slide 6-106)
20
spectra of MSK
o MSK decays as
the inverse
fourth power of Figure 6.9
frequency.
o QPSK decays as
the inverse
(See Slide 6-38 and (6.40) in text.)
second power of
frequency.
6.5 Coherent frequency-shift keying – GMSK

o Gaussian-filtered MSK (GMSK)
n MSK has the merits of
o Constant envelope
o Relatively narrow bandwidth (compared with QPSK)
o Same coherent detection performance as QPSK
n Can we further improve out-of-band characteristics of
MSK (to fulfill the stringent requirements of certain
applications such as GSM)?
o Answer: GMSK
21
Gaussian MSK
Filter modulation
Gaussian filter
22
o Approximate
(truncate and
time-shift) the
noncausal
Gaussian filter
by a causal filter
n Shift in time
by 2.5Tb and
truncate at
+2.5T
- b

o More compact in power
spectra for time-
bandwidth product WTb
less than unity
23
o In the limiting case, GMSK corresponds to the case of ordinary
MSK.

o Error probability of GMSK
It is known from Slide 6-138 that
Assume that
Find a empirically.
24
Performance degrades (from MSK) due to intersymbol interference
that is introduced by Gaussian filter.
Figure 6.34
GSM
0.46 dB (a=1.8)
0.3

o Power spectra of GMSK
for GSM
Down by more than
40 dB at adjacent
subchannels
Figure 6.35
25
6.5 Coherent frequency-shift keying – M-ary FSK
o M-ary FSK
E is the transmitted energy per symbol, and

T is the symbol duration.
o Orthogonality

o Error probability bound of M-ary FSK
(See the next slide.)

o Power spectra of M-ary FSK (No derivation is provided!)
26
(1)
Union (upper) bound
Lower bound
Upper bound = Lower bound, when M = 2
Appendix (Excluded from exam)
This is outside the current scope of the text. Just provide it for those
who are interested in the derivation.
27

o Spectral efficiency of M-ary FSK
M 2 4 8 16 32 64
r 1 1 0.75 0.5 0.3125 0.1815
Larger M implies worse spectral efficiency.
28
6.6 Detection of signals with unknown phase
o How to deal with unknown phase q, e.g., in FSK?
n Answer: Noncoherent receiver

o How to remove the requirement of phase information at
noncoherent receiver?
o Answer: Take the expectation with respect to all
possible q.

o (Conditional) likelihood ratio test
n For known q,
29
o However, q is unknown! So, let’s average it out.
Hence,
Since
where
30
Assume q is uniform distributed over [-p, p).
The modified Bessel function of zero

kind is a monotonically increasing
function.
The modified Bessel function of

zero kind is a monotonically
increasing function.
The receiver is therefore named as quadratic receiver.
31

o Alternative realization of quadratic receiver
n Quadrature receiver using (coherent) matched filter
32
o Another alternative realization of quadratic receiver
n Noncoherent matched filter
y(t)

o Envelope detector = squarer + lowpass filter + square-rooter
o Final note
n The merit of noncoherent matched filter over coherent matched filter
is that the latter actually samples the output before the lowpass filter
(i.e., high-frequency signal) while the former samples the output after
the lowpass filter (i.e., true envelope signal). Hence, the latter has a
much higher demand on the accuracy of sampling time.
33
6.7 Noncoherent orthogonal modulation
o Definition of noncoherent orthogonal modulation
n The signals remain orthogonal and have the same
energy regardless of the unknown carrier phase.
n Example. Binary FSK introduced previously

o Noncoherent matched filter
s1(t) and s2(t) are orthogonal.

34
o Noncoherent matched filter = Quadrature receiver if
6.7 Noncoherent orthogonal modulation & 6.8

Noncoherent binary frequency-shift keying– Error rate of
noncoherent receiver
o Error rate of noncoherent receiver for

binary orthogonal modulated signals
n Assume s1(t) is transmitted and q is
known.

xI xQ See Slide 6-162.
35
6.7 Noncoherent orthogonal modulation & 6.8
Noncoherent binary frequency-shift keying– Error rate of
noncoherent receiver
o Based on the decision rule
All four are independent.
(Cf. Section 1.12)
Similarly,
Consequently,
Note that the resultant error rate has nothing to do with q.
36
6.9 Differential phase-shift keying
o Transmitter of differential phase-shift keying
o Receiver of differential phase-shift keying
37
All four are independent.
© Po-Ning Chen@cm.nctu
All four are also independent. Chapter 6-176
38
Similarly,
Consequently,
Note that the resultant error rate has nothing to do with q.
6-10 Comparison of digital

modulation schemes using a
single carrier
o The performance degradation

from coherent to noncoherent
counterpart is less than 1 dB
(101/10=1.259). 3 dB
< 1 dB
39
6-10 Comparison of digital modulation schemes
using a single carrier
o The power-bandwidth requirement of M-ary PSK with
respect to binary PSK
n M = 4 offers the best tradeoff between power and
bandwidth requirement, which explains why QPSK is
widely used in practice.
Under the same Tb (Slide 6-51) (Slide 6-45)

(Or Slides 6-64 ~ 6-65) (Slide 6-66)
6-10 Comparison of digital modulation schemes

using a single carrier
o Comparison between M-ary PSK and M-ary QAM
n M-ary QAM outperforms M-ary PSK as can be easily
seen from the below constellation.
n However, M-ary QAM requires higher linearity in, e.g.,
power amplifier design.
40
6.11 Voiceband (PSTN) modems
o Modem to/from an ISP over PSTN
n The connection from a home to the central office
(coined as local loop) remained analog nowadays.
n Example Study
o Symmetric modem (1991): V.32

o Two modulation schemes of V.32
n Nonredundant coding
o 2 inputs à 2 code bits in QPSK
o 4 inputs à 4 code bits in 16QAM
n Trellis coding
o 4 inputs à 5 code bits
41
o V.32 16-QAM = Hybrid amplitude/phase modulation
scheme
Quadrant code

o Example
(See the Table above.)
Operational example at the receiver
42
o V.32 Trellis coding = Hybrid 90-degree rotational
amplitude/phase modulation invariant with respect to
scheme : 4 dB coding gain over Q3,n = Y3,n and Q4,n = Y4,n .
16 QAM at high SNR
11 00
01 10
43
o Rotationally invariant
n “Rotation” of the constellation points must be “invariant’
with respect to Q3,n = Y3,n and Q4,n = Y4,n .
For every (000xx) to (111xx),

90-degree counterclockwise rotation
is observed.
0
correction
44
For every (110xx) to (000xx),
90-degree counterclockwise rotation
is oberved.
0
0
© Po-Ning Chen@cm.nctu

o At the receiver end, a convoludional decoder is performed
first, followed by the differential decoder.
o In presence of a (fixed) phase error, e.g., 90 degree, as long
as the phase transition between two consecutive signal
points remains the same, the differential decoder can
correctly determine Q1,n and Q2,n based on the output of the
convolutional decoder.
45
n Example Study
o Asymmetric modem
n The communication between PSTN and ISP becomes
digital.
6.11 Voiceband (PSTN)

modems
o A digital modem and
analogue modem pair for
use on the Public
Switched Telephone
Network (PSTN) at data
signalling rates of up to
56 000 bit/s downstream
and up to 33 600 bit/s
upstream
46
o Typical realization of digital modem
n With PCM sampling rate = 8 KHz, and 256 levels per
PCM sample, the PCM should result in 64 Kbps in
theory.
n However, since the conventional PCM transmit filter
has a bandwidth of about 3.5 KHz, the theoretical
speed of 64 Kbps cannot be achieved.
o One solution: Use the recurrent non-uniform
equivalent form of the sampling theorem.

o The below non-uniform sampling is equivalent to the
sampling rate of 7 KHz.
47
o The bandlimited signal is now interpolated as:
n Properties of the seven standard pulse functions
No ISI
Convolution once
48
o The above realization resolves the A/D problem, subject to
the 3.5 KHz bandwidth constraint of the transmit filter.
o How about D/A? Is 64 Kbps achievable for the digital
modem?
n Still, 56 Kbps is the feasible rate.
o The reconstructed voice as below is susceptible to
(quantization) residual ISI.

o Most importantly, LSBs in each octet are robbed
from the data stream for various purposes internal to
the PSTN (which is called “bit-robbing”, and which
makes worse the susceptibility to residual ISI).
49
n Example Study
o Asymmetric modem
V.34 modem (1998)

o Due to the 3.5KHz bandwidth restriction of anti-aliasing
and interpolation filters, and the quantization noise, the rate
of the analog modem operates at 33.6 kbps with the below
features:
n 960-QAM super-constellation
o You may view it as a selection of 240 message points,
and their rotated versions through 90, 180, and 270
degrees.
50

Chap6 5to6 11

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Chap6 5to6 11

Încărcat de

Drepturi de autor:

Formate disponibile

6.

5 Coherent frequency-shift keying – Error

© Po-Ning Chen@cm.nctu Chapter 6-101

6.5 Coherent frequency-shift keying – Error

© Po-Ning Chen@cm.nctu Chapter 6-102

© Po-Ning Chen@cm.nctu Chapter 6-103

6.5 Coherent frequency-shift keying – Power

© Po-Ning Chen@cm.nctu Chapter 6-104

n The text derives the (time-averaged) power spectra of

n This may not be “correct” in general (See the next

© Po-Ning Chen@cm.nctu Chapter 6-105

6.5 Coherent frequency-shift keying – Power

© Po-Ning Chen@cm.nctu Chapter 6-106

The cross-correlation also affects the resultant power spectra.

© Po-Ning Chen@cm.nctu Chapter 6-107

6.5 Coherent frequency-shift keying – Power

© Po-Ning Chen@cm.nctu Chapter 6-108

© Po-Ning Chen@cm.nctu Chapter 6-109

o Equivalent baseband signal

© Po-Ning Chen@cm.nctu Chapter 6-110

© Po-Ning Chen@cm.nctu Chapter 6-111

© Po-Ning Chen@cm.nctu Chapter 6-112

1. Decay much faster (as 1/f 4)

© Po-Ning Chen@cm.nctu Chapter 6-113

6.5 Coherent frequency-shift keying – Power

© Po-Ning Chen@cm.nctu Chapter 6-114

© Po-Ning Chen@cm.nctu Chapter 6-115

6.5 Coherent frequency-shift keying – Memoryless

In order to maintain phase-continuity, s(t) must be of the form:

© Po-Ning Chen@cm.nctu Chapter 6-116

In order to maintain phase-continuity, s(t) must be of the form:

© Po-Ning Chen@cm.nctu Chapter 6-117

6.5 Coherent frequency-shift keying – Memoryless

© Po-Ning Chen@cm.nctu Chapter 6-118

o Sunde’s FSK is a special case of the continuous-phase FSK

o For general CPFSK, the system requires to memorize

© Po-Ning Chen@cm.nctu Chapter 6-119

6.5 Coherent frequency-shift keying – Memoryless

Phase tree of CPFSK

© Po-Ning Chen@cm.nctu Chapter 6-121

6.5 Coherent frequency-shift keying – Minimum

© Po-Ning Chen@cm.nctu Chapter 6-122

© Po-Ning Chen@cm.nctu Chapter 6-123

6.5 Coherent frequency-shift keying – Minimum

From textbook (Table 6.5)

© Po-Ning Chen@cm.nctu Chapter 6-124

© Po-Ning Chen@cm.nctu Chapter 6-126

© Po-Ning Chen@cm.nctu Chapter 6-128

© Po-Ning Chen@cm.nctu Chapter 6-129

6.5 Coherent frequency-shift

© Po-Ning Chen@cm.nctu Chapter 6-130

By following similar procedure,

© Po-Ning Chen@cm.nctu Chapter 6-133

6.5 Coherent frequency-shift keying – Error

© Po-Ning Chen@cm.nctu Chapter 6-134

© Po-Ning Chen@cm.nctu Chapter 6-135

o Error probability of MSK (Quick derivation)

© Po-Ning Chen@cm.nctu Chapter 6-136

© Po-Ning Chen@cm.nctu Chapter 6-137

6.5 Coherent frequency-shift keying – Error

n In such case, the error rate of MSK becomes the one

© Po-Ning Chen@cm.nctu Chapter 6-138

© Po-Ning Chen@cm.nctu Chapter 6-139