This action might not be possible to undo. Are you sure you want to continue?
Pulse Modulation
Problem 3.1
Let 2W denote the bandwidth of a narrowband signal with carrier frequency Ie The inphase and quadrature components of this signal are both lowpass signals with a common bandwidth of W According to the sampling theorem, there is no information loss if the inphase and quadrature components are sampled at a rate higher than 2 W For the problem at hand, we have
t; = 100 kHz 2W= 10 kHz
Hence, W = 5 kHz, and the minimum rate at which it is permissible to sample the inphase and quadrature components is 10 kHz.
From the sampling theorem, we also know that a physical waveform can be represented over the interval 00 < t < 00 by
(1)
n=oo
where {<!>n(t)} is a set of orthogonal functions defined as
sin {nlsCt  nils)} nlsCt  nils)
where n is an integer and j, is the sampling frequency. If get) is a lowpass signal bandlimited to W Hz, and j, ;:::: 2W, then the coefficient an can be shown to equal g(n/fs). That is, for Is;:::: 2W, the orthogonal coefficients are simply the values of the waveform that are obtained when the waveform is sampled every 1115 second.
As already mentioned, the narrowband signal is twodimensional, consisting of inphase and quadrature components. In light of Eq. (1), we may represent them as follows, respectively:
143
n=oo
Hence, given the inphase samples g {;J and quadrature samples g Q(;J ' we may reconstruct the narrowband signal get) as follows:
00
= L [g{;J cos(2nIet)  gQ(;J sin (2nIet)]<I>n(t)
n=oo
where t; = 100 kHz and Is ~ 10 kHz, and where the same set of orthonormal basis functions is used for reconstructing both the inphase and quadrature components.
144
Problem 3.2
(a) Consider a periodic train c(f) of rectangular pulses, each of duration T. The Fourier series expansion of c(t) (assuming that a pulse of the train is centered on the origin) is given by
00
c(t) = L fs sinc(nfs T) exp(j 21tnfst)
n=oo
where fs is the repetition frequency, and the amplitude of a rectangular pulse is assumed to be ltr (i.e., each pulse has unit area). The assumption that fsT»l means that the spectral lines (i.e., harmonics) of the periodic pulse train c(t) are well separated from each other.
Multiplying a message signal g(t) by c(t) yields
s(t) = e(t)g(t)
= L fs sine (nfs T)' g(t) expG21tnfs(t)
(1)
n=oo
Taking the Fourier transform of both sides of Eq .. (1) and using the frequencyshifting property of the Fourier transform:
00
8(0 = L r, sine(nfs T) G(f nfs)
(2)
n= 00
where G(f) = F[g(t)]. Thus, the spectrum S(f) consists of frequencyshifted replicas of the original spectrum G(f), with the nth replica being scaled in amplitude by the factor fssinc(nfsT).
145
(b) In accordance with the sampling theorem, let it be assumed that
• The signal g(t) is bandlimited with
G(f) = 0 for  W < f < W
_ The sampling frequency f8 is defined by
Then, the different frequencyshifted replicas of G(f) involved in the construction of S(f) will not overlap. Under the conditions described herein, the original spectrum G(f), and therefore the signal g(t), can be recovered exactly (except for a trivial amplitude scaling) by passing s(t) through a lowpass filter of bandwidth W.
Problem 3.3
(a) get) = sinc(200t):
This sinc pulse corresponds to a bandwidth W and the Nyquist interval is 1/200 seconds.
(b) get) = sinc2(200t}:
100 Hz. Hence, the Nyquist rate is 200 Hz,
This signal may be viewed as the product of the sinc pulse sinc(200t)." with itself. Since multiplice.tion in the time domain corresponds to convolution in the frequency domain, we find that the signal get) has a bandwidth equal to twice that of the sinc pulse sin(200t):~ that is, 200 Hz. The Nyquist rate of get) is therefore 400 Hz, and the Nyquist interval is 1/400 seconds.
(c) get) = sinc(200t): + sinc2(200t}:
The bandwidth of get) is determined by the highest frequency component of sinc(200t) or sinc2(200t):~. whichever one is the largest. With tile bandwidth (Le., highest frequency component of) the sinc pulse sinc(200 t) equal to 100 liz and that of the squared sinc pulse sinc2(200t}:: equal to 200 Hz, it follows thaL the bandwidth of get:) is 200 Hz. Correspondingly, the Nyquist rate of get) is 400 lIz, and its Nyquist interval is 1/400 seconds.
146
Problem 3.4
(a) The PAM wave is
co
set) = E [1 + ~m'(nTs)]g(tnTs)'
n=_co
where get) is the pulse shape, and m'(t) = m(t)/A = cos(2'1rf t). The PAM wave is
m m
equivalent to the convolution of the instantaneously sampled [1 + um! (t)] and the pulse
shape g( t) :
co
set) = { E
n=_co
{;I. get)
co
= {[1+~m'(t)] E
n=co
6(tnT )} ~ get)
s
The spectrum of the PAM wave is,
S( f)
co
= Ir 6( f) + ~M' (f)] «::r t 1:
s m=_co
m
6(fI)} G(f)
s
co
= t G ( f) E [ 6 (f  ~ ) + ~M' (f _ ~ )]
s m=_CO s s
For a rectangular pulse get) of duration T=0.45s, and with AT = 1, we have:
G(f) = AT sinc(fT)
= sinc(0.45f)
For
m'(t) = cOs(2'1rfmt), and with fm.= 0.25 Hz, we have: 1
M'(f) = 2 [6(f0.25) + 6(f+0.25)]
co
1s, the ideally sampled spectrum is S6(f) =
sm m=_co
s
[6(fm) + ~M'(fm)].
I
4'~A /,1.
JI . lot . __ 1_
 .15 . O.i~ 0.2.5
The actual sampled spectrum is
o
147
00
S~f) = 1: sinc(O.45f)[6(fm) + ~M'(fm)]
m=OII SC;)
6.157 o.~5~ " .'l~"'~ o.'18~ o·8SEA O.iS7
O.b37)(. g.~3~
1 2. 2 r 1
2 l. ! __ t a
t ! t :f ( H 1)
1.2!' 1.0 0. 7) o.~ 0 O.2S O.'~ \.0 I.'l{ (b) The ideal reconstruction filter would retain the centre 3 delta functions of S(f) or:
O. 'I'all..iu r o. ~gi;.LI.
2.' 
t 2
lf
0.1$ 0 O. "2. 5' With no aperture effect, the two outer delta functions would have amplitude~. Aperture effect distorts the reconstructed signal by attenuating the high frequency portion of the message signal.
Problem 3.5
The spectrum of the flattop pulses is given by
H(f) = Tsinc(fF)exp(jnfF)
Let set) denote the sequence of flattop pulses:
n=r=
The spectrumS(!) = F[s(t)] is as follows:
S(f) = i, L M(f  kfs)H(f)
k=oo
k=oo
The magnitude spectrum IS(f)1 is thus as shown in Fig. lc. 148
o
l/T
(a)
liT
f
(b)
S(f)
liT
(c)
l/T
f
lIT = 1O,OOOHz fs = 1,OOOHz W = 400Hz
Figure 1
149
Problem 3.6
. m nent of the message signal for a
Atf= 1/2T" which corresponds to. the hlghest~rei~re:~y ~o. (:'q) that the amplitude response
sampling rate equal to the Nyquist rate. we n q
1.2
sine (0.5TITsl 1 .1
0.2
0.4
Duty cycle TITs
0.6
0.8
a
Figure'
of the equalizer normalized to that at zero frequency is equal to
I (rr/2)(T/T,J
sinc(0.5 itt; sin[ (rr/2)( T/'fs)]
where the ratio T[T, is equal to the duty cycle of the sampling pulses. In Fig. t , this result is plotted as a function of Tl'T; Ideally, it should be equal to one for all values of Tl'T; For a duty cycle of 10 percent, it is equal to 1.0041. It follows therefore that for duty cycles of less than 10 percent, the aperture effect becomes negligible, and the need for equalization may be omitted altogether.
150
Problem 3.7
Consider the fullload test tone A cos(2lTfmt). Denoting the kth sample amplitude of this signal by Ak, we find that the transmitted pulse is Ak g(t), where g(t) is defined by
the spec trl.lll :
G(f) =
0, otherwise
The mean value of the transmitted signal power is
LT L Ak g(t)]2dt1
{. 1 J s [ E
P = E l~m 2LT
L+o> s LT k= L
s
LT L L g2(t)dt]
1 s E E AkAn
= E[lim 2LT J
L+o> s LT k=L n=L
s
L L LT l(t)dt
1 E E E[AkA ] J s
= lim 2LT
k=L n=L n LT
L+o> s s where Ts is the sampling period. However,
\A2
2 '
0,
k = n
otherwsie
Therefore,
2 IXI
P = 2~ J g2(t)dt
S _IXI
Using Rayleigh's energy theorem, we may write
IXI IXI
J g2(t)dt = J IG(f)12df
_IXI
2
df
151
Therefore,
P =
The average signal power at the receiver output is A2/2. Hence, the output signaltonoise ratio is given by
(SNR)O
2T P s
=~
By·choosing BT=1/2Ts' we get P
(SNR)O = BTNO
This shows that PAM and baseband signal transmission have the same signaltonoise ratio for the same average transmitted power, with 'additive white Gaussian noise, and assuming the use of the minimum transmission bandwidth possible.
Problem 3.8
(a) The sampling interval is T = 125 ~s. There are 24 channels and 1 sync pulse, so the time alloted to each channel i~ T = T 125 = 5 ]..Is. The pulse duration is 1 ~s, so the
c s
time between pulses is 4 ]..Is.
(b) If sampled at the nyquist rate, 6.8 kHz, then Ts = 147 ]..IS, Tc = 6.68 ]..Is, and the time between pulses is 5.68 ~s.
Problem 3.9
(a) The bandwidth required for each single sideband channel is 10 kHz. bandwidth for 12 channels is 120 kHz.
The total
(b) The Nyquist rate for each signal is 20 kHz. For 12 TDM signals, the total data rate is 240 kHz. By using a sinc pulse whose amplitude varies in accordance with the modulation, and with zero crossings at multiples of (1/240) ms, we need a minimum bandwidth of 120 kHz.
152
Problem 3.10
(a) The Nyquist rate for S1(t) and S2(t) is 160 Hz. 160, and the maximum R is 3.
2400
Therefore, ~ must be greater than
(b) With A = 3, we may use the following signal format to multiplex the signals s1(t) and S2(t) into a new signal, and then multiplex S3(t) and s4(t) and S5(t) including markers for synchronization:
5
300
!
(V2400)sr
Sa Si $\ ~ S" Sl $3 5.", s~
_J ..
H'/1200)S .sa.;r,p& s
I • zero Based on this signal format, we may develop the following multiplexing system:
2~oo Hz _r::gi _!._s 25
c~oc.lt ...
2400 2<400
~ l)eQ"j J)J~ \
Ma.rRH 1 $O.mfi¢ 6(1:) so..rr.rt9  s (t)
1genefCltoi I 2
\ ~M sir)
u Sa.I'I)P~o/If
 X
"....; j
.j, :
I '5 4 M
7200S
~ 7200 S(~} +1 SOJ'r\ptt.A U ~
~Jfl:J », ~".J 3 ~ X.

S (!:) J
SQmp'~
.. Iooi t 153
Time
M \A ~ b' j(p t.X4c1. S~t\~
Problem 3.11
In general, a line code can be represented as
N
s (r) = L ang(t  nT b)
n=N
Let g(t) .# G(f). We may then define the Fourier transform of set) as
S(f)
N . T
loon b
= L anG(f)e
n=N
N . T
loon b
= G(f) Lane
n=N
where co = 2nJ The power spectral density of s(t) is
where T is the duration of the binary data sequence, and E denotes the statistical expectation operator. Define the autocorrelation of the binary data sequence as
By letting m = n + k and T = (2N + l)Tb' we may write
Replacing the outer sum over the index n by 2N+ 1, we get
S (f) = IG(f)12 lim [2N+1 kk=I_NN_nnR(k)/kOOTb]
S T b N 7 00 2N + 1
154
~ »«,
L.J R(k)e
(1)
k=oo
where
I
R(k) = E[anan + k] = L (anan + k)Pi i=l
(2)
where Pi is the probability of getting the product (an an+k)i and there are I possible values for the an an+k product. G(j) is the spectrum of the pulseshaping signal for representing a digital symbol. Eqs. (1) and (2) provide the basis for evaluating the spectra of the specified line codes.
(a) Unipolar NRZ signaling
For rectangular NRZ pulse shapes, the Fouriertransform pair is
For unipolar NRZ signaling, the possible levels for a's are +A and o. For equiprobable symbols, we have the following autocorrelation values:
4
R(k) = L(anan+k)iPi i=l
A2 0 0 0 A2
=  +  +  +  =  for Ikl > 0
4 4 4 4 4
Thus
R(k)
= {
for k = 0 for k * 0
(3)
155
Therefore, the power spectral density for unipolar NRZ signals, using formulas (1) and (3), is
2 [00 ]
A T b 2 j2nkfT b
= 4sinc (fT b) 1 + k~oo e
But,
where 8(j) is a delta function in the frequency domain. Hence,
n
We also note that sinc(fT b) = 0 at f = , n 7; 0 ; we thus get
Tb
(b) Polar Nonretumtozero Signaling
For polar NRZ signaling, the possible values for a's are +A and A. Assuming equiprobable symbols, we have
2
R(O) = L(anan)iPi
i=l
A2 (_A)2 2
=+=A
2 2
156
For k 7; 0 , we have
4
R(k) = L(anan+k)iPi i=l
= A2 + 2(A)(A) + 2(A)(A) + (_A)2
4 4 4 4
= 0
Thus,
R(k) ~ { ~2
for k = 0 for k::j:. 0
(4)
The power spectral density for this case, using formulas (1) and (4), is
(c) Returntozero Signaling
The pulse shape used for returntozero signaling is given by g( T :12) . We therefore have
The autocorrelation for this case is the same as that for unipolar NRZ signaling. Therefore, the power spectral density of RZ signals is
(d) Bipolar Signals
The permitted values of level a for bipolar signals are +A, A, and 0, where binary symbol 1 is represented alternately by +A and A, and binary 0 is represented by level zero. We thus have the following autocorrelation function values:
157
2
R(O) = ~
4 2
R(1) = I,(anan+ l)iPi = ~
i=l
For k > 1,
Thus,
A2 for k = 0

2
R(k) = A2 j (5)
 for Ikl = 1
4
0 for Ikl > 1 The pulse duration for this case is equal to T il2. Hence,
(6)
Using Equations (1), (5) and (6), the power spectral density of bipolar signals is
2
A Tb. 2(fTb)[ 1 jwTb jWTbJ
=  SInC  1   (e + e )
8 2 2
2
A r; 2(fTb)
= 8sinc 2 [1 cos (2TCfTb)]
158
2
A r; 2(fTb) 2
= 8sinc 2 sin (nfTb)
(e) Manchester Code
The permitted values of a's in the Manchester code are +A and A. Hence,
121 21 212
R(O) = A + (A) + (A) + (A)
4 4 4 4
For k:f:. 0,
422
R(k) = ~(a a ).p. = ~ + (A)(A) + A(A) + (A)
L.; n n+k: I I 4 4 4 4
i=l
= 0
Thus,
R(k) ~ t ~2 for k = 0 }
for k:f:. 0 The pulse shape of Manchester signaling is given by
The pulse spectrum is therefore
G(f)
(fT b) (2nfT b)
= jT bsinc 2 sin 4
159
Therefore, the power spectral density of Manchester NRZ has the form
Problem 3.12
Power spectral density of a binary data stream will not be affected by the use of differential encoding. The reason for this statement is that differential encoding uses the same pulse shaping functions as ordinary encoding methods. If the number of bits is high, then the probability of a symbol one and symbol zero are the same for both cases.
Problem 3.13
(a)
set)
{ (1[~
cos 
(b) get) = T'
0,
T T
_b<t<_!!.
2  2
otherwise
Equivalently, we may write
get) = Cos(;~Arec{;~
where rect(t) is a rectangular function of unit amplitude and unit duration. The Fourier transform of get) is given by
where A denotes the pulse amplitude and * denotes convolution in the frequency domain.
160
Using the replication property of the delta function 8(j), we get
Using Eq. (1.52) of the textbook, the power spectral density of the binary data stream is
S(J)
(1)
Note that the two spectral components sine ( T b( J  ~ ~) and sine ( T b( J + ~ ~) overlap in the frequency interval (lITb) ~J ~ (lITb), hence the presence of crossproduct terms in Eq. (1).
Figure 1 plots the normalized power spectral density S(j)/(A2T ,)4) versus the normalized frequency [Tb. The interesting point to note in this figure is the significant reduction in the power spectrum of the pulseshaped data stream x(t) in the interval lITb ~J ~ 11Tb.
(c) The power spectral density of the standard form of polar NRZ signaling is
(2)
Comparing this expression with that of Eq. (1), we observe the following differences:
Polar NRZ signals using Polar NRZ signals using
cosine pulses rectangular pulses
J=O 0 A2Tb
J=±2ITb A2T,)4 0 161

:=..
CI)
0.6
1
?
~
I~
'"
~0.8
4
3
2
1
1
2
3
4
5
162
Problem 3.14
+A~
(()..)
OL·+~~+_~~A
(C.j
.. (12.)0
It
163
Problem 3.15Ca)
 r
·t·_i .:~~~'";~
t.~J+~I!t+~~l~~+""!'r~I~~~~~~F~~~~~~~C~~~~~~!. {~~~ ~ 
i
.~++ ~t  _ fr
~_""''''''_'''''_'''' __ J,.~~ I "._ .. __ ~I~r __  ~~~
_ i[ __ ~+ ~_4~L~,:r . __ ~ __
_ ~~,,
. ; .~ vr t ,' _,___L L_+_~_'__ __ L __
i '
:  L_. ; __ i
''~";~+'':"';+r~, __ ; __ c _
L_ .. __:__ ,_,. __
 . J
.. : __ ' __ :...... ' __ : ;_ .. _' _I __ f __ :_·~·_ :,:.:: .~~ ~~~.
j ~ :..
..: .. '._!__:+' +~l_j( :_,'___j;i ··'"J_:i;it'~~~~
__ : i.__: ' __ : __ : __ ._+ __ ,, __ ~,i _,' __ ,[__ .,:...._i.~~i _~ __ i_. __ ,;' __ ,:.:.' __ .• __ ',' __ ,i ,: __ >,J__ .  
;': ~, ~..:__
i,:~:·:; ~!+: ~:~: rri( _! i :_j __ ' j __ i__ _
__ ' __ i ; __ ~_+' ~~_: __ .~,L;~ ++~___!_+__;_~! J=r~;~~~+l~~J~
: i : 1 : i : : ! I : ! LLJ I I i 1 i
.;+;~+~~i___,I!~ i_____;__; 164 jHlrj+j 'I! I I ~l;
,++++~+, ! '1+1++ I i I I
,"' ',. :;,
r++r~...~r_r+~~r__++_~_rt__+~r__++_~+t__+~!__t+~'r+++j+~,I L_L~ __ L_L __ L__L~_~_L_L__L~_~_L_L__~~_~__L_
Problem 3.15Cb) .
d \ l \ 0 0 l b t 0 ,0
.n
.e.n 0 D 0 (J I t.
0 0 (I) 0 ( t
ta)+A
0 (t;.) .,., O~~~~~~~~+~~~~~~·
(e) .It
O~~~~~~~+~+~~~~~~+~~~
.. lime t
165 .. _ . . . . .... __
Problem 3.16
The minimum number of bits per sample is 7 for a signaltoquantization noise ratio of 40 dB. Hence,
(The number of samPles) = 8000 x 10
in a duration of lOs _ 8 X 104 I
samp es
The minimum storage is therefore
= 7 x 8 x 104 = 5.6 x 105 = 560 kbits
166
Problem 3.17
Suppose that baseband signal m(t) is modeled as the sample function of a Gaussian random process of zero mean, and that the amplitude range ofm(t) at the quantizer input extends from
"'41\ms to'41\ms· We find that samples of the signal m(t) will fall outside the amplitude range 81\ms with a probability of overload that is less than 1 in 104. If we further assume the use of a binary code with each code word having a length n, so that the number of quantizing levels is 2", we find that the resulting quantizer step size is
(1)
Substituting Eq. (1) to the formula for the output signaltoquantization noise ratio, we get
(2)
Expressing the signaltonoise ratio in decibels:
1010g0(SNR)o = 6R  7.2
(3)
This formula states that each bit in the code word of a PCM system contributes 6dB to the signaltonoise ratio. It gives a good description of the noise performance of a PCM system, provided that the following conditions are satisfied:
1. The system operates with an average signal power above the error threshold, so that the effect of transmission noise is made negligible, and performance is thereby limited essentially by quantizing noise alone.
2. The quantizing error is uniformly distributed.
3. The quantization is fine enough (say R > 6) to prevent signalcorrelated patterns in the quantizing error waveform.
4. The quantizer is aligned with the amplitude range from 41\ms to 41\ms.
In general, conditions (1) through (3) are true of toll quality voice signals. However, when demands on voice quality are not severe, we may use a coarse quantizer corresponding to ~ < 6. In such a case, degradation in system performance is reflected not only by a lower signaltonoise ratio, but also by an undesirable presence of signaldependent patterns in the waveform of quantizing error.
167
Problem 3.18
(a) Let the message bandwidth be W. Then, sampling the message signal at its Nyquist rate, arid using an Iibit code to represent each sample of the message signal, we find that the bit duration is
The bit rate is
The maximum val ue of message bandwid th is therefore
W max
=
50 x 106 2 x 7
= 3.57 x 106 Hz
(b) The output signaltoquantizing noise ratio is given by (see Example 2): 10 log10(SNR)0 = 1.8 + 61
=1.8+6x7
= 43.8 dB
Problem 3.19
Let a signal amplitude lying in the range
1 1
Xi  2" °i ~ x s Xi + 2" °i
be represented by the quantized amplitude xi. The instantaneous square value of the error
2
is (xxi). Let the probability density function of the input signal be fX(x). If the
step size 0i is small in relation to the input signal excursion, then fX(X) varies little within the quantum step and may be approximated by fx(xi). Then, the meansquare value of the error due to signals falling within this quantum is
168
1
xi + 2" ci 2
::: f (xxi) f x( Xi) dx
1
xi  '2 ci
1
xi + 2" ci 2
= fX(xi) f (xxi) dx
1
xi  2" ci
1
2" Ci X2
= fX(Xi) f dx
1
  15
2 i
1 3 fX(xi)
= 12 ci (1)
The probability that the input signal amplitude lies within the ith interval is
1 1
xi + 2" ci xi +  15
2 i
Pi = f f X( x) dx ::: fX(xi) f dx = fX(xi)Ci
1 1
xi   15 xi   15
2 i 2 i (2 )
Therefore, eliminating fX(xi) between Eqs , (1) and (2), we get
.The total meansquare value of the quantizing error is the sum of that contributed by each of the several quanta. Hence,
2 1 .1'2
~ E[Qi] = 12 ~ Pi vi
169
Problem 3.20
(a)
~        _ ...
2
( b)
1
    
I I
1 t _ ......  "
I'.;..__;, I" pILI . I; rJI e
r 1 ~  \"oR:~"''''
1,1·  
:: '1    rl 1"    .I II
I i·l· Q
\A. 0. A~z.e._i
I I I
I , I c::>...._. \pv,J
.,I. .,Q __ ..!....::::t:Io.od..4  ~
) ~ 0
"'09 ~
;t. ~
" ..1,
Problem 3.21
The quantizer has the following inputoutput curve:
0000
IIII
.i 5.3
At the sanpling instants we have:
t met) code
3/8 3/~ 0011
1/8 3/~ 0011
+1/8 3/~ 1100
+3/8 312 1100
And the coded waveform is (assuming onoff sign ali ng): J I J I 11. I l o
,
"8
3 r
Problem 3.22
The transmitted code words are:
t/Tb code
1 001
2 010
3 011
4 100
5 101
6 110 171
The sampled analog signal is
Problem 3.23
(a) The probability P1 of any binary symbol being inverted by transmission through the system is usually quite small, so that the probability of error after n regenerations in the system is very nearly equal to n P1' For very large n , the probability of more than one inversion must be taken into account. Let p denote the probability that a binary symbol is in error after tr ansmission through the gompl ete system. Then, p is al so the probability of an odd number of errors, since an even number of errors nrestores the original val ue. Counting zero as an even number, the probabll ity of an even number of errors is 1p. Hence
n
Pn+1 = Pn(1P1)+(1Pn)P1 = (12p1)Pn+P,
This is a linear difference equation of the first order. Its solution is
1 n
Pn = 2 [1(12p1) ]
(b) If P1 is very small and n is not too large, then n
(12p1) '" 12p1n
and
172
Problem 3.24  Regenerative repeater for PCM
Three basic functions are performed by regenerative repeaters: equalization, timing and decisionmaking.
Equalization: The equalizer shapes the incoming pulses so as to compensate for the effects of amplitude and phase distortion produced by the imperfect transmission characteristics of the channel.
Timing: The timing circuitry provides a periodic pulse train, derived from the received pulses, for sampling the equalized pulses at the instants of time where the signaltonoise ratio is maximum.
Decisionmaking: The extracted samples are compared to a predetermined threshold to make decisions. In each bit interval, a decision is made whether the received symbol is 1 or 0 on the basis of whether the threshold is exceeded or not.
Problem 3.25
m ( t) = A tanh ( B t)
To avoid slope overload, we require
~ 2. maxldm(t)1
Ts dt
(1)
dm(t) 2
_ = ABsech (Bt) dt
(2)
Hence, using Eq. (2) in (1):
2
~ 2. maxt Ab sech (Bt» x Ts
(3)
Since sech (Bt) =
1
coshtjtr)
2
=
it follows that the maximum value of sechtjiz) is 1, which occurs at time t = O. Hence, from Eq. (3) we find that ~ 2. ABTs.
173
Problem 3.26
The mod ul ating wave is
m( t) = Am cos (21rfm t) The slope of met) is
The max imum slope of m( t) is equal to 21Tf mAm.
The maximum average slope of the approximating signal ma(t) produced by the delta modulator is olTs' where 0 is the step size and Ts is the sampling period. The limiting val ue of Am is therefore given by
or
A > 0
m 21Tfm Ts
Assuming a load of 1 ohm, the transmitted power is A~/2.
Ther efore, the max imum
power that may be transmitted without slopeoverload distortion is equal to 02/8n2f2T2. m s
174
Problem 3.27
Is. = lOINyquist
INyquist = 6.S kHz
Is = 10 x 6.S x 103 = 6.S x 104 Hz
For the sinusoidal signal met) = Amsin(2nImt), we have
Hence,
or, equivalently,
Therefore,
=
=
0.1 x 6.S x 104 2n x 103
= 1.0SV
175
Problem 3.28
(a) From the solution to Problem 3.27, we have
I). is 2ni rnA
A =  or I). = 
2ni m is
(1)
A2 The average signal power = 2
With slope overload avoided, the only source of quantization of noise is granular noise. Replacing M2 for peM with I). for delta modulation, we find that the average quantization
noise power is 1).2/3; for more details, see the solution to part (b) of Problem 3.30. The waveform of the reconstruction error (i.e., granular quantization noise) is a pattern of bipolar binary pulses characterized by (1) duration = T, = Vis, and (2) average power = M3. Hence, the autocorrelation function of the quantization noise is triangular in shape with a peak value of 1).2/3 and base 2Ts' as shown in Fig. 1:
Fig. 1
From random process theory, we recall that
which, for the problem at hand, yields
=
Typically, in delta modulation the sampling rate is is very large compared to the highest frequency component of the original message signal. We may therefore approximate the power spectral density of the granular quantization noise as
176
SQ(f) Z { IJ? /s], W:::; f:::; W
. 0, otherwise
where W is the bandwidth of the reconstruction filter at the demodulator output. Hence, the average quantization noise power is
(2)
Substituting Eq. (2) into 0), we get
_ (21[f mA)2 W N  2 I, 3fs
(b) Correspondingly, output signaltonoise ratio is
Problem 3.29
177
3 ~>2xnx 10 xl
 50 X 103
= 0.126V
3 (50 X 103)3
=  X ''"
16n2 106 X 5 X 103
= 475
In decibels,
(SNR)out = 10log 10475
= 26.8 dB
Problem 3.30
(a) For linear delta modulation, the maximum amplitude of a sinusoidal test signal that can be used without slopeoverload distortion is
A ~Is
=
2nIm
3
0.1 X 60 X 10 3
= Is = 2 x 3 x 10
3
2n x 1 x 10
= 0.95V (b) (i) Under the prefiltered condition, it is reasonable to assume that the granular quantization noise is uniformly distributed between ~ and +~. Hence, the variance of the quantization noise is
178
~
2 f 1 2
(JQ = 2~q dq
~
1 3 ~
= 6~ [q ]_L\
~2
= 
3 The signaltonoise ratio under the prefiltered condition is therefore
A2/2 (SNR)prefiltered = 2 ~ 13
3A2
=
2~2
2
3 x 0.95
=
2
2 x 0.1 = l35
= 21.3 dB
(ii)The signaltonoise ratio under the postfiltered condition is
3
(8) 3 Is
N postfiltered = 16n? x I~ W
3 (60)3
=  X ''
16n2 (1)2x3
= l367
= 31.3 dB
The filtering gain in signaltonoise ratio due to the use of a reconstruction filter at the demodulator output is therefore 31.3  21.3 = 10 dB.
179
Problem 3.31
Let the sinusoidal signal m(t) = Asinwot, where Wo = 2nfo
The autocorrelation of the signal is
For this problem, we thus have
(a) The optimum solution is given by
=
= cos(O.I)
= 0.995
A2 A2 A2 2
= 2  2cos(0.1) x 2cos(0.1)/(A 12)
180
A2 2
= 2(1 cos (0.1»
= 0.005A2
Problem 3.32
[ 1 0.8 0.6j
n, = 0.8 1 0.8
0.6 0.8 1
T
rx = [0.8, 0.6, O.~
[ 1 0.8 0.6j 1 ~0.8~
= 0.8 1 0.8 0.6
0.6 0.8 1 0.4
= [0.~75J
0.125
[0.875J
= 1  [0.8, 0.6, O.~ 0
0.125
= 1  (0.8 x 0.875  0.4 x 0.125)
= 1  0.7 + 0.05
= 0.35
181
Problem 3.33
R = x
[1 0.8l 0.8 1 J
rx = [0.8,
T
0.6J
= [0.8889] 0.1111
= 1 0.6444
= 0.3556
which is slightly worse than the result obtained with a linear predictor using three unit delays (i.e., three coefficients). This result is intuitively satisfying.
Problem 3.34
Input signal variance = Rx(O)
The normalized autocorrelation of the input signal for a lag of one sample interval is
182
Rx(O)
Processing gain = 2
Rx(O)(I px(I))
1
= :
IP;(l)
1
=
1  (0.75)2
= 2.2857
Expressing the processing gain in dB, we have
10log 10(2.2857) = 3.59 dB
Problem 3.35
(a) Threetap predictor:
Processing gain = 2.8571 = 4.56 dB
(b) Twotap predictor:
Processing gain = 2.8715 = 4.49 dB
Therefore, the use of a threetap predictor in the DPCM system results an improvement of 4.56  4.49 = 0.07 dB over the corresponding system using a twotap predictor.
Problem 3.36
(a) For DPCM, we have 100oglO(SNR)o = ex + 6n dB
For PCM, we have 100oglO(SNR)o = 4.77 + 6n  20loglO(log(1 + ~))
where n is the number of quantization levels SNRofDPCM
183
SNR = a + 6n, where 3 < a < 15
For n=8, the SNR is in the range of 45 to 63 dBs.
SNR ofPCM
SNR = 4.77 + 6n  20l0g1000g(2.56)) = 4.77 + 48  14.8783
= 38 dB
Therefore, the SNR improvement resulting from the use of DPCM is in the range of 7 to 25 dB.
(b) Let us assume that nl bits/sample are used for DPCM and n bits/sample for PCM
If a = 15 dB, then we have
15 + 6nl = 6n  10.0
Rearranging: (n  n 1) =
10 + 15 6
= 4.18
which, in effect, represents a saving of about 4 bits/sample due to the use of DPCM.
If, on the other hand, we choose a = 3 dB, we have
3 + 6nl = 6n  10
Rearranging: (n  n 1) =
103 6
7
=
6
= 1.01
which represents a saving of about 1 bit/sample due to the use of DPCM.
184
Problem 3.37
The transmitting prediction filter operates on exact samples of the signal, whereas the receiving prediction filter operates on quantized samples.
Problem 3.38
Matlab codes
% Problem 3.38, CS: Haykin %flattopped PAM signal %and magnitude spectrum
% Mathini Sellathurai
%data
fs=8000; % sample frequency ts=i.25e4; %i/fs pUlse_duration=5e5; %pulse duration
% sinusoidal sgnal;
td=i.25e5; %sampling frequency of signal fd=80000;
t=(O:td:i00*td);
fm=i0000;
s=sin(fm*t) ;
% PAM signal generation pam_s=PAM(s,td,ts,pulse_duration); figure(i);hold on
185
plot (t , s , '' ) ; plot(t(l:length(pam_s)),pam_s); xl abe Lf t t ime" ) ylabel('magnitude') legend('signal','PAMsignal');
% Computing magnitude spectrum S(f) of the signal a=((abs(fft(pam_s)).2));
a=a/max(a); f=fs*(fs/fd:fs*(fs/fd):(length(a))*fs*(fs/fd); figure(2)
plot(f,a);
xlabel('frequency');
ylabel('magnitude')
% finding the zeros index=find(a<le5);
% finding the first zero
fprintf('Envelopes goes through zero for the first time %6d\n', min(index)*fs*(fs/fd))
186
function pam_s=PAM(s,td,ts,pulse_duration)
% Problem 3.38, CS: Haykin %flattopped PAM signal
%used in Problem 3.38, CS: Haykin % Mathini Sellathurai
potd=pulse_duration/td; tsotd=ts/td;
y=zeros(1,length(s»; tt=1:(tsotd):length(s);
for kk=1:length(tt); y(tt(kk):tt(kk)+potd1)=s(tt(kk».*ones(1,potd); end
pam_s=y(1:length(s)potd);
187
Answers: 3.38
' " ' ,
\ \
\
0.8 \
I \ I
I \ I
0.6
I
I \
0.4 I \
I
0.2 I
<1> I I
~ I I
0
g> I
E \ I
0.2 \
\
\
0.4 I
I
0.6
\ \
\ \
0.8 \
\
1
0 0.2 0.4 0.6 0.8
time
Figure 1: Flattopped PAM signal signal PAMsignal
I I
I I
188
1.2
1.4 X 103
0.9
U 1\", 1\ A A 1\ /\ 1\ ~}\~~ , 0.8
0.7
0.6
{!l
~ 0.5 g>
E
0.4
0.3
0.2
0.1
o o
6
7
2
3
4 frequency
5
Figure 2: Magnitude spectrum of flattopped PAM signal
0.9
l
\J L 0.8
0.7
0.6
Q)
io.5 l¥
0.4
0.3
0.2
0.1
0.5
2
2.5
1.5 frequency
Figure .'.3: Zoomed magnitude spectrum of flattopped PAM signal
190
Problem 3.39
Matlab codes
%problem 3.39, CS: Haykin
%muelaw pCM and uniform quantizing %Mathini Sellathurai
clear all
%sinusoidal signal t=[0:2*pi/100:2*pi]; a=s in Ct ) ;
% input signal to noise ratio in db SNRdb=[20 15 10 5 0 5 10 15 20 25 ];
for nEN=1: 10 sqnrfm=O; sqnrfu=O;
for k=1:100
snr = 10(SNRdb(nEN)/10);
wn= randn(1,length(a»/sqrt(snr); % noise a1=a+wn; %signal plus noise
[a_quanu,codeu,sqnr_u]=u_pcm(a1,256); %call uPCM [a_quanm,codem,sqnr_m]=mue_pcm(a1,256,255); %call muePCM
sqnrfm=sqnrfm+sqnr_m; sqnrfu=sqnrfu+sqnr_u; end
SNROm(nEN)=sqnrfm/k; %binSNRMUEPCM SNROu(nEN)=sqnrfu/k; %binSNRUPCM end
'i.plots
figure;hold on; plot(SNRdb,SNROu,'+') plot(SNRdb,SNROm,'o')
xlabel('input signaltonoiseration in db') ylabel('output signaltonoiseration in db') legend('uniform PCM, 256 levels' ,'muelaw PCM, mue=255')
191
function [a_q,snr]=u_pcm(a,n)
% function to generate uniform PCM for sinwave %used in problem 3.39, CS: Haykin
%Mathini Sellathurai
n=length(a); amax=max(abs(a)); a_q=a;
b_q=a_q;
d=2/n; q=d.*[O:n1]; q=q((n1)/2)*d; for i=1:n
a_q(find((q(i)d/2<= a_q) & (a_q <=q(i)+d/2)))= ... q(i).*ones(1,length(find((q(i)d/2 <=a_q) & (a_q<=q(i)+d/2)))); b_q(find(a_q==q(i)))=(i1).*ones(1,length(find(a_q==q(i)))); end
a_q =a_q*amax;
snr=20*log10(norm(a)/norm(aa_q));
192
function [a_q,snr]=mue_pcm(s,n,mue)
% function to generate muelaw PCM for sinwave %used in problem 3.39, CS: Haykin
%Mathini Sellathurai
a=max(abs(s»;
% muelaw y=(log(1+mue*abs(s/a»./log(1+mue».*sign(s); [y_q,code,sqn]=u_pcm(y,n);
%inverse muelaw a_q=«(1+mue).~(abs(y_q»1)./mue).*sign(y_q); a_q=a_q*a;
%SNR snr=20*log10(norm(s)/norm(sa_quan»;
193
Answer to Problem 3.39
50
+
E>
uniform PCM, 256 levels mueIaw PCM, mue=255
40
3
..c
"0
oS
C
.s 20
~
I
'"
.'"
0
c: 10
I
.if!
~
·in 0
"5
%
0
10
20 10
5
o
5
10
15
20
25
input signaltonoiseration in db
Figure 1 . input signaltonoise ratio Vs. output signaltonoise ratio for Jllaw PAM and uniform PCM
194
Problem 3.40
Matlab codes
% Problem 3.40, CS: Haykin %Normalized LMS prediction %of AR process/ speech signal % Mathini Sellathurai
clear all
mue=0.05; % step size parameter, a value between 0 ans 2 p=2; % filter order
N=10; % size of data
M=l;% number of realizations
% initializing counters errl=zeros(l,Np); xhatl=zeros(l,Np); x=zeros (1 ,N) ;
for m=l:M % 100 realizations
x(1:2)= [0.1 0.2J;
%AR process
for k=3:N x(k)=(0.8*x(kl)0.1*x(k2))+0.1*rand(1); end
% LMS prediction
[err, xhatJ=LMS(x,mue,p); errl=errl+err.~2; xhatl=xhatl+xhat;
end
plot(errl/m,'');
195
function [err, xhat]=LMS(xx,mue,p) % function Normalized LMS %porder of the filter
%muestep size parameter
%used in problem 3.40, CS: Haykin %Mathini Sellathurai
% length of the data N=length(xx);
% initializing weights and erros w=zeros(p,Np);
err=ones(l,Np); xhat=zeros(l,Np);
%prediction 1=1;
for k=l :Np
h=xx(k:p+k1); err(l)=(xx(k+p)h*w(:,l)); xhat(l)=h*w(:,l); xxx=xx(1+p1)+xx(1+p2);
we: ,l+l)=w(: ,l)+(mue/xxx)*h'*err(l); 1=1+1;
end
196
Answer to Problem 3.40
AR process (a , = O.BO, a2=O.10)+random noise
o.3rr~..__..__,.__,,
0.25
0.2 ~~I l~l~r lll~j'
'" .~~
."
.E
a.
E
«
0.15
0.1 0.050~1~0~0~~2~0~0~~3~0~0~4~0~0~5~0~0~60~0~7~0~0~8~0~0~~9~0~0~~1~000
Sample number
Figure'
NoisyARprocess, ao = 0.80, al
0.10
11 = 0.00751:':
11=0.05 ..
11=0.5 .
Number of iterations
Figure 2 : Learning curves for {l
0.0075, 0.05, 0.5
197
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.