You are on page 1of 33

CHAPTER 5

Problem 5.1

(a) Unipolar NRZ code.

The pair of signals sl (t) and s2(t) used to represent binary symbols 1 and 0, respectively are defined by

where Eb is the transmitted signal energy per bit and Tb is the bit duration. From the definitions of sl (t) and s2(t), it is clear that, in the case of unipolar NRZ signals, there is only one basis function of unit energy. The basis function is given by

<PI (t) = g.,

Then, we may expand the transmitted signals sl (t) and s2(t) in terms of <PI (t) as follows:

Hence, the signal-space diagram for unipolar NRZ code is (+ ~,O), as shown

(b) Polar NRZ code.

In this code, binary symbols 1 and ° are defined by

266

The basis function is given by

<1>1 (t) = /*'

Then, the transmitted signals in terms of <1>1 (t) are as follows:

Hence, the signal-space diagram for the polar NRZ code is (+~, -~) as shown below:

--------~--------O~------~~----~~<I>1

-VEb +-VEb

(c) Unipolar retum-to-zero code.

In this third code, binary symbols 1 and 0 are defined by

Sl (t) =

=0

S2(t) = 0

267

The energy of signal sl (t) is

The energy of signal s2(t) is zero.

The basis function is given by

s 1 (t) <1>1 (t) =

~

The signal-space diagram for the RZ code is as follows:



(d) Manchester code

Binary symbols 1 and 0 are defined by

~ o < i-: T bl2
,
s 1 (t) b
=
-~ Tb12:S; t:S; r;
T'
b -~ O:S; t:S; Tbl2
T'
s2(t) b
=
+ ~ T bl2 :s; t :s; T b
,
b 268

The energy of signal sl (t) is

Similarly, the energy of symbol s2(t) is

The basis function is given by

The signal-space diagram of the Manchester code is thus as follows:

--------~----------r-------~~------~~1

---.JEb 0 +--.JEb

Thus all the four line codes in this problem are one-dimensional.

Problem 5.2

The given 8-level PAM signal is defined by

The energy of signal si(t) is given by

T

e, = f (A)2dt o

269

2

= Ai T, Ai = ±l, ±3, ±5, ±7

The basis function is given by

<1>1 (t) =

The signal-space diagram of the 8-level PAM signal is as follows:

3ff 5ff

Problem 5.3

Consider the signals sl (t), s2(t), s3(t), and s4(t) shown in Fig. la. We wish to use the GramSchmidt orthogonalization procedure to find an orthonormal basis for this set of signals.

Step 1 We note that the energy of signal sl (t) is

IT/3 2

= 0 (1) dt

T

=

3

The first basis function <1>1 (t) is therefore

= {J3/T'

0,

0::; t ::; T /3 otherwise

}

270

s, (t) S2 (t) S3 (t) S4 (t)

'b_, 'b_, 1[l, tn,

o T 0 2T 0 T TOT

"333

(a)

4>, (t) 4>2 (t) 4>3 (t)

~L, ~b_, .r.

o TOT 2T 0 2T T

"3 "33 3

tb)

Figure 1

Step 2 Evaluating the projection of s2(t) onto <1>1 (t), we find that

The energy of signal s2(t) is

IT 2 E2 = 0 s2(t)

2T/3 2

= fo (1) dt

2T

=

3

271

The second basis function <1>2(t) is therefore

= {J3/T' T/3 ~ 2T /3

° otherwise

Step 3 Evaluating the projection of s3(t) onto <1>1 (t),

= °

and the coefficient s32 equals

The corresponding value of the intermediate function gi(t), with i = 3, is therefore

{I, 2T/3~t~T - 0, elsewhere

Hence, the third basis function <1>3(t) is

272

= {J3/T' 2T/3":;'t":;'T

0, elsewhere

The orthogonalization process is now complete.

The three basis functions <I> 1 (t), <l>2(t) , and <l>3(t) form an orthonormal set, as shown in Fig. lb. In this example, we thus have M = 4 and N = 3, which means that the four signals sl (t), s2(t), s3(t), and s4(t) described in Fig. la do not form a linearly independent set. This is readily confirmed by noting that sit) = sl (t) + s3(t). Moreover, we note that any of these four signals can be expressed as a linear combination of the three basis functions, which is the essence of the Gram-Schmidt orthogonalization procedure.

273

Problem 5.4

(a) We first observe that s1(t), s2(t) and s/t) are linearly independent.
The energy of s1(t) is
1
E1 = f (2)2dt = 4
0
The first basis function is therefore 4'1(t) =

o <t < 1

otherwsie

Define

T

= f s2(t) <P1(t)dt o

1

= f (-4)(1)dt = -4 o

g2( t) = s2( t) - s21 4'1 (t)
r' < t < 2
=
0, otherwise Hence, the second basis function is

274

)

= { -1, 1 < t < 2
0, otherwise
Define
T
s31 = f s3(t) 4>1 (t)dt
0
1
= f (3)(1)dt = 3
0
2'1"
s32 = f s3(t) 4>2(t)dt
T 2

= f (3)(-1)dt = -3 1

= (3, lo,

otherwise

Hence, the thrid basis function is

// g~(t)dt

o

= {1.,

0,

otherwise

The three basis functions are as follows (graphically)

~(.t) I

o ~---,--.--~-

-1.0 1- - - - - i

275

(b) S,(t) = 2~,(t)

S2(t) = -4~,(t) + 4~2(t)

S3(t) = 3~,(t) - 3~2(t) + 3~3(t)

276

Problem 5.5

Signals sl (t) and s2(t) are orthogonal to each other. The energy of sl (t) is

TI2 T

E 1 = f 12 dt + f (-1) 2 dt = T

o TI2

The energy of s2(t) is

To represent the orthogonal signals sl (t) and s2(t), we need two basis functions. The first basis function is given by

<1>1 (t) =

=

The second basis function is given by

=

The signal-space diagram for sl (t) and s2(t) is as shown below:

----------O+---~-----------4~<I>l

{f

Problem 5.6

The common properties of PDM and PPM are as follows: In both cases a time parameter of the pulse is modulated and the pulses have a constant amplitude. In PDM, the samples of the message signals are used to vary the duration of the individual pulses, as illustrated in Fig. 1a for M = 4 on the next page. In PPM, the position of the pulse is varied in accordance with the message, while keeping the duration of the pulse constant, as illustrated in Fig. 1 b for M = 4.

From these two illustrative figures, it is perfectly clear that the set of PDM signals is nonorthogonal, whereas the PDM signals form an orthogonal set.

277

___ J. , .. , ,1 .1 J l______!

$of·) t ' · , .

I ... ,i, I

I"e~,:t~l

; --a : : I I .l.. '

.. J_~~ .. 1_4!j.

S Ct) Iii : ' :

., I"

=l=;.Hh-~

Oil i I I I' ;-,'

I , j , ,_IT

J__ _ 1-

i

.

I

______ 1 _

_j J

___ . __ J_. __ .

,t! j

278

Problem 5.7

(a) The biorthogonal signals are defined as the negatives of orthogonal signals. Consider for example the two orthogonal signals sl (t) and s2(t) defined as follows:

where <1>1 (t) and <1>2(t) are orthonormal basis functions. The biorthogonal signals are given by -sl (t) and -s2(t), which are respectively expressed in terms of the basis functions as --.)E<1>1 (t) and - -{E<1>2(t). Hence, the inclusion of these two biorthogonal signals leaves the dimensionality of the signal-space diagram unchanged. This result holds for the general case of M orthogonal signals.

(b) The signal-space diagram for the biorthogonal signals corresponding to those shown in Fig.

P5.5 is as shown in Fig. 1a. Incorporating this diagram with that of the solution to Problem 5.5, we get the 4-signal constellation shown in Fig. lb.

-ff

-ff

------_.----+---~-----<1>1

-------....,,0+------ <1>1

-ff

(a)

(b)

Figure 1

Problem 5.8

(a) A pair of signals si(t) and sk(t), belonging to an N-dimensional signal space, can be represented as linear combinations of N orthonormal basis functions. We thus write

N

si(t) = ~>ij<1>/t), j=l

o s r s T i = 1,2

(1)

where the coefficients of the expansion are defined by

279

T

-» = f Si(t)cp/t)dt, o

l = 1,2

(2)

j = 1,2

The real-valued basis functions CPi (t) and CP2(t) are orthonormal. Hence,

= 8ij = { 1, 0,

if i= j otherwise

(3)

The set of coefficients {s ij} ~=1 may be viewed as an N-dimensional vector defined by

i = 1,2, "" M

(4)

where M is the number of signals in the setl, with M ~ N. The inner product of the pair of signal siCt) and skCt) is given by

T

f si(t)sk(t)dt o

(5)

By substituting (1) in (5), we get the following result for the inner product:

N N T

= L LSijSklf cP /t)CPI(t)dt

j=l 1=1 0

(6)

Since the cpit) form an orthonormal set, then, in accordance with the two conditions of Eq. (3) and (4), the inner product of s/t) and sit) reduces to

T N

f si(t)sk(t)dt = LSijSkj

o j=l

280

(b) Consider next the squared Euclidean distance between si and sk' which can be expressed as follows:

T T T

= f s7(t)dt + f s~(t)dt - 2 f si(t)skdt

o

o

o

T

= f (si( t)-s k( t»2 dt o

Problem 5.9

Consider the pair of complex-valued signals sl (t) and s2(t), which are defined by

(1)

(2)

The basis functions <1>1 (t) and <l>2(t) are real-valued and the coefficients all, alb a2l and a22 are complex-valued. We may denote the complex-valued coefficients as follows:

all = all + iBll al2 = al2 + iBl2 a21 = a2l + iB21 a22 = a22 + iB22

On this basis, we may represent the signals sl (t) and s2(t) by the following respective pair of vectors:

281

all a21
~l = Bll g2 = B21
al2 a22
Bl2 B22 The angle subtended between the vectors gl and g2 is

=

H

S I s2

Ilsdl'lls211

[a21] .. h

and S2 = are complex vectors. Recognizing t at

a22

we may go on to write

f S I (t) s; ( t) dt

_00 < 1

00 112 00 1/2 -

( f lsI (t)12 dt) ( f 182(t)12 dt)

_00 -00

The complex form of the Schwarz unequality becomes

2

f sl (t)s;(t)dt ::; f lSI (t)12 dt f IS2(t)12 dt

282

The equality holds when sl (t) and s2(t) are co-linear, that is, sl (t) = ks2(t) where k is any realvalued constant.

Problem 5.10

E[XjW'(tk)] = E[(sij + Wj)W'(tk)] E[sij W'(tk)] = Sij E[W'(tk)] = 0 We also note that

N

W'(tk) = W(tk) - L Wi $i(tk) i=1

We therefore have

E[X.W' (t )] = E[W. W'(tk)]

. J k J

N

= E [ WJ. W( t k) - L '" • (t ) E [W W ]

't' k ..

i=1 ~ J 1

But

T T

E[W. W(tk)] = E[W(tk) J Wet) $.(t)dt] = J $ .(t) E[W(tk)W(t)]dt

J 0 J 0 J

o h'j

Hence, we get the final result

= o.

283

Problem 5.11

For the noiseless case, the received signal ret) = set), 0 ~ t ~ T.

(a) The correlator output is

T

yeT) = f r("C)s("C)dt o

T

yeT) = f i("C)d"C

o

T

2(81t"C)

= f sin T dt

o

= TI2

(b) The matched filter is defined by the impulse response

h(t) = seT - t)

The matched filter output is therefore

yet) = f r('A)h(t - 'A)dA

= f s('A)s(T - t + 'A)dA

284

00 00

= ! f [81t(T - t)] .n _! f [81t(T - t + A)]dA

2 cos T UI\, 2 cos T

Since -T < A s 0, we have

1 (81t(t - T») ]A = 0

yet) = -2cOS A

T A=-T

_ ! . .I.sin(81t(T-t+2A»)]A=O

2 81t T A =-7

= !. cos(81t(t - T») _ _I_ sin(81t(t - T») _ _I_ sin(81t(t + T»)

2 T 161t T 161t T

(c) When the matched filter output is sampled at t = T, we get

yeT) = T 12

which is exactly the same as the correlator output determined in part (a).

Problem 5.12

(a) The matched filter for signal sl (t) is defined by the impulse response

The matched filter for signal s2(t) is defined by the impulse response

h2(t) = s2(T - t)

The matched filter receiver is as follows

I
Matched filter .. ~XI
hI (t) 1
1
x(t) 1
1
Matched filter 1"'- x2
hI (t) ..
I
Sample at
t = T
285 The receiver decides in favor of s2(t) if, for the noisy received signal,

X(t) = sk(t) + wet),

O'5:t'5:T k = 1,2

we find that xl > x2. On the other hand, if x2 > xl, it decides in favor of s2(t). If xl = x2, the decision is made by tossing a fair coin.

(b) Energy of signal sl (t) is given by

T 2T 3T

S'l = f ( 1) 2 dt+ f (-1) 2 dt + f (1) 2 di

o

T

2T

= 3T = E

Energy of signal s2(t) is

T12 3T12 5T12 3T

E 2 = f (-1) 2 dt + f ( 1 ) 2 dt + f ( -1 ) 2 dt + f ( 1 ) 2 dt

o

TI2

3T12

5T12

= 3T = E

The orthomomal basis functions for the signal-space diagram of these two orthogonal signals are given by

<1> I (t) S 1 (t)
=
J3T
and
<1>2 (t) s2(t)
= --
J3T The signal-space diagram of signals sl and s: is as follows:

286

----+-- -----1~ <1>1

o ,]3T

,]3T

The distance between the two signal points sl (t) and s2(t) is

The average probability of error is therefore

= ~erfc(~ !JJ

For ElNo, we therefore have

= ~erfc(J2)

-2 = 4 x 10

Problem 5.13

Energy of binary symbol 1 represented by signal sl (t) is

TI2 T

E 1 = f (+ 1 ) 2 dt + f (-1) 2 dt = T

o

TI2

Energy of binary symbol 0 represented by signal s2(t) is the same as shown by

287

TI2 T

E 2 = f (-1) 2 dt + f ( + 1) 2 dt = T

o TI2

The only basis function of the signal-space diagram is

SI (t) SI (t)

<Pl(t) = ~ = JT

The signal-space diagram of the Manchester code using the doublet pulse is as follows:

--ff

Hence, the distance between the two signal points is d = 2JT. The average probability of error over an AWGN channel is given by

P = !erfc(_d_) = ~erfc( fl-)

e 2 2Fo ~NoJ

Problem 5.14

(a) Let Z denote the total observation space, which is divided into two parts Zo and Z,. Whenever an observation falls in Zo' we say HO' and whenever an observation falls in Z" we say H,. Thus, expressing the risk R in terms of the conditional probability density functions and the decision regions, we may write

288

(1)

For an N-dimensional observation space, the integrals in Eq. (,) are N-fold integrals.

To find the Bayes test, we must choose the decision regions Zo and Z, in such a manner that the risk R will be minimized. Because we require that a decision be made, this means that we must assign each point ~ in the observation space Z to Zo or Z,; thus

Z = Zo + Z,

Hence, we may rewrite Eq. (,) as

(2)

We observe that

Hence, Eq. (2) reduces to

R = POC,o + P,C"

+ fz {-[po(C,o-Coo)fX1H (~IHO)] + 0-0

The first two terms in Eq. (3) represent the fixed cost. The integral represents the cost controlled by those points ~ that we assign to ZOo Since C'O > COO and CO, > C", we find

·that the two terms inside the square brackets are positive. Therefore, all values of x where the first term is larger than the second should be included in Zo because thej contA.:bute a negative amount to the integral. Similarly, all values of x where the second term is larger than the first should be excluded from Zo (i.e., assigned to Z,) because they would contribute a positive anount to the integral. Values of x where the two terms are equal have no effect on the cost and may be assigned arbitrarily. Thus the decision regions are defined by the following statement: If

289

aSSign ~ to z, and consequently say that H, is true. If the reverse is true, assign ~ to Zo and say HO is true.

Alternatively, we may write

po(C,o-Coo) P, (CO,-C,,)

The quantity on the left is the likelihood ratio:

Let

Thus, Bayes criterion yields a likelihood ratio test described by

(b) For the minimum probability of error criterion, the likelihood ratio test is described by

. Thus, we may view the mi.mmum probability of error criterion as a special case of the Bayes criterion with the cost values defined as

Coo = C'l = 0 Cl0 = COl

That is, the cost of a correct decision is zero, and the cost of an error of one kind is the same as the cost of an error of the other kind.

290

Problem 5.15

From the signal-space diagrams derived in the solution to Problem 5.1, we immediately observe the following:

l. Unipolar NRZ and unipolar RZ codes are non-minimum energy signals. 2. Polar NRZ and Manchester codes are minimum energy signals.

Problem 5.16

The orthonormal matrix that transforms the signal constellation shown in Fig. 5.11(a) of the textbook into the one shown in Fig. 5.11(b) is

1 1
Q= J2 J2
1 1
-
J2 J2 To prove this statement, we note that the constellations of Fig. 5.11(a) is defined by the four points {( a, a), (-a, a), (-a, -a), (a, -a)}. The new constellation is defined by

8i rotate = Q8i, which for i = 1 yields

,

81, rotate =

1

J2

1

J2

1

J2

1

J2

for a = l.

(:) = (~)

= (~l

Similarl y, 82, rotate '" 2)

= (-J202)

83, rotate

Hence, the transformation from Fig. 5.11(a) to Fig. 5.11(b) is given by Q, except for a scaling factor.

291

Problem S.17

(a) The minimum distance between any two adjacent signal points in the constellation of Fig.

PS .17 a of the textbook is

d(a) = 2a nun

The minimum distance between any two adjacent signal points in the constellation of Fig. PS .17b of the textbook is

which is the same as d~~. Hence, the average probability of symbol error usmg the constellation of Fig. PS .17 a is the same as that of Fig. PS .17b.

(b) The constellation of Fig. PS.17a has minimum energy, whereas that of Fig. PS.17b is of nonminimum energy. Applying the minimum energy translate to the constellation of Fig. PS.17b,

which involves translating it bodily to the left along the <Praxis by the amount ,/2 a , we get the corresponding minimum energy configuration:

<P2

----~--~O+---~----<Pl

-fla fla

- -Y2a

Problem S.18

Consider a set of three orthogonal signals denoted by {s j( t) } ~=O' each with energy Es. The average of these three signals is

aCt)

Appl ying the minimum energy translate to the signal set {s / t)} ~=O' we get a new signal set defined by

292

si(t) = si(t) - aCt),

i = 0,1,2

(1)

The signal energy of the new set is

00 2

E = f (si(t» dt

_00

= [ s7(t)dt - 2 foo si(t)a(t)dt + [ a(t)dt

_00 _00 _00

The correlation coefficient Pij between the signals si(t) and sj(t) is given by

E[si(t)(sj(t»]

Pij = E

(2)

Since si(t) and sjCt) are orthogonal by choice, Eq. (2) reduces to

1 2

for i::f:. j

which is the maximum negative correlation that characterizes a simplex signal with M = 3. thus, 2

the signal set {si( t) } i=O defined in Eq. (1) is indeed a simplex signal.

293

2

To represent the signal set {si( t) } i=O In geometric terms, we use the Gram-Schmidt

orthogonalization procedure. Specifically, we first set

so(t) <Po(t) = - JE

(3)

or equivalently

so(t) = JE<Po(t)

The projection of s'l (t) unto <Po(t) is

SlO = r s'l (t)<Po(t)dt

_00

= ~r sl (t)so(t)dt '" E _00

The second basis function is therefore

sI(t) + (JEI2)(<Po(t»

=

JE-(EI4)

Accordingly, we may express sl (t) in terms of the basis functions <Po(t) and <PI (t) as

294

(4)

The remaining signal s2 (t) may be expressed in terms of <1>0(t) and <1>1 (t) as

(5)

Thus, using Eqs. (3) to (5), we may represent the simplex code by the following signal-space diagram:

<1>1
,.,
I <,
~3E <,
I
"2 <,
I <,
I -.JE/2 <,
<,
<, <1>0
-: -{E
~3E ./
./
"2 ./
./ Problem 5.19

(a) An upper bound on the complementary error function is given by

2 exp(-u ) erfc (u) < ------"-=-----'-

Jicu

Hence, we may bound the given Pe as follows:

(1)

For large positive u, we may further simplify the upper bound on the complementary error function as shown here:

295

2 exp(-u ) e rf c ( u) < ---"--'--------'-

In

Correspondingly, we may bound Pe as follows:

(2)

(b) For EblNo = 9, we get the following results:

(i) The exact calculation of Pe yields

= 1.0 x 10-5

(ii) Using the bound in (1), we have the approximate value:

P ::::: exp( -9) e 6Jn

-5 = 1.16 x 10

(iii) Using the looser bound of (2), we have

P ::::: exp( -9)

e 2Jn

-5 = 3.48 x 10

As expected, the first bound is more accurate than the second bound for calculating Pe'

Problem 5.20

According to Eq. (5.91) of the textbook, the probability of error is over-bounded as follows:

296

(1)

where dik is the distance between message points si and sk' With the M transmitted messages assumed equally likely, the average probability of symbol error is overbounded as follows:

(2)

The second line of Eq. (2) defines the union bound on the average probability of symbol error for any set of M equally likely signals in an AWGN channel. Equation (2) is particularly useful for the special case of a signal set that has a symmetric geometry, which is of common occurrence in practice. In such a case, the conditional error probability Pe(m) is the same for all i, and so we may simplify Eq. (2) as

1 M (dik J

s 2: L erfc !TV' for all

k=l 2",No

k""i

(3)

The complementary error function may be upper-bounded as follows:

( J (2 J

d'k 1 d'k

erfc 2Fo :::; ,Jicexp -2~o

Hence, we may rewrite Eq. (3) as

for all

(4)

297

Provided that the transmitted signal energy is high enough compared to the noise spectral density No, the exponential term with the smallest distance dik will dominate the summation in Eq. (4).

Accordingly, we may approximate the bound on P; as

[ ( 2 J]

Mmin dik

P; ~ --exp +min --

2Jn i, k 2No

i*k

(5)

where Mmin is the number of transmitted signals that attain the minimum Euclidean distance for each mi. Equation (5) describes a simplified form of the union bound for a symmetric signal set, which is easy to calculate.

298