2919 2838 Mobile Wireless Communications - Solutions Manual CH 2-6

Mobile Wireless
Communications
Solutions Manual
(Chapters 2-6)
Department of Electrical Engineering

Columbia University
 2005 All Rights Reserved

Chapter 2
2-1 Explain, in your own words, the distinction between average path loss, shadow fading, and
multipath fading. How are they related to one another?
Average path loss is the decrease of the far-field average power of the transmitted EM wave
over distance at a rate of 1/(distance)n, where n ≥ 2 , n = 2 being the free-space case, n = 4
if we assume a two-ray propagation model, with the received signal composed of a direct
line-of-sight component and an indirect component by perfect reflection from a flat ground
surface.
Shadow fading is the long-term variations in radio signal power about the average power due
to terrain obstructions such as hills or buildings. This type of fading is slowly varying, being
manifested over relatively long distances (many wavelengths), from tens to hundreds of
meters. A good approximation to the effect of shadow fading is to assume that the power
measured in decibels (dB) follows a Gaussian or Normal distribution centered about its
average value, with some standard deviation ranging, typically, from 6 – 10 dB. The power
probability distribution is thus commonly called a log-normal distribution.
Multipath fading is the small-scale variation of the received signal attributed to the
destructive/constructive phase interference of many received signal paths. The signal power
fluctuates substantially on the order of wavelengths. For macrocellular systems, the
amplitude of the received signal due to multipath fading is often modeled as varying
randomly according to a Rayleigh distribution. For microcellular systems, with the existence
of a direct line-of-sight signal, the small-scale variation is better approximated by a Ricean
distribution.
Putting these three phenomena together, the statistically-varying received signal power PR
may be modeled by the following equation:
x
PR = α 21010 g (d ) PT GT GR
where PT is the transmitted signal power, GT is the transmitter gain, GR is the receiver gain
and g(d) is the 1/(distance)n relation which models the effect of average path loss. The
average received power is given by PR = g (d ) PT GT GR . The terms 10 x /10 and α 2 represent
the shadow fading and multipath fading effects respectively. The shadow-fading random
variable x, measured in dB, is Gaussian-distributed with a zero mean. The multipath-fading
random variable α is either Rayleigh-distributed for macrocellular systems, or
Ricean-distributed for microcellular systems.
Copyright  2005 1
2-2 Using Table 2-1, plot PR ,db / PT ,db for Orlando as a function of distance d, in meters, with 0 <
d < 200m. Assume transmitter and receiver antenna gains are both 1.
From (2-5),
PR = g (d ) PT GT GR
Assume GT = GR = 1 ,
PR / PT = g (d )
10 log10 ( PR / PT ) = 10 log10 g (d )
d
PR , db − PT ,db = 10 log10 [d − n1 (1 + ) − n 2 ]
db
Plot of PR , db − PT ,db with respect to Distance for Orlando
0
10log10d-n1(1+d/db)-n2
-5
10log10d-n1
10log10(1+d/db)-n2
-10
-15
-20
Power (dB)
-25
-30
-35
-40
-45
-50
0 20 40 60 80 100 120 140 160 180 200
Distance d (m)
2-3 Determine the shadow-fading parameter σ for each of the four measured curves of Figs.
2-4 and 2-5, and compare. Hint: First calculate the average value of each curve and then the
root mean-squared value about these averages.
For each curve, we choose 30 data points to compute the shadow-fading parameter σ .
Case (i) 836 MHz; winter values:

-40 -42 -39 -35 -29 -22 -25 -27 -24 -30 -30 -32 -25 -20 -15
-20 -25 -20 -9 -4 0 -10 0 -5 -7 -11 -17 -11 -6 -12
Average received signal level = -19.7 dB; σ = 12 dB
Case (ii) 836 MHz; summer values:

-44 -45 -46 -42 -33 -28 -28 -36 -26 -35 -39 -40 -30 -30 -24
-25 -28 -28 -12 -14 -4 -10 -8 -3 -8 -20 -21 -17 -13 -20
Average received signal level = -25.2 dB; σ = 12.6 dB
Case (iii) 11.2 GHz; winter values:

-46 -45 -37 -30 -30 -29 -39 -28 -42 -42 -35 -30 -28 -26 -28
-34 -28 -15 -23 -2 -20 -6 -6 -4 -10 -16 -17 -15 -16 -20
Case (iv) 11.2 GHz; summer values:

-55 -57 -57 -54 -48 -42 -48 -50 -44 -57 -57 -55 -54 -44 -40
-46 -50 -45 -32 -16 -15 -15 -15 -17 -21 -30 -34 -24 -29 -38
2-4 The average power received at mobiles 100 m from a base station is 1 mW. Log-normal,
shadow, fading is experienced at that distance.
a. What is the probability that the received power at a mobile at that distance from the base
station will exceed 1 mW? Be less than 1 mW?
0.5
b. The log-normal standard deviation σ is 6 dB. An acceptable received signal is 10 mW or

higher. What is the probability that a mobile will have an acceptable signal? Repeat for σ
= 10 dB. Repeat both cases for an acceptable received signal of 6 mW. Note: Solutions
here require the integration of the Gaussian function. Most mathematical software
packages contain the means to do this. Most books on probability and statistics have
tables of the error function used for just this purpose. The error function is defined in
chapter 3 of this text. See (3-12).
Case (i) p0 ≥ 10 mW; σ = 6 dB
1 1 10 log10 10
P( p ≥ p0 ) = − erf ( ) = 0.0478
2 2 2 ×6
Case (ii) p0 ≥ 10 mW; σ = 10 dB

1 1 10 log10 10
P( p ≥ p0 ) = − erf ( ) = 0.1587
2 2 2 ×10
Case (iii) p0 ≥ 6 mW; σ = 6 dB

1 1 10 log10 6
P( p ≥ p0 ) = − erf ( ) = 0.0973
2 2 2 ×6
Case (iv) p0 ≥ 6 mW; σ = 10 dB

1 1 10 log10 6
P( p ≥ p0 ) = − erf ( ) = 0.2182
2 2 2 ×10
2-5 a. Fill in the details of the derivation of the two-ray average received power result given by
(2-13a).
Step 1: Derivation of ∆φ ≈ 4π ht hr / λ d
With reference to the first diagram on the

right, we can prove by Pythagoras’
ht d1
Theorem that
d1 + d 2 = r 2 + (ht + hr ) 2 (1) d2 hr
With reference to the second diagram on hr

d2
the right, we have r
r 2 = d 2 − (ht − hr ) 2 (2)
Substitute (2) into (1), we have

d
d1 + d 2 = d 2 − (ht − hr ) 2 + (ht + hr ) 2 ht - hr
d1 + d 2 = d 2 + 4ht hr (3)
hr hr
r
Assume d 2 >> 4ht hr ,
∆d = (d1 + d 2 ) − d = d 2 + 4ht hr − d
≈ d 2 + 4ht hr + (2ht hr / d ) 2 − d
= (d + 2ht hr / d ) − d = 2ht hr / d (4)
Substitute (4) into ∆φ ≡ 2π∆d / λ , we have ∆φ ≡ 4π ht hr / λ d .
2
d
Step 2: Derivation of 1 − e − j∆φ ≈ (∆φ )2
d1 + d 2
Assume ∆φ is small and d /(d1 + d 2 ) ≈ 1

2
d 2
1− e − j∆φ ≈ (1 − cos ∆φ ) 2 + j sin ∆φ ≈ j sin ∆φ = (sin ∆φ ) 2 ≈ (∆φ ) 2
2
(5)
d1 + d 2
Finally, we can obtain (2-13a) by substituting the results in Steps 1 and 2:

2
 λ   λ∆φ 
2 2
d − j ∆φ (ht hr )2
PR = PT GT GR   1 − e ≈ P G G
T T R  ≈ P G G
 4π d   4π d 
T T R
d1 + d 2 d4
b. Superimpose a 1/d4 curve on the measured curves of Figs. 2-7 and 2-8, and compare with
the measured curves.
SIGNAL POWER RELATIVE TO 0dBm FREE SPACE AT 1km
+20
+10
-10
FREE SPACE
-20 TRANSMISSION
-30
-40
-50
-60 A
-70
A: f = 836 MHz, ht = 150 m PHILADELPHIA B 1/d4
-80 B: f = 900 MHz, ht = 137 m NEW YORK
C: f = 922 MHz, ht = 140 m TOKYO C
-90
-100
1 10 100
DISTANCE FROM BASE STATION ANTENNA (km)
Fig. 2-7 Received signal power as function of distance
SIGNAL POWER RELATIVE TO 0dBm FREE SPACE AT 1km
0
-10
-20 1/d4 FREE SPACE
-30
-40
-50
-60 hb = 820 m
hb = 45 m
-70
-80
f = 922 MHz
-90 hb = 220 m
hm = 3 m
-100
hb = 140 m
-110
-120
1 2 4 6 10 20 40 60 100
DISTANCE (km)
Fig. 2-8 Effect of base station antenna height on received power
In Fig. 2-7, we can see that for all the measured curves, the rates of decrease of the
received signal power only come close to that of the 1/d4 curve when the distance from
the base station antenna is large enough. This is due to the fact that in the derivation of
(2-13a), ∆φ ≡ 4π ht hr / λ d is assumed to be small. For instance, take ∆φ ≤ 0.6 radian ,
signal frequency f c = 836MHz ( λ = 0.36m ), ht = 150m and hr = 1m , we have
d ≥ 4π ×150 /(0.36 × 0.6) = 8.7km . In Fig. 2-8, all the measured curves have slopes
similar to that of the 1/d4 curve, and raising the base station antenna tends to increase the
strength of the received signal.
c. Do the results of Fig. 2-8 validate (2-13a)? Explain.
Yes. First, all the measured curves in Fig. 2-8 have the received signal powers decrease
with distance at a rate similar to that of the 1/d4 curve. Second, as the height of the base
station antenna hb increases, the received signal power increases. Both observations agree
with (2-13a).
2-6 a. Verify, as indicated in the text, that, for the Rayleigh-distributed random variable α in
(2-15), σr2 must equal 1/2.
x
From (2-4), the instantaneous received signal power PR = α 10 g (d ) PT GT GR . Taking 2 10
expectation on both sides, we have
x
PR = E[ PR ] = E[α ]E[10 ]g (d ) PT GT GR
2 10
x
Since the average received power is just PT g (d )GT GR and E[1010 ] = 1 , this implies
E[α 2 ] = 1 . From (2-15),
∞ α 3 −α 2 / 2σ r2
E[α 2 ] = ∫ e dα
0 σ r2
Substitute y = α 2 , dy = 2α dα , we have
∞ y ∞
E[α 2 ] = ∫ e − y / 2σ r dy = − ∫ yd (e − y / 2σ r ) = − [ ye− y / 2σ r + 2σ r2 e− y / 2σ r ]∞0 = 2σ r2
2 2 2 2
0 2σ 2 0
r
Therefore, E[α 2 ] = 2σ r2 = 1 or σ r2 = 1/ 2
b. Derive (2-17a) from (2-17) and show that x and y are zero-mean random variables, each
with variance σR2 as defined.
The actual received normalized signal S R (t ) is given by (2-17):

L
S R (t ) = ∑ ak cos[ωc (t − t0 ) + φk ]
k =1
Expanding (2-17) by trigonometry, we get

L
S R (t ) = ∑ ak [cos φk cos ωc (t − t0 ) − sin φk sin ωc (t − t0 )]
k =1
L L
= ∑ ak cos φk cos ωc (t − t0 ) − ∑ ak sin φk sin ωc (t − t0 )
k =1 k =1
= x cos ωc (t − t0 ) − y sin ωc (t − t0 )
L L
where x ≡ ∑ ak cos φk and y ≡ ∑ ak sin φk .
k =1 k =1
Consider the means of x and y.
L L
E[ x] = E[∑ ak cos φk ] = L ⋅ E[ak ]E[cos φk ] E[ y ] = E[∑ ak sin φk ] = L ⋅ E[ak ]E[sin φk ]
k =1 k =1
Since φk is uniformly distributed between 0 and 2π, E[cos φk ] = E[sin φk ] = 0 . Hence,

E[ x ] = E[ y ] = 0
Now, consider the variances of x and y.

L
σ x2 = E[( x − x )2 ] = E[(∑ ak cos φk − 0)2 ]
k =1
L
= E[∑ ak2 cos 2 φk ] + 2 E[∑ ai a j cos φi cos φ j ]
k =1 i≠ j
1 + cos 2φk
L
1 L
= E[∑ a ( )] + 0 = ∑ E (ak2 ) = σ R2
2
k
k =1 2 2 k =1
L
σ = E[( y − y ) ] = E[(∑ ak sin φk − 0) 2 ]
2
y
2
k =1
L
= E[∑ ak2 sin 2 φk ] + 2 E[∑ ai a j sin φi sin φ j ]
k =1 i≠ j
1 − cos 2φk
L
1 L
= E[∑ ak2 ( )] + 0 = ∑ E (ak2 ) = σ R2
k =1 2 2 k =1
Therefore, both x and y have the same variance σ R2 .
c. Starting with the Rayleigh distribution (2-21) for the received signal envelope a, show the
instantaneous received power PR obeys the exponential distribution of (2-23).
From (2-22), PR = ca 2 / 2 ⇒ a 2 = 2 PR / c
da 1
Differentiate (2-22) with respect to PR on both sides, we have =
dPR ca
From simple probability theory,
da
f PR ( PR ) = f a (a)
dPR
Substitute (2-21) and a 2 = 2 PR / c , we have
1 1
f PR ( PR ) = 2 e − a / 2σ R = 2 e− PR / cσ R
2 2 2
cσ R cσ R
From (2-20), cσ R2 = p . Put it into the above equation and we can obtain the exponential
distribution of (2-23):
1
f PR ( PR ) = e− PR / p
p
2-7 a. Show, following the hints provided in the text, that the Ricean distribution (2-25)
approaches a Gaussian distribution centered about A for A2 / 2σ R2 1 .
a 2 + A2
a −( 2σ R2
)
aA
From (2-25), f a (a) = 2 e I0 ( )
σR σ R2
According to the hints, I 0 ( z ) → e z / 2π z when z 1 . This corresponds to the case

A2 / 2σ R2 1 . With this assumption, (2-25) becomes
a 2 + A2 aA a 2 − 2 aA+ A2 ( a − A )2
a −( 2σ R2
)
σ R2 σR 1 a −( 2σ R2
)
1 a − 2σ R2
f a (a) = 2 e ⋅e ⋅ = ⋅ e = ⋅ e
σR 2π aA σ R 2π A σR 2π A
Therefore, it is readily seen that f a (a ) peaks at about A, and, in the vicinity of that value
of the amplitude, is closely Gaussian.
b. Verify that the instantaneous received power distribution in the case of a direct ray is
given by (2-27). Show that, as the K-factor gets smaller (the direct line-of-sight ray
decreases relative to the scattered signal terms), the fading distribution of (2-27)
approaches a Rayleigh distribution.
From (2-22), PR = ca 2 / 2 ⇒ a 2 = 2 PR / c (i)
Take expectation on both sides of (2-22),

p = E[ PR ] = cE[a 2 ] / 2
Since
E[a 2 ] = E[( A + x) 2 + y 2 ] = A2 + 2σ R2 = 2σ R2 (1 + K )
we have
p = cσ R2 (1 + K ) or c = p / σ R2 (1 + K ) (ii)
Put (ii) into (i)
a = 2 PRσ R2 (1 + K ) / p (iii)
da 1
Differentiate (2-22) with respect to PR on both sides, we have = (iv)
dPR ca
From simple probability theory,
da
f PR ( PR ) = f a (a)
dPR
Substitute (iii), (iv) and (2-25) into the above expression,

1+ K A2
1 −( p
⋅ PR + 2 )
2σ R 4(1 + K ) A2
f PR ( PR ) = 2 e I0 ( ⋅ PR ⋅ 2 )
cσ R p 2σ R
Put K ≡ A2 / 2σ R2 and cσ R2 = p /(1 + K ) , we have

1+ K
(1 + K )e− K − ⋅ PR 4 K (1 + K )
f PR ( PR ) = e p
I0 ( ⋅ PR )
p p
Put K=0, the expression for the instantaneous received power distribution becomes
P
1 − pR
f PR ( PR ) = e
p
Therefore, it is shown that as the K-factor gets smaller, f PR ( PR ) approaches an
exponential distribution.
2-8 As will be seen throughout the text, simulation is commonly used to determine the
performance, as well as verify analysis, of cellular systems. Most critical here is the
simulation of fading conditions. This problem provides an introduction to the simulation of
Rayleigh fading.
a. Consider a sequence of n random numbers xi, j =1 to n, uniformly distributed from 0 to 1.
(Pseudo-random number generators are often available in mathematical software
packages.) Let x = (b / n)∑ j =1 ( x j − 1/ 2) . Show x approximates a Gaussian random
n
variable of zero average value and variance σ2 = b2/12n. Repeat for another set of n
(independent) uniformly-distributed random numbers, calling the sum obtained in this
case y. Using x and y, generate a Rayleigh-distributed random variable. Comparing with
(2-21), what is the Rayleigh parameter σR2 in this case? Hint: Consider the derivation of
(2-21) starting with (2-19) and the discussion in the text following.
By the Central Limit Theorem of probability, for large n, the random variable x, defined
as the sum of n random variables, becomes approximately Gaussian-distributed. Now we
will derive the mean and variance of x.
E ( x) = (b / n)∑ j =1[ E ( x j ) − 1/ 2] = 0
n
Mean:
σ 2 = E[( x − 0)2 ] = (b / n)2 ∑ j =1 E[( x j − 1/ 2) 2 ]

n
Variance:
1/ 2
b2  y j 
3
b 2 1/ 2 b2
= ( ) ⋅ n ⋅ ∫ y j dy j = ⋅  
2
=
n −1/ 2 n  3  −1/ 2 12n
We can generate a Rayleigh-distributed random variable a from x and y by the formula

a = x 2 + y 2 and in this case the Rayleigh parameter σ R2 = b 2 /12n .
b. A different method of obtaining the Rayleigh distribution directly from a

uniformly-distributed random variable x is to write the expression
a = −2σ 2 log e x
Show the variable a is Rayleigh-distributed. How would you now use a sequence of
uniformly-distributed random numbers to generate a Rayleigh distribution?
Rewrite the expression of a, we get

a2
x = exp(− 2 )
2σ
As a increases from 0 to ∞ , x decreases from 1 to 0. Since x is bounded by 0 and 1, the
probability density function f x ( x) = 1 .
exp(0) a02
P(a < a0 ) = ∫ dx = 1 − exp(− 2 )
a02
exp( − 2 )
2σ 2σ
d a2 a a2
f a (a) = [1 − exp(− 2 )] = 2 exp(− 2 )
da 2σ σ 2σ
Therefore, a is Rayleigh-distributed. We can obtain Rayleigh-distributed random

numbers ai from a sequence of uniformly-distributed random numbers xi by directly
applying the formula ai = −2σ 2 log e xi .
c. Choosing various values for n, generate the Rayleigh distribution using the two
methods of a. and b., and compare, both with each other and with a plot of the Rayleigh
distribution. What is the effect on these results of varying n?
Choose σ R2 = 9 . Hence, in method a., the value of the parameter b is 108n . Each plot
is based on the statistics of k = 10, 000 data points generated by the approximation
methods. To obtain a plot of the relative frequency density, we first divide the x-axis into
equal intervals each of width w. Then we count the number of data points ci in each
interval. The relative frequency density of interval i is then given by ci /(k × w) . From the
plots we find that by increasing the value of n, method a. gives a better approximation to
the Rayleigh distribution.
Relative Frequency Density of the Rayleigh-Distributed Random Variables Generated by Two Different Methods
0.25
Method a; n = 5
Method a; n = 20
Method b
Rayleigh Distribution fa(a)
0.2
0.15
Density
0.1
0.05
0
0 2 4 6 8 10 12
a
d. Show how you would extend the method of a. to generate the Ricean distribution.
Generate and plot this distribution using the pseudo-random numbers generated in c. for
various values of the Ricean K-factor.
We can extend the method of a. to generate the Ricean distribution by the formula
a = ( A + x) 2 + y 2 , where A is a positive constant.
Relative Frequency Density of the Ricean-Distributed Random Variables Generated by Method a (n = 5)

0.16
K=1
K=2
0.14 K=3
Ricean Distribution fa(a); K = 1
0.12
0.1
Density
0.08
0.06
0.04
0.02
0
0 2 4 6 8 10 12 14 16 18
a
Relative Frequency Density of the Ricean-Distributed Random Variables Generated by Method a (n = 20)
0.16
K=1
K=2
0.14 K=3
Ricean Distribution fa(a); K = 1
0.12
0.1
Density
0.08
0.06
0.04
0.02
0
0 2 4 6 8 10 12 14 16 18
a
2-9 Consider the average fade duration equation (2-39). Take the case of a vehicle moving at a
speed of 100 km/hr. The system frequency of operation is 1 GHz. Say the ratio ρ = 1. Show
the average fade duration is 8 msec, as noted in the text. Now let the received signal
amplitude be 0.3 of the rms value. Show the average fade duration is now 1 msec.
v 100 ×103 /(60 × 60) e −1

Case (i) ρ = 1 , fm = = = 92.59 Hz τf = = 7.4m sec.
λ 3 ×108 /109 92.59 × 2π
e0.3 − 1
2
Case (ii) ρ = 0.3 , f m = 92.59 Hz τ f = = 1.35m sec.

0.3 × 92.59 × 2π
2-10 a. Summarize, in your own words, the discussion in the text on time dispersion and
frequency-selective fading.
In a wireless medium, due to reflection, scattering, diffraction and refraction, a signal

usually reaches a destination through multiple paths of various distances. Thus, a
receiver, instead of getting a copy of the signal, often obtains replicas of a signal which
arrive at different times. The superposition of the delayed replicas results in the
broadening of the signal. This is known as time dispersion, which corresponds to a
non-flat frequency response of the channel. For relatively large signal bandwidth, one
encounters frequency-selective fading, with different frequency components of the
signal being handled differently over the channel, leading to signal distortion. For the
case of digital signals, this distortion introduced by frequency-selective fading
manifests itself in intersymbol interference (ISI), with successive digital symbols
overlapping into adjacent symbol intervals.
b. Consider several cases: a delay spread of 0.5 µsec, one of 1 µsec, and a third one of 6
µsec. Determine whether individual multipath rays are resolvable for the two
transmission bandwidths, 1.25 MHz used in IS-95 and cdma2000, and 5 MHz used in
WCDMA. (See Chapter 10)
For IS-95 and cdma2000, multipath echoes appearing much greater than
1/(2π ×1.25) µ sec ≈ 0.13µ sec apart will be resolvable. Therefore, multipath rays
should be resolvable for delay spreads of 1µ sec and 6 µ sec , but probably not for the
delay spread of 0.5µ sec .
For WCDMA, multipath echoes appearing much greater than 1/(2π × 5) µ sec
≈ 0.032 µ sec apart will be resolvable, which is valid for all the three cases of delay
spread.
2-11 Indicate the condition for flat fading for each of the following data rates:
8 kbps, 40 kbps, 100 kbps, 6 Mbps.
Indicate which, if any, radio environments would result in flat fading for each of these data
rates.
Data Rate Symbol Flat Fading Radio Environment for Flat Fading
Interval Condition
8 kbps 125 µsec τav < 25 µsec Most environments
40 kbps 25 µsec τav < 5 µsec Some urban environments and
most suburban and rural environments
100 kbps 10 µsec τav < 2 µsec Most suburban and rural environments
6 Mbps 0.167 µsec τav < 0.033 µsec Some indoor picocellular environments
2-12 a. Consider the transversal filter equalizer of Fig. 2-20. A training sequence of K binary
digits is used to determine the 2N + 1 tap gains, as described in the text. Show that,
under a minimum mean-squared performance objective, the optimum choice of tap
gains is given by (2-53).
Under a minimum mean-squared performance objective, we want to choose the tap

K
gains hn, − N ≤ n ≤ N , such that ∑ (s
j =1
j − sˆ j ) 2 is minimum.
To obtain the optimum choice of tap gains, we find the solution of

∂ K K ∂sˆ j
∑ j j ∑
∂hl j =1
( s − s
ˆ ) 2
=
j =1
2( s j − s
ˆ j )
∂hl
=0
N ∂sˆ j
Since sˆ j = ∑hr
n =− N
n j −n and
∂hl
= rj −l , the above equation becomes
K
∑ 2(s
j =1
j − sˆ j )rj −l = 0
K K
⇒ ∑ s j rj −l = ∑ sˆ j rj −l
j =1 j =1
K K N
⇒ ∑s r
j =1
j j −l =∑ ∑hr
j =1 n =− N
r
n j − n j −l −N ≤ l ≤ N
b. Show the vector form of (2-53) is given by (2-56), with the solution given by (2-57).
K K
Define Rl , n ≡ ∑ rj − n rj −l and gl ≡ ∑ s j rj −l , − N ≤ l , n ≤ N . Equation (2-53) then takes
j =1 j =1
on the simpler looking form

N
gl = ∑hR
n =− N
n l ,n −N ≤ l ≤ N
In matrix notation,
 g − N   R− N , − N R− N ,− N +1 R− N , N   h− N 
   
R− N +1,− N +1 R− N +1, N   h− N +1 
 g − N +1  =  R− N +1,− N or g = Rh
    
    
 g N   RN ,− N RN ,− N +1 RN , N   hN 
Multiply R-1 to both sides, we have

R-1g = (R-1R)h or h = R-1g
2-13 a. Work out a simple example of the transversal filter equalizer: Say the equalizer has
three taps to be found using the minimum mean-squared performance objective.
Choose a set of K = 10 arbitrarily-chosen binary digits as the training sequence and
then let some of these digits be received in “error”, i.e. some are converted to the
opposite polarity. Find the “best” set of taps in this case. Try to choose the training
sequence so that there are equal numbers of +1 and –1 digits. Compare the tap
coefficients with those found using the approximation of (2-58).
Choose the training sequence {sj} = {1 1 -1 1 -1 -1 -1 1 1 -1}

And the received sequence {rj} = {1 1 1 1 -1 -1 -1 -1 1 -1}
with the third and eighth bits received in errors.
From equations (2-54), (2-55) and (2-57), we find that

 3 10 3 2   91 -24 -11  3   0.1716 
    −1 1     
g =  6  R =  3 10 3  and h = R g =  -24 96 -24  6  =  0.6471 
-1  2 3 10  816  -11 -24 91  -1  -0.3284 
        
By using the approximation of (2-58), the tap coefficients
 3   0.3 
1 1    
happrox = ⋅ g = ⋅  6  =  0.6 
10 10    
 -1  -0.1
The approximated coefficients obtained by (2-58) are very close to the exact
coefficients obtained by (2-57). The mean-squared difference given by the
approximated tap coefficients is just 0.0763 higher than the optimum value.
b. Repeat this example for a different set of transmitted digits and errors in reception.
{sj} = {1 -1 -1 1 -1 1 -1 -1 1 1}
{rj} = {1 -1 1 1 1 1 -1 -1 1 1} (third and fifth bits received in errors)
 -3   10 1 -2   99 -12 21   -3   -0.3718 
    1     
g= 6  R =  1 10 1  h=  -12 96 -12   6  =  0.6410 
1  -2 1 10  936     
     21 -12 99   1   -0.0385 
 -3   -0.3 
1    
happrox = ⋅  6  =  0.6 
10    
 1   0.1 
c. Choose a larger example of a transversal filter equalizer and repeat a. and b.,
comparing with the results obtained there.
In the following two examples, we use transversal filter equalizers of 11 taps (N = 5)

adapted with 30-bit training sequences (K = 30). Since the matrix manipulation is
cumbersome, a computer program was written to solve the optimum tap coefficients h.
Example 1
{sj} = {1 1 -1 1 -1 -1 -1 1 1 -1 1 1 -1 1 -1 -1 -1 1 1 -1 1 1 -1 1 -1 -1 -1
1 1 -1}
{rj} = {1 1 -1 1 1 1 1 1 1 -1 1 1 -1 1 -1 -1 -1 1 1 -1 1 1 -1 1 -1 -1 -1
1 1 -1} (fifth to seventh bits received in error)
 -1   30 -1 -2 7 0 1 -2 9 -6 -7 14   0.0080 
     -0.0472 
 -2   -1 30 -1 -2 7 0 1 -2 9 -6 -7   
 5   -2 -1 30 -1 -2 7 0 1 -2 9 -6   -0.0598 
     
 -8   7 -2 -1 30 -1 -2 7 0 1 -2 9   -0.1492 
 -7   0 7 -2 -1 30 -1 -2 7 0 1 -2   -0.2215 
     
g =  24  R =  1 0 7 -2 -1 30 -1 -2 7 0 1  h =  0.7994 
 -7   -2 1 0 7 -2 -1 30 -1 -2 7 0   -0.1814 
     
 -4   9 -2 1 0 7 -2 -1 30 -1 -2 7   -0.0115 
 5   -6 9 -2 1 0 7 -2 -1 30 -1 -2   -0.0245 
     
 -2   -7 -6 9 -2 1 0 7 -2 -1 30 -1   -0.0222 
     
 -3   14 -7 -6 9 -2 1 0 7 -2 -1 30   -0.1230 
happrox = (-0.0333 - 0.0667 0.1667 - 0.2667 - 0.2333 0.8000
- 0.2333 - 0.1333 0.1667 - 0.0667 - 0.1000)T
Example 2
{sj} = {1 -1 -1 1 -1 1 -1 -1 1 1 1 -1 -1 1 -1 1 -1 -1 1 1 1 -1 -1 1 -1 1 -1
-1 1 1}
{rj} = {1 -1 -1 1 -1 1 -1 -1 -1 -1 -1 -1 -1 1 -1 1 -1 -1 1 1 1 -1 -1 1 -1 1
-1 -1 1 1} (ninth to eleventh bits received in error)
 13   30 -3 2 1 -2 11 0 -3 0 -5 12   0.1146 
     0.0548 
 -4   -3 30 -3 2 1 -2 11 0 -3 0 -5   
 -1   2 -3 30 -3 2 1 -2 11 0 -3 0   0.0016 
     
 -4   1 2 -3 30 -3 2 1 -2 11 0 -3   -0.2065 
 -9   -2 1 2 -3 30 -3 2 1 -2 11 0   -0.2417 
     
g =  24  R =  11 -2 1 2 -3 30 -3 2 1 -2 11  h =  0.7024 
 -9   0 11 -2 1 2 -3 30 -3 2 1 -2   -0.2417 
     
 -4   -3 0 11 -2 1 2 -3 30 -3 2 1   -0.2065 
 -1   0 -3 0 11 -2 1 2 -3 30 -3 2   0.0016 
     
 -4   -5 0 -3 0 11 -2 1 2 -3 30 -3   0.0548 
     
 13   12 -5 0 -3 0 11 -2 1 2 -3 30   0.1146 
happrox = (0.4333 -0.1333 -0.0333 -0.1333 -0.3000 0.8000
-0.3000 -0.1333 -0.0333 -0.1333 0.4333)T
A comparison of the four examples of traversal filter equalizers is summarized in the

table below.
K 2N+1 Avg. distance between Optimum ms diff. Increase in ms
optimum and approx. hj’s 1 K diff. with
∑ ( s j − sˆ j )
2
1 N
K j =1 approx. h
∑ h( j ) − happrox ( j )
2N + 1 j =− N
(sj - training seq.)
1. 10 3 0.1346 0.5137 0.0763
2. 10 3 0.0838 0.4937 0.0243
3. 30 11 0.0772 0.2128 0.0917
4. 30 11 0.1313 0.1397 0.4456
In general, the approximated tap coefficients are very close to the exact optimum
coefficients obtained by (2-57). Larger traversal filter equalizers with more tap
coefficients and adapted by longer training sequences usually give better estimated
sequences, i.e. resulting in a smaller mean square difference between the transmitted
and the estimated signal sequences.
2-14 a. Plot the improvement in performance obtained with the use of dual selection diversity
as the ratio of local-mean power to the threshold varies. Use at least the following cases:
(1) the local-mean power 20 times the threshold; (2) local-mean power 10 times the
threshold; (3) local-mean power equal to the threshold; (4) local-mean power 0.1 of the
threshold. Note: Performance may be defined as outage rate or the probability that at
least one of the channels has an instantaneous power greater than the threshold.
Outage Rate vs. Ratio of Local-Mean Power to the Threshold for Dual Selection Diversity
1
0.9
0.8
0.7
0.6
Outage Rate
0.5
0.4
0.3
0.2
0.1
0
0 2 4 6 8 10 12 14 16 18 20
Ratio of Local-Mean Power to the Threshold
b. Repeat a. for four- and eight-fold diversity and compare all three orders of diversity.
Outage Rate vs. Ratio of Local-Mean Power to the Threshold for N-Fold Selection Diversity
1
0.9
N=2
N=4
0.8
N=8
0.7
0.6
Outage Rate
0.5
0.4
0.3
0.2
0.1
0
0 2 4 6 8 10 12 14 16 18 20
Ratio of Local-Mean Power to the Threshold
In general, the higher the order of diversity, the better the performance. However, as the
ratio of local-mean power to threshold increases, the improvement becomes less
significant. For a ratio larger than three, there is no much difference between four- and
eight-fold diversities. Besides, if the order of diversity is high enough, it will be hard to
improve the performance further by more diversity branches.
2-15 Show the optimum maximal-ratio combining gain for the kth diversity branch is given by
(2-64). Explain the statement that the SIR is then the sum of the SIRs, summed over the N
diversity branches.
N 2 N
SIR ≡ s / I 0 = ∑ g k ak / ∑ g k nk′
2 2 2
From (2-61)
k =1 k =1
According to Schwartz’s inequality for complex numbers,

N
 N 2 
N
2
∑
k =1
c *
k kd ≤  ∑ k   ∑ dk 
 k =1
c ⋅
  k =1 
Choose ck = ak* / nk′ and d k = g k nk′ , we have

N
 N 2 
N
2
∑ ak g k ≤  ∑ ak / nk  ⋅  ∑ g k
′ nk′ 
2 2
k =1  k =1   k =1 
N N N
SIR ≡ ∑a g /∑ g nk′ ≤ ∑ ak / nk′
2 2 2 2
⇒ k k k
k =1 k =1 k =1
SIR is maximized when equality is obtained with d k = Kck , i.e. g k nk′ = Kak* / nk′ or
g k = Kak* / nk′ , which is the expression in (2-64).
2
∑a / nk′ , and the SIR on the kth diversity branch is just

2 2
Since the SIR in this case is k
k =1
given by ak / nk′ , we can state that the resultant SIR is the sum of the SIRs over the N
2 2
diversity branches.
2-16 a. Explain how equal-gain combining differs from maximal-ratio combining. In particular,
write an expression for the SIR in the case of equal-gain combining. Hint: How would
this expression compare with (2-61)?
Equal-gain combining differs from maximal-ratio combining in that the gains gk over
all diversity branches are the same, rather than being adjusted according to the SIR on
the respective branch. Hence, in equal-gain combining the diversity channel outputs are
simply added together. The SIR in this case is
N 2 N
SIR = ∑ ak / ∑ nk′
2
k =1 k =1
b. Why would you expect the performance of diversity schemes to be ranked in the order
maximal-ratio combining best, equal-gain next best, selection diversity last?
We can rank the performance of the three diversity schemes by comparing their
respective SIR’s. First, the SIR of the maximal-ratio scheme must excel that of the
equal-gain scheme, based on the fact that the branch gains of the maximal-ratio
schemes are adapted to maximize the SIR. Next, we will show that the SIR of the
equal-gain scheme should in general be larger than that of the selection scheme.
Consider the SIR of the selection scheme. Suppose the signal on the ith branch has the
largest SIR, then
2 2
ak ai
SIRselection = max{ }=
nk′ ni′
k 2 2
Since the SIR on each branch ak / nk′

2 2
is usually greater than one, we can write the
following inequality:
2 N 2
2 ai + ∑ ak ∑a k
ai k ≠i k =1
SIRselection = < = = SIRequal − gain
ni′ + ∑ nk′
N
ni′
2 2 2
∑ n′
2
k ≠i k
k =1
Therefore, it can be concluded that the performance of maximal-ratio combining is the

best, equal-gain combining is the next best, and selection combining is the last.
2-17 a. Explain the operation of the RAKE receiver in your own words.
A RAKE receiver boosts up the signal reception performance by combining

separately-arriving rays of a signal transmitted over a fading channel. The technique
can only be applied to very wideband wireless systems, such as the CDMA, in which
the delay spread over a fading channel is greater than the symbol period, and as a result
individual components of the multipath signal can be separately distinguished. The
differential delays as well as relative phases and amplitudes of the individual multipath
components need to be estimated accurately, so that the different received rays can be
shifted in time, compensating for the differential delays, and then be combined using
the maximal-ratio scheme. In practice, only a few of the earliest arriving rays, normally
the strongest in power, are used to carry out the RAKE processing.
b. Two third-generation CDMA systems are discussed in chapter 10. The first, W-CDMA,
uses a chip rate of 3.84 Mcps (million chips per second), with a corresponding chip
duration of 0.26 µsec; the second system, cdma2000, uses a chip rate of 1.2288 Mcps,
with a chip duration of 0.81 µsec. (This is the same chip rate used by the
second-generation CDMA system IS-95 discussed in chapters 6 and 8.) Explain the
statements made in chapter 10 that RAKE receivers can be used to provide multipath
time-diversity for paths differing in time by at least those two chip durations,
respectively. Which system potentially provides better RAKE performance?
In CDMA, multipath components separated in time over a chip duration are orthogonal
to each other and hence can be individually received by applying the same
pseudorandom code at different times which correspond to the arrivals of the delayed
signals. The receiving system can rapidly scan through different delay values of the
received rays, searching for the sequence corresponding to the one transmitted. Once a
number of such differentially-delayed received signals are identified, delay
compensation can be carried out and maximal-ratio combining used to recover the
transmitted sequence.
W-CDMA, having a shorter chip duration, should potentially provide a better RAKE
performance.
Chapter 3
3-1 a. Verify the entries in Table 3-1.
For 3-cell reuse, D/R = 6, SIR = 1 / (5-n + 7-n). Substituting n = 3 and n = 4, we have SIR =
19.6 dB and 27 dB respectively.
For 4-cell reuse, D/R = 8, SIR = 1 / (7-n + 9-n). Substituting n = 3 and n = 4, we have SIR =
23.7 dB and 32.4 dB respectively.
b. Compare with the effect of including second-tier interferers as well. Verify that SIRdB =
19.2 dB for the case of D/R = 6 and n = 3, as indicated in the text.
Including second-tier interferers, the expression for SIR becomes

R −n
SIR =
( D − R )− n + ( D + R ) − n + (2 D − R )− n + (2 D + R)− n
Put D/R = 6 and n = 3,
1
SIR = = 82.5 or 19.2dB
(5) + (7) + (11) −3 + (13) −3
−3 −3
3-2 a. Consider a hexagonal cell of radius R, as shown in Fig. 3-4. Verify the area of the cell is
3 3R 2 / 2 = 2.6 R 2 . Show that, for reuse cluster size C, the center-to-center spacing
between the closest interfering cells is given by DC / R = 3C .
The area of a hexagonal cell is the sum of the areas of two trapezoids, therefore
3R 1 3 3 R 2
Cell Area = 2 × [( R + 2 R) × × ]= = 2.6 R 2
2 2 2
Assume a C-cell reuse pattern. The area of a cluster consisting of C cells can be
approximated by a larger hexagon whose distance between opposite edges is equal to the
distance between the closest interferers DC. This implies the hexagon has a radius
RC = DC / 3 . Its area is then AC = 3 3RC2 / 2 = 3DC2 / 2 . The number of cells within a
cluster is given by the ratio of the cluster area to the cell area, i.e.
2
Cluster Area 3DC2 / 2 1  DC 
C= = =   ⇒ DC / R = 3C
Cell Area 3 3R 2 / 2 3  R 
The geometries of the cells and the hexagonal clusters for C = 3 and C = 4 are depicted in
the figure below.
2 1
2
1 1 3 4 3
3
1 2
R3 D3
R4 D4
b. Referring to Fig. 3-6, fill in the details of the derivation of (3-4).

With reference to Fig. 3-6, a vector D between the
centers of any two hexagonal cells can be given by a

linear combination of the vectors 3Ru and 3Rv ,
3Rv
i.e. D = 3R(iu + jv ) , where u , v are unit vectors,

and i, j are any integers. 3Ru

Substitute u = (cos 30 ) x + (sin 30 )v , we have

D = 3R[(i cos 30 ) x + (i sin 30 + j )v ] 3Rx

Since x ⊥ v ,
D 2 = 3R 2 [(i cos 30 ) 2 + (i sin 30 + j ) 2 ] = 3R 2 (i 2 + j 2 + ij )
Combining this equation with the relation C = DC2 / 3R 2 , we get C = i 2 + j 2 + ij .
3-3 a. Verify the entries of Table 3-3.
C D/R SIR SIRdB

n=3 n=4 n=3 n=4
3 3 3.46 8.64 5.39 9.36
7 4.58 14.48 61.84 11.61 17.91
12 6 33.97 195.97 15.31 22.92
b. Consider the entries in Table 3-3 for the 7-reuse system. Compare these numbers with
those obtained using the two other approximations suggested in the text. Can you propose
another approximation for the location of the interferers?
Stuber’s SIR approximation:

C D/R SIR SIRdB
n=3 n=4 n=3 n=4
3 3 2.81 6.35 4.49 8.03
7 4.58 13.20 53.38 11.20 17.27
12 6 32.16 179.33 15.07 22.54
Rappaport’s SIR approximation:

C D/R SIR SIRdB
n=3 n=4 n=3 n=4
3 3 2.56 5.76 4.09 7.61
7 4.58 12.20 48.69 10.86 16.87
12 6 30.12 165.53 14.79 22.19
Among the three approximations, Rappaport’s method is the most conservative, since it
always gives the smallest SIR approximation for the same values of C and n. Stuber’s
method is the next conservative. The method based on (3-6) is the most optimistic.
Another possible approximation is

1
SIR ≈
( D / R − 1) + ( D / R + 1) + 2 D − n + 2( D / R + 0.5) − n
−n −n
With reference to Fig. 3-7, the interferers at distances (D - R) and (D + R) are located at
points 1 and 4 respectively. The two interferers at D correspond to points 2 and 6. Finally,
the two interferers at (D + 0.5R) correspond to points 3 and 5. The results obtained by this
approximation are summarized as follows:
C D/R SIR SIRdB
n=3 n=4 n=3 n=4
3 3 3.83 9.58 5.83 9.81
7 4.58 15.75 68.34 11.97 18.35
12 6 36.42 213.67 15.61 23.30
3-4 Calculate the worst-case uplink SIR assuming the co-channel interference is caused only by
the closest interfering mobiles in radio cells a distance D = 3.46R away from the cell. Assume
the simplest path-loss model g ( d ) = 1/ d 4 . Repeat for n = 3.
The worst-case uplink SIR occurs when the mobile is at the edge of the cell, i.e. at a distance R
from the base station, and all the six first-tier interferers are at a distance (D – R) from the base
station under consideration. The worst-case uplink SIR is given by
1
SIR =
6( D / R − 1) − n
For n = 4, SIR = 6.1 or 7.86 dB; for n = 3, SIR = 2.48 or 3.95 dB.
3-5 Calculate the Erlang loads on a system for the following cases:
1. An average call lasts 200 sec.; there are 100 call attempts per minute.
traffic load = (100 / 60) × 200 = 333 Erlangs
2. There are 400 mobile users in a particular cell. Each user makes a call attempt every 15
minutes, on the average. Each call lasts an average of 3 minutes.
traffic load = 400 × (1/15) × 3 = 80 Erlangs
3. The number of users in 2. is increased to 500; each user makes a call attempt every 20
minutes, on the average. Repeat if the average call length doubles to 6 minutes.
Case (i) 1/ µ = 3 mins traffic load = 500 × (1/ 20) × 3 = 75 Erlangs
Case (ii) 1/ µ = 6 mins traffic load = 500 × (1/ 20) × 6 = 150 Erlangs
3-6 Consider a mobile system supporting 100 channels per cell. A call blocking probability of 1%
is desired. Mobile users typically use their cell phones once per 10 minutes, on the average,
their calls lasting an average of 10 minutes. Say the system is concentrated in an urban area
with a density of 500 cell phones per km2. Calculate the required cell radius if a hexagonal
topology is assumed. Repeat if the mobile users “stay on the line” for 4 min., on the average.
What happens to the cell size if, in addition, they start using their phones more often?
For 100 channels per cell, to keep the blocking probability PB at 1% or less, based on the
Erlang-B formula, the total traffic load cannot exceed 84 Erlangs. Since the traffic load
contributed by each user Aper user = (1/10) ×10 = 1 , 84 users can be supported per cell. The cell
radius should be chosen such that on average there are 84 users in a cell, i.e.
cell area × user density = 2.6 R 2 × 500 = 84 ⇒ R = 0.254km or 254m
With average call length reduced to 4 mins., Aper user = (1/10) × 4 = 0.4 and the number of
users allowed per cell increases to 84/0.4 = 210. In this case,
R = 210 /(2.6 × 500) = 0.402km or 402m
If the users start using their phones more often, the traffic load per person will increase. Hence,
fewer users can be accommodated per cell and the cell size should be reduced accordingly so
as to maintain the same total traffic and call blocking probability.
3-7 The worst-case uplink SIR is to be calculated for a C = 3 reuse system, for which path loss
may be modeled by
g ( d ) = d −1.3 0 ≤ d ≤ 150m.
−1.3 −3.5
= 150m. (d /150m.) 150m ≤ d [See (2-6) in Chapter 2]
Do this for two cases: cell radius = 1000m and cell radius = 100m.
The worst-case uplink SIR occurs when the mobile is at the edge of the cell, i.e. at a distance R
from the base station, and all the six first-tier interferers are at a distance (D – R) from the base
station under consideration. For C = 3, D = 3R.
Case (i) cell radius R = 1000m, D = 3000m

150−1.3 ( R /150) −3.5 (1000) −3.5
SIR = = = 1.886 or 2.75dB
6 × 150−1.3 [( D − R ) /150]−3.5 6 × (2000) −3.5
Case (ii) cell radius R = 100m, D = 300m

R −1.3 (100) −1.3
SIR = = = 0.773 or − 1.12dB
6 ×150−1.3[( D − R) /150]−3.5 6 ×150−1.3 (200 /150) −3.5
3-8 Refer to section 3-4 of the text. The statement is made that IS136 or D-AMPS systems capable
of handling three times the traffic of the analog AMPS system can cover cells with radii
3 = 1.7 times as large. Explain. What assumptions are implicit in this statement?
In this statement, the implicit assumptions are that the user density is uniform and the same
blocking probability is maintained. With TDMA technology, IS136 can support three times as
many users per cell without increasing the blocking probability. If the user density is uniform,
that means the cell size can be three times as large, or equivalently the cell radius can be
3 = 1.7 times as large.
3-9 Consider a GSM system with a reuse parameter C = 3. For this system, 330 channels per cell
become available. Assume that the allowable Erlang traffic load is about the same number.
(The reader is encouraged to consult tables or graphs of the Erlang-B formula to determine the
resultant call blocking probability.) Assuming a hexagonal cell model, calculate the allowable
cell radius for a suburban region with a mobile density of 200 mobiles/km2. Repeat for a
system with a reuse parameter of 4. How do these results change for an urban region with a
mobile density of 1000 mobiles/km2? In all cases assume a typical user makes a call once per
15 minutes on the average, the call lasting, on the average, 200 sec.
With an Erlang traffic load approximately equal to the number of channels per cell, the
blocking probability is about 10%.
Traffic load per user Aper user = [1/(15 × 60)] × 200 = 2 / 9

2 3 3 2 R2
Total traffic load per cell A = ⋅ R ⋅ mobile density = ⋅ mobile density
9 2 3
Case (i) C = 3, mobile density = 200 mobiles/km2

A = ( R 2 / 3) ⋅ 200 = 330 ⇒ R = 1.69km
Case (ii) C = 4, mobile density = 200 mobiles/km2

A = ( R 2 / 3) ⋅ 200 = 330 × 3 / 4 ⇒ R = 1.46km
Case (iii) C = 3, mobile density = 1000 mobiles/km2

R = 1.69 / (1000 / 200) = 0.76km
Case (iii) C = 4, mobile density = 1000 mobiles/km2

R = 1.46 / 5 = 0.65km
3-10 Refer to tables of the error function defined by (3-12). Equation (3-15b) represents the
fraction of acceptable service area within a cell assuming shadow fading with log-normal
variance σ2 and power-variation parameter n. Show this fraction is 0.71 for σ = 9 dB and n =
3. Show the fraction increases to 0.83 for σ = 6dB and n = 4. Try various other combinations
of σ and n, and repeat.
1
0.9
0.8
0.7
0.6
erf(x)
0.5
0.4
0.3
0.2
0.1
0
0 0.5 1 1.5 2 2.5 3
x
σ
For σ = 9 dB and n = 3, = 1 , erf (1) = 0.84 , and Fu = 0.5 + 0.5e(1 − 0.84) = 0.72 .
3n
σ
For σ = 6 dB and n = 4, = 0.5 , erf (0.5) = 0.52 , and Fu = 0.5 + 0.5e0.5 (1 − 0.52) = 0.81 .
2
3n
As the ratio σ / n increases, the fraction of useful service area Fu decreases. A plot of Fu vs.
σ / n is shown below.
0.95
0.9
0.85
0.8
Fu
0.75
0.7
0.65
0.6
0.55
0 1 2 3 4 5 6 7 8 9 10
σ /n
Chapter 4
4-1 Refer to Fig. 4-1. Say there are 10 channels assigned per cell.
a. How many channels are there in the system?
70.
b. The DCA algorithm BDCL described in the text is to be implemented for this system:
Make a channel table numbering the channels from 1 to 10. Apply channel ordering as
described in the text, assigning local calls priority from 1 up, assigning channels to be
borrowed by neighboring cells priority from 10 down. Focus on one cell. Work out
different cases of the application of this algorithm, with different numbers of calls in
progress, both local, and those using channels borrowed from neighboring cells.
Demonstrate channel locking for this example, and then show how directional locking can
help alleviate this problem. Include examples that demonstrate the features of immediate
channel reallocation described in the text.
With reference to Fig. 4-1, consider cell 7.

Nominal Borrowed
Events \ Channels
1 2 3 4 5 6 7 8 9 10 11 12
1. Initial Configuration. X1 X2 X3 X4 X5 B2(3) B1(3) L1(1)
2. Arrival of a local call. X1 X2 X3 X4 X5 X6 B2(3) B1(3) L1(1)
3. Cell 3 requests to borrow a channel. X1 X2 X3 X4 X5 X6 B3(3) B2(3) B1(3) L1(1)
4. Cell 2 requests to borrow a channel. With X1 X2 X3 X4 X5 X6 B3(3) B2(3) B1(3) L1(1)
directional blocking, channel 10 can be B4(2)
borrowed by cells 1, 2 and 4.
5. Arrival of two local calls. X1 X2 X3 X4 X5 X6 B3(3) B2(3) B1(3) L1(1) X8 X7
B4(2)
6. Channel 10 unlocked by interfering cell. X1 X2 X3 X4 X5 X6 B3(3) B2(3) B1(3) B4(2) X8 X7

7. Local call on channel 12 terminates. X1 X2 X3 X4 X5 X6 B3(3) B2(3) B1(3) B4(2) X8
8. Local call on channel 5 terminates. X1 X2 X3 X4 X8 X6 B3(3) B2(3) B1(3) B4(2)
9. Local call on channel 4 terminates. X1 X2 X3 X6 X8 B3(3) B2(3) B1(3) B4(2)
10. Foreign call on channel 10 terminates. X1 X2 X3 X6 X8 B2(3) B1(3) B3(3)
Xj – channel used by local call j; Bj(k) – j th borrowed channel to neighboring cell k; Lj(k) – j th locked
channel due to interfering cell k
c. Compare this algorithm with another one, either taken from the literature, or one you have
invented yourself.
(Leave to readers.)
4-2 Problem 4-1 above is to be repeated for the one-dimensional case of Fig. 4-2.
a. How many channels are there in the system if 10 channels are assigned per cell?
20.
b. Show how the BDCL algorithm is implemented for this system, with different numbers of
calls in progress.
With reference to Fig. 4-2, choose i = 2, consider cell 2 in the middle.

Nominal Channels
Events
1 2 3 4 5 6 7 8 9 10
1. Initial Configuration. X1 X2 X3 X4 X5 X6 L1(1)
2. Arrival of a local call. X1 X2 X3 X4 X5 X6 X7 L1(1)
3. Cell 3 requests to borrow a channel. With X1 X2 X3 X4 X5 X6 X7 L1(1)
directional blocking, channel 10 can be B1(3)
borrowed by cell 3.
4. Cell 1 requests to borrow two channels. X1 X2 X3 X4 X5 X6 X7 B3(1) B2(1) L1(1)
B1(3)
5. Cell 3 requests to borrow one more channel. X1 X2 X3 X4 X5 X6 X7 B3(1) B2(1) L1(1)

B4(3) B1(3)
6. A new local call arrives. Since all available X1 X2 X3 X4 X5 X6 X7 B3(1) B2(1) L1(1)
channels in the system are occupied, the new B4(3) B1(3)
call must be blocked.
7. Local call on channel 1 terminates. X7 X2 X3 X4 X5 X6 B3(1) B2(1) L1(1)
B4(3) B1(3)
8. Two foreign calls on channels 9 and 10 due X7 X2 X3 X4 X5 X6 B3(1) B2(1) L1(1)

to cell 3 terminates
9. Channel 10 unlocked by cell 1. X7 X2 X3 X4 X5 X6 B2(1) B3(1)
Xj – channel used by local call j; Bj(k) – j th borrowed channel to neighboring cell k; Lj(k) – j th locked
channel due to interfering cell k
c. Consider the case of 10 calls in progress in cell i. Say cell i-1 has 7 calls in progress. A
mobile user in cell i now attempts to make a call. The base station for that cell borrows a
channel from cell i-1. Applying the BDCL algorithm, are any channels now locked?
Which one(s)? Show, by example, that borrowing of channels does not propagate, as
stated in the text.
The channel borrowed from cell i-1 is locked out in cell i+1 because of reuse constraints.
An example that borrowing of channels does not propagate is shown in row 6 of the table
in part b.
4-3 Calculate the blocking probability for the one-dimensional BDCL algorithm for the case of 2m
= 2 channels in the system. Assume homogeneous traffic conditions. Show that (4-8) results.
Plot this probability vs. the traffic intensity A in erlangs and compare with the FCA result.
With reference to Fig. 4-2, let x1, x2, x3 be the number of channels occupied in cells i, i+1, and
i-1 respectively. The triplet (x1, x2, x3) represents the vector of active states of the three-cell
group. The tables below list all the blocking and non-blocking states and their corresponding
steady-state probability obtained by Eqn. (4-2). With the assumption of homogeneous traffic
conditions, the average call arrival rate λ and the average call holding time 1/µ are the same in
all cells, and A=λ/µ.
Blocking States Non-Blocking States

P[x1, x1, x3] P[x1, x1, x3]
x1 x2 x3 x1 x2 x3
2 0 0 0 0 0 G-1
0 2 0 G-1A2/2 1 0 0
0 0 2 0 1 0 G-1A
1 0 1 0 0 1
G-1A2
1 1 0 0 1 1 G-1A2
0 1 2
G-1A3/2
0 2 1
1 1 1 G-1A3
0 2 2 G-1A4/4
∑
all states
P[ x1 , x2 , x3 ] = ∑
blocking states
P[ x1 , x2 , x3 ] + ∑
non − blocking states
P[ x1 , x2 , x3 ] = 1
−1 A2 A3 A4
⇒ G [(3 × + 2A + 2×
2
+ A + ) + (1 + 3 A + A2 )] = 1
3
2 2 4
7 A 4
9 A4
⇒ G = [( A2 + 2 A3 + ) + (1 + 3 A + A2 )] = 1 + 3 A + A2 + 2 A3 +
2 4 2 4
B= ∑
blocking states
P[ x1 , x2 , x3 ]
7 2 A4
A + 2 A3
+
7 A4 2 4
= G −1 ( A2 + 2 A3 + ) =
2 4 9 2 A4
1 + 3 A + A + 2 A3 +
2 4
Comparing the blocking probability curves for DCA and FCA in this case of m = 1 channel per
cell, DCA outperforms FCA at traffic loads below 0.75 erlangs. Above this value of load, FCA
provides slightly better performance.
0
Blocking Probability vs. Erlang Traffic Load (m = 1)
10
Blocking Probability B
-1
10
FCA
DCA
-2
10
0 0.5 1 1.5 2 2.5 3
Traffic Load A (Erlang)
4-4 Consider the case of 2m = 4 channels in a one-dimensional BDCL system.
a. Show there is a total of 55 possible states.
At x1 = 4, there is only one state (4, 0, 0). At At x1 = 3, there are 22 = 4 states: (3, 1, 0), (3, 1,
1), (3, 0, 1), and (3, 0, 0). Continuing to x1 = 0, there are 52 = 25 states. Therefore, the total
number of possible states is given by
5
∑k
k =1
2
= 55 states
b. Indicate some of the 25 blocking states. Explain, in words, why they are blocking states.
Blocking States 2 0 2 A4/4 0 4 0 A4/24

G⋅P[x1, x1, x3]
x1 x2 x3 2 1 2 A5/4 0 4 1 A5/24
4 0 0 A4/24 1 3 0 A4/6 0 4 2 A6/48
3 1 0 A4/6 1 3 1 A5/6 0 4 3 A7/144
3 1 1 A5/6 1 3 2 A6/12 0 4 4 A8/576
3 0 1 A4/6 1 3 3 A7/36 0 0 4 A5/24
2 2 0 A4/4 1 0 3 A4/6 0 1 4 A6/48
2 2 1 A5/4 1 1 3 A5/6 0 2 4 A7/144
2 2 2 A6/8 1 2 3 A6/12 0 3 4 A7/144
They are blocking states because the number of active states in any two adjacent cells: (x1
+ x2) or (x1 + x3), is equal to the total number of channels (=4) in the system. Therefore, all
channels are occupied and any new call must be blocked.
c. Calculate and plot the blocking probability as a function of the Erlang load A, assuming
homogeneous traffic conditions. Compare with the plot of Fig.4-5. Plot the FCA blocking
probability as well and compare with the DCA result. Using the results of both this
problem and problem 4-3, discuss the effect of increasing the number of channels on both
the FCA and DCA algorithms.
Non-Blocking 1 1 2 A4/2 0 2 2 A4/4

States G⋅P[x1, x1, x3] 1 1 1 A3 0 2 1 A3/2
x1 x2 x3 1 1 0 A2 0 2 0 A2/2
3 0 0 A3/6 1 0 2 A3/2 0 1 3 A4/6
2 1 1 A4/2 1 0 1 A2 0 1 2 A3/2
2 1 0 A3/2 1 0 0 A 0 1 1 A2
2 0 1 A3/2 0 3 3 A6/36 0 1 0 A
2 5
2 0 0 A /2 0 3 2 A /12 0 0 3 A3/6
1 2 2 A5/4 0 3 1 A4/6 0 0 2 A2/2
1 2 1 A4/2 0 3 0 A3/6 0 0 1 A
3 5
1 2 0 A /2 0 2 3 A /12 0 0 0 1
A8 3 1 2 2 1 2 3 2 2 4 2
∑
blocking states
P[ x1 , x2 , x3 ] = G −1[ +( + ) A7 + ( + + ) A6 + ( + + ) A5 + ( + + ) A4 ]
576 144 36 48 12 8 24 6 4 24 6 4
1 8 7 7 1 6 13 5 5 4
= G −1 ( A + A + A + A + A )
576 144 3 12 4
A6 2 1 1 2 3 6 3 3
∑
non − blocking states
P[ x1 , x2 , x3 ] = G −1[ + ( + ) A5 + ( + + ) A4 + ( + + 1) A3 + ( + 3) A2 + 3 A + 1]
36 12 4 4 6 2 2 6 2
1 6 5 5 25 4 9 3 9 2
= G −1 ( A + A + A + A + A + 3 A + 1)
36 12 12 2 2
1 8 7 7 13 6 3 5 10 4 9 3 9 2
∑
all states
P[ x1 , x2 , x3 ] = 1 ⇒ G = (
576
A +
144
A +
36
A + A + A + A + A + 3 A + 1)
2 3 2 2
1 8 7 7 1 6 13 5 5 4
A + A + A + A + A
B = ∑ P[ x1 , x2 , x3 ] = 576 144 3 12 4
1 8 7 7 13 6 3 5 10 4 9 3 9 2
blocking states
A + A + A + A + A + A + A + 3A + 1
576 144 36 2 3 2 2
Comparing the blocking probability curves for DCA and FCA in this case of m = 2
channels per cell, DCA outperforms FCA at traffic loads below 1.6 erlangs. In general, the
lighter the load, the greater the improvement over FCA. As the number of channels per
cell increases, the overall blocking probability decreases and the range of loads over
which the DCA advantage is manifested increases.
0
Blocking Probability vs. Erlang Traffic Load
10
DCA
FCA
FCA
-1
10
m=1
Blocking Probability B DCA
-2
10
m=2
-3
10
-4
10
0 0.5 1 1.5 2 2.5 3
Traffic Load A (Erlang)
4-5 Refer to Figs. 4-7 and 4-8. The DBA power algorithm is to be applied to each of these
examples.
a. Find the R matrices in each case. In particular, show that (4-26) is the R matrix for the
case of Fig. 4-7.
Let dij be the distance from an interfering mobile in cell j to the base station in cell i. The
elements of the matrix R is defined as rij = (dii / dij)n, where n is the path loss exponent. In
the examples, n is chosen to be four and each cell is defined to have a width of one unit.
Referring to the figures, obviously the distance of a mobile to its service base station is
always 0.5 unit. For the scenario in Fig. 4-7,
 0.5 4 0.5 4 0.5 4 0.5 4 0.5 4 0.5 4   1 1 1 1 1 
( 0.5 ) ( ) ( ) ( ) ( ) ( ) 1
1.5 3.5 5.5 7.5 9.5   34 74 114 154 194 
   
( )4   4
( 0.5 )4 0.5 0.5 0.5 0.5 0.5 1 1 1 1 1 
( )4 ( )4 ( )4 ( )4 1
 1.5 0.5 1.5 3.5 5.5 7.5   3 34 74 114 154 
   
( 0.5 )4 0.5 0.5 0.5 0.5 0.5 4   1 1 1 1 1 
( )4 ( )4 ( )4 ( )4 ( ) 1
 3.5 2.5 0.5 1.5 3.5 5.5   7 4 54 34 74 114 
R1 =  =
0.5 0.5 0.5 0.5 0.5 0.5   1 1 1 1 1 
( )4 ( )4 ( )4 ( )4 ( )4 ( )4   4 1 
 5.5 4.5 2.5 0.5 1.5 3.5   11 94 54 34 74 
 0.5 4 0.5 4 0.5 4 0.5 0.5 0.5   1 1 1 1 1 
( ) ( ) ( ) ( )4 ( )4 ( )4   4 1 
 7.5 6.5 4.5 2.5 0.5 1.5  15 134 94 54 34 
 0.5 4 0.5 0.5 4 0.5 4 0.5 0.5   1 1 1 1 1 
( ) ( )4 ( ) ( ) ( )4 ( )4   4 1 
 9.5 8.5 6.5 4.5 2.5 0.5  19 17 4 134 94 54 
Similarly, for the scenario in Fig. 4-8,
 0.5 4 0.5 4 0.5 4 0.5 4 0.5 4 0.5 4   1 1 1 1 1 
( 0.5 ) ( ) ( ) ( ) ( ) ( ) 1
1.5 4.5 5.5 8.5 9.5   34 94 114 17 4 194 
   
( )4   4
( 0.5 ) 4 0.5 0.5 0.5 0.5 0.5 1 1 1 1 1 
( )4 ( )4 ( )4 ( )4 1
 1.5 0.5 2.5 3.5 6.5 7.5   3 54 74 134 154 
   
( 0.5 ) 4 0.5 0.5 4 0.5 0.5 4 0.5 4   1 1 1 1 1 
( )4 ( ) ( )4 ( ) ( ) 1
 3.5 2.5 0.5 1.5 4.5 5.5   7 4 54 34 94 114 
R2 =  =
0.5 0.5 0.5 0.5 0.5 0.5   1 1 1 1 1 
( ) 4 ( )4 ( )4 ( )4 ( )4 ( )4   4 1 
 5.5 4.5 1.5 0.5 2.5 3.5   11 94 34 54 74 
 0.5 4 0.5 4 0.5 0.5 0.5 0.5   1 1 1 1 1 
( ) ( ) ( )4 ( )4 ( )4 ( )4   4 1 
 7.5 6.5 3.5 2.5 0.5 1.5  15 134 74 54 34 
 0.5 4 0.5 4 0.5 4 0.5 4 0.5 4 0.5   1 1 1 1 1 
( ) ( ) ( ) ( ) ( ) ( )4   4 1 
 9.5 8.5 5.5 4.5 1.5 0.5  19 17 4 114 94 34 
b. Find the six eigenvalues of R in each case and show they agree with the values appearing
in the text.
For the case of Fig. 4-7, the six eigenvalues of R1 are 1.0137, 1.0073, 1.0018, 0.9969,
0.9934, and 0.9871.
For the case of Fig. 4-8, the six eigenvalues of R2 are 1.0140, 1.0124, 1.0108, 0.9884,
0.9876, and 0.9869.
c. Show Tr(R) = M = ∑zi i in each case.
In both cases, ∑z i i = 6 , which is equal to the trace of the R matrix.
d. Run the DBA algorithm for each of the cases and show the algorithm converges to the two
values of SIR, 18.7 dB and 18.5 dB, indicated in the text. Plot the SIR at a number of
base stations as a function of iteration n for at least 20 iterations. Comment on the results.
In the simulations, the initial power vector is arbitrarily chosen to be [1 2 3 2 1 3] for both
scenarios, and c ( j ) = 1/ max( P ( j ) ) as described in the text. The simulation results show
that the algorithm converges to the theoretical limiting values of SIR, 18.7dB and 18.5dB.
For the case of Fig. 4-7, it takes 987 iterations to have the SIR’s of all cells fall within ±
0.1 dB about the limiting value, while for the case of Fig. 4-8, it only takes 175 iterations.
In general, the DBA algorithm has a slow convergence, which is not practical for a
dynamic system with mobile users roaming around. By the time convergence is reached,
the resultant power vector will no longer be optimum for the new mobile positions.
For the case of Fig. 4-7:
Convergence of SIR at Different Base Stations as Number of Iterations Increases
32
Cell 5
30
28
26
24
SIR γ(n) (dB)
22
i
20 Cell 2
Cell 3
18 Cell 1
Cell 0
16
Cell 4
14
12
0 100 200 300 400 500 600 700 800 900 1000
Number of Iterations n

24
Cell 5
22
Cell 1
20
Cell 2
SIR γ(n) (dB)
18
i
Cell 3
16 Cell 0
14 Cell 4
12
0 20 40 60 80 100 120 140 160 180
4-6 This problem is concerned with the analysis of the DPC algorithm, as defined by (4-29).
a. Starting with (4-29) show the matrix-vector form of this algorithm may be written in the
form of (4-30), with the matrix A defined as indicated in the text.
From (4-29), Pi ( n ) = c ( n −1) Pi ( n −1) / γ in −1

Substitute (4-15) into (4-29), we have
M
Pi ( n ) = c ( n −1) (∑ Pj( n −1) rij − Pi ( n −1) )
j =1
In matrix-vector form, it becomes

P ( n ) = c ( n −1) (RP ( n −1) − IP ( n −1) ) = c ( n −1) (R − I )P ( n −1) = c ( n −1) AP ( n −1) , where A = R – I.
b. Relate the eigenvalues of the A matrix to those of the matrix R defined for the DBA
algorithm.
Let {αi}, {βi} be the eigenvalues of R and A respectively. By definition, it means

det(R − α i I ) = 0 (1)
and det( A − β i I ) = 0 (2)
Substitute A = R – I into (2), we have

det[(R − I ) − β i I ] = 0 or det[R − ( β i + 1)I ] = 0 (3)
Comparing (3) and (1), we get

βi = α i − 1
c. Prove the DPC algorithm converges to the same SIR as the DBA algorithm. Show why
this algorithm is expected to converge more rapidly than the DBA algorithm.
The proof for the limiting SIR of DPC is very similar to that for DBA as stated in the text,
except that matrix R is now replaced by matrix A, and accordingly z* by β* which denotes
the maximum eigenvalue of A, and P* now represents the eigenvector corresponding to β*.
Therefore, (4-23) becomes
n−2
Pi ( n ) = c ( n −1) Pi*a1 ( β * ) n ∏ c ( j ) (4)
j =0
Substitute (4) into (4-29), we have

n−2
1
Pi ( n ) = c ( n −1) ⋅ Pi * a1 ( β * ) n −1 ∏ c ( j ) (5)
γ i( n −1) j =0
Comparing (4) and (5) and take the limit in n, we get

lim n →∞ γ i( n −1) ≡ γ * = 1/ β *
From the result in part (b), β * = z * − 1 . Hence, the limiting SIR for the DPC algorithm is
γ * = 1/( z* − 1) , which is the same for the DBA algorithm.
For the SIR’s to converge, ( z * ) n and ( β * ) n must dominate all the terms in ∑ j
a j z nj φ j
and ∑ j
a j β jn φ j respectively. Thus, the speed of convergence should increase as the
ratios ( z2 / z * ) or ( β 2 / β * ) decreases, where z2 and β 2 denote the second largest
eigenvalues for the two algorithms. Suppose z * = z2 + ∆ and z * = 1 + ε . Then,
z2 ∆
for DBA = 1− *
z *
z
β2 ∆ ∆ ∆
and for DPC = 1− * = 1− * = 1−
β *
β z −1 ε
In order that DPC has a faster convergence,
β2 ∆ z z ∆ z ∆ ∆ z
= 1 − < 2* ⇒ 1 − 2* < < 1 + 2* ⇒ < < 1 + 2* ≈ 2
β *
ε z z ε z z *
ε z
The condition holds in general. Therefore, the DPC algorithm is expected to converge
more rapidly than the DBA algorithm.
4-7 Repeat problem 4-5 for the DPC algorithm: Find the A matrices for the two cases of Figs. 4-7
and 4-8. Find the eigenvalues in each case. Compare with those of the DBA algorithm. Run
and plot the algorithm in each case for at least 20 iterations and compare with the DBA
algorithm.
For the case of Fig. 4-7,

 0 1/ 34 1/ 7 4 1/114 1/154 1/194 
 4 
 1/ 3 0 1/ 34 1/ 7 4 1/114 1/154 
 1/ 7 4 1/ 54 0 1/ 34 1/ 7 4 1/114 
A1 = R1 − I =  4 4

1/11 1/ 9 1/ 54 0 1/ 34 1/ 7 4 
1/154 1/134 1/ 94 1/ 54 0 1/ 34 
 
1/19 1/17 1/13 1/ 9 0 
4 4 4 4
1/ 54
The eigenvalues of A1 are 0.0137, 0.0073, 0.0018, -0.0031, -0.0066, and -0.0129.
For the case of Fig. 4-8,

 0 1/ 34 1/ 94 1/114 1/17 4 1/194 
 4 
 1/ 3 0 1/ 54 1/ 7 4 1/134 1/154 
 1/ 7 4 1/ 54 0 1/ 34 1/ 94 1/114 
A2 = R 2 − I =  4 4 4

1/11 1/ 9 1/ 3 0 1/ 54 1/ 7 4 
1/154 1/134 1/ 7 4 1/ 54 0 1/ 34 
 
1/19 1/17 1/11 1/ 9 0 
4 4 4 4
1/ 34
The eigenvalues of A2 are 0.0140, 0.0124, 0.0108, -0.0116, -0.0124, and -0.0131.
In both cases, the eigenvalues of the A matrices are one less than the eigenvalues of their
corresponding R matrices of the DBA algorithm.

23
Cell 0
22 Cell 2
21
20
SIR γ(n) (dB)
19
i
18
17
16
15
0 10 20 30 40 50 60

20.5
Cell 0
20 Cell 2
19.5
19
18.5
SIR γ(n) (dB)
18
i
17.5
17
16.5
16
15.5
0 5 10 15 20 25 30 35 40 45 50
In both cases, the convergence values of SIR are the same as those obtained by the DBA
algorithm. However, convergence of DPC is much faster. For the case of Fig. 4-7, it takes only
53 iterations to have the SIR’s of all cells fall within ± 0.1 dB about the limiting value, and for
the case of Fig. 4-8, it takes only 40 iterations. In DPC, the SIR’s of different base stations
approach the limiting value with oscillations of decreasing amplitudes, whereas in DBA the
curves are much smoother.
4-8 The DBA and DPC distributed power control algorithms are to be compared for a
one-dimensional cellular network of your choice, consisting of at least 13 cells.
a. Carry out the iterative analysis for both the two-reuse and three-reuse cases, plotting the
resultant SIRs as a function of iteration parameter at a number of base stations. Try to
choose worst-case mobile locations.
Case (i) Two-reuse, 15-cell linear cellular system
0 1 2 3 4 5 6 7
 1 1/ 54 1/ 7 4 1/114 1/154 1/ 214 1/ 234 1/ 27 4 

 4 
 1/ 3 1 1/ 34 1/ 7 4 1/114 1/17 4 1/194 1/ 234 
 1/ 7 4 1/ 34 1 1/ 34 1/ 7 4 1/134 1/154 1/194 
 
 1/114 1/ 7 4 1/ 54 1 1/ 34 1/ 94 1/114 1/154 
R1 =  A1 = R1 − I
1/154 1/114 1/ 94 1/ 54 1 1/ 54 1/ 7 4 1/114 
 
1/19 1/15 1/13 1/ 9 1/ 34 1/ 7 4 
4 4 4 4
1/ 54 1
1/ 234 1/194 1/17 4 1/134 1/ 94 1/ 34 1 1/ 34 

1/ 27 4 1/ 234 1/ 214 1/17 4 1/134 1/ 7 4 1/ 54 1 
Convergence of SIR at Different Base Stations as Number of Iterations Increases (DBA)

30
Cell 7
28
26
24
(dB)
22 Cell 4
SIR γ(n)
Cell 3
i
20
Cell 0
18 Cell 6 Cell 2
16 Cell 1
14 Cell 5
12
0 50 100 150 200 250 300 350 400 450 500
Convergence of SIR at Different Base Stations as Number of Iterations Increases (DPC)
24
Cell 0
23 Cell 2
22
21
(dB)
20
SIR γ(n)
i 19
18
17
16
15
0 5 10 15 20 25 30 35 40 45 50
Case (ii) Three-reuse, 16-cell linear cellular system
0 1 2 3 4 5
 1 1/ 7 4 1/114 1/17 4 1/ 234 1/ 314 

 4 
 1/ 5 1 1/ 54 1/114 1/17 4 1/ 254 
1/114 1/ 54 1 1/ 54 1/114 1/194 
R2 =  4 4 4
 A2 = R 2 − I
1/17 1/11 1/ 7 1 1/ 54 1/134 
1/ 234 1/17 4 1/134 1/ 7 4 1 1/ 7 4 
 
1/ 29 1/ 23 1/19 1/13 1/ 7 1 
4 4 4 4 4
Convergence of SIR at Different Base Stations as Number of Iterations Increases (DBA)

38
Cell 5
36
34
(dB)
32
SIR γ(n)
i
30
Cell 0
Cell 3
28
Cell 4
Cell 2
26
Cell 1
24
0 500 1000 1500 2000 2500 3000 3500 4000 4500
Convergence of SIR at Different Base Stations as Number of Iterations Increases (DPC)
30
Cell 0
29.5 Cell 2
29
28.5
28
SIR γ(n)
i
(dB) 27.5
27
26.5
26
25.5
25
0 5 10 15 20 25 30
b. Compare the SIR to which the DPC algorithm converges to the largest eigenvalue of the R
matrix of the DBA algorithm.
For the two-reuse case, the largest eigenvalue of R1 is 1.0144 and the DPC algorithm
converges to 18.42 dB (or 69.5) after 50 iterations.
For the three-reuse case, the largest eigenvalue of R2 is 1.0021 and the DPC algorithm
converges to 26.863 dB (or 485.6) after 30 iterations. Since the interfering cells in this
case are further away from each other, the resultant SIR is larger than the previous case.
In both cases, the limiting SIR, γ*, and the largest eigenvalue of R, z* are related by the
expression γ* = 1 / (z* - 1). For case (i), γ* = 1 / (0.0144) = 69.44, and for case (ii) γ* = 1 /
(0.0021) = 476.19. The values of γ* are very close to the SIR of convergence obtained by
simulations.
c. Find the relative powers at each of the mobile terminals and comment on the results.
Case (i) Two-reuse, 15-cell linear cellular system
Mobile 0 1 2 3 4 5 6 7
Relative power 0.1385 0.9837 1 0.1843 0.0494 0.0967 0.0975 0.0147
Cell 0 1 2 3 4 5 6 7
SIR (dB) when transmission 26.75 15.99 15.92 18.34 24.06 18.36 16.04 26.83
powers in all cells are equal
Case (ii) Three-reuse, 16-cell linear cellular system
Mobile 0 1 2 3 4 5
Relative power 0.2305 0.9628 1 0.3057 0.0908 0.0291
Cell 0 1 2 3 4 5
SIR (dB) when transmission 33.00 24.84 24.76 26.71 30.54 33.33
powers in all cells are equal
Apart from the relative powers of the mobile terminals, we also include tables listing the
SIR’s of the cells when the powers of all terminals are assumed equal. A smaller SIR
indicates that the cell is more affected by the interfering mobiles. In both cases, we can see
that a mobile is assigned a larger transmission power if its cell is more affected by
interference.
4-9 Consider the Foschini-Miljanic algorithm defined by (4-31).
a. Compare this algorithm with the DPC algorithm defined by (4-29). Show the two
algorithms converge to the same SIR.
Compare with the DPC algorithm defined by (4-29), we found that the only difference in
this algorithm is the replacement of c(n-1) by the parameter γ. Since in DPC it is only
required that c(n-1) be a positive value agreed by all mobiles in the nth round of iteration, its
replacement by a constant γ should not have any impact on the SIR of convergence.
b. Show why no inter-mobile communication is needed in carrying out this algorithm.
This algorithm is defined by (4-31), Pi ( n ) = γ Pi ( n −1) / γ i( n −1) . Since γ i( n −1) is the SIR of cell i
which can be measured directly at the base station before the nth round of iteration, and γ
is a pre-defined system parameter, no inter-mobile communication is needed in carrying
out this algorithm.
c. Apply this algorithm to the 11-cell linear system of Fig. 4-7 and compare with the results
shown in the text.
Choose γ = 70 ( < γ * = 1/ 0.0137 = 73 ). It is found that after 53 iterations the algorithm

converges to within ±0.1dB about the theoretical maximum achievable SIR of 18.65 dB.
In general, the results obtained by this algorithm are the same as those shown in the text
obtained by DPC.
Mobile 0 1 2 3 4 5
Relative power 0.8967 1 0.1914 0.0529 0.0161 0.0043
23
Cell 0
22 Cell 2
21
20
SIR γ(n) (dB)

19
i
18
17
16
15
0 10 20 30 40 50 60
4-10 The object of this problem is to provide an example of the application of the fixed-increment
power control algorithm of (4-32). Start with a cellular network of your own choice. (A
simple choice would be a one-dimensional network such as those used as examples in the
text.) Let each potentially-interfering mobile choose an initial transmit power. The base
station for each mobile estimates its SIR and then commands the mobile to reduce or increase
its transmit power by a fixed increment, as discussed in the text. This procedure is then
iterated until convergence is reached. Carry out this algorithm for various values of the
increment. Discuss your results.
Assume the fixed-increment power control algorithm is applied to the two-reuse 11-cell linear
cellular system in Fig. 4-7. Choose ∆ = 0.5 dB and γ = 18.65 dB. The plot of SIR vs. number
of iterations shows that the SIR’s of all cells converge to ± 0.25 dB about γ after 39 iterations.
However, the convergence is unstable: the SIR of each cell keeps fluctuating about γ in a
zigzag pattern. This is due to the fact that the resultant SIR’s can never be exactly equal to γ.
The problem can be remedied by stopping the adjustment of transmission power when the
SIR has fallen within a margin about the target value, i.e. Pi will only be adjusted if
SIRi − γ > δ . In general, the value of δ should be larger than ∆/2 for equilibrium to be
possible. In the second plot, we choose a margin δ of 0.25 and all the curves stabilize after 39
iterations.
In the third plot, we choose ∆ = 0.1 dB, γ = 18.65 dB and δ = 0.05dB. It takes 173 iterations
for the system to reach equilibrium. With a smaller value of ∆, it takes a longer time to reach
convergence. However, it gains the advantage of having a closer convergence of the SIR’s to
the target value.
(∆ =0.5dB; γ=18.6476dB; δ=0dB)
32 Cell 5
30
28
26
24
SIR γ(n) (dB)

22
i Cell 2
20
Cell 3
18 Cell 1
Cell 0
16
14 Cell 4
12
0 5 10 15 20 25 30 35 40 45 50

(∆ =0.5dB; γ=18.6476dB; δ=0.25dB)
32 Cell 5
30
28
26
24
SIR γ(n) (dB)
22
i
Cell 2
20
Cell 3
18 Cell 1
Cell 0
16
14 Cell 4
12
0 5 10 15 20 25 30 35 40 45 50

(∆ =0.1dB; γ=18.6476dB; δ=0.05dB)
32 Cell 5
30
28
26
24
SIR γ(n) (dB)
22
i
Cell 2
20
Cell 3
18
Cell 1
Cell 0
16
14 Cell 4
12
0 20 40 60 80 100 120 140 160 180
Chapter 5
5-1 a. Why are constant-envelope modulation techniques preferred for use on radio channels for
mobile communications?
Constant-envelope modulation techniques are preferred because they are more resistant to
random amplitude variations caused by fading in radio channels.
b. Explain, on the basis of a., why FM, rather than AM, was adopted for use with the
first-generation AMPS.
In FM, it is the frequency of the carrier that is varied in direct proportion to the amplitude
of the baseband or information signal, whereas in AM, it is the amplitude of the carrier
being modulated. Therefore, FM is a constant-envelope modulation technique and is
preferred over AM for use with the first-generation AMPS.
c. Again referring to a., why do digital mobile wireless systems avoid using some form of
QAM requiring multiple amplitude levels, despite the ability thereby to attain higher bit
rates over the channels provided?
Except for QPSK, a special type of QAM which uses four different phase angles to
distinguish four two-bit sequences while keeping the amplitude of the carrier constant, all
other forms of QAM use a combination of amplitude and phase to identify a k-bit
sequence. If the amplitude of the symbol is altered by fading in the wireless channels, the
original k-bit sequence cannot be correctly recovered. Therefore, in general, QAM is
susceptible to fading and is avoided by digital wireless systems despite its ability to attain
higher bit rates.
d. Why are bandwidth and power efficiency important considerations in the design of mobile
wireless systems?
Bandwidth efficiency is important because it is directly related to the overall capacity, i.e.
number of users, supported by the system. Power efficiency is another important
consideration mainly because of the limited energy supply via batteries for mobile
devices.
5-2 a. QPSK is to be used to transmit data over a wireless link operating at a carrier frequency of
871.5 kHz. The transmission bandwidth available at that frequency is 1.25 MHz.
Sinusoidal rolloff shaping is to be used. Calculate the data rate that can be used for the
following rolloff factors: r = 0.25, 0.5, 1. What would the comparable rates be if PSK
were to be used? Would the results differ at a different carrier frequency? Explain.
r Baud Rate = BT/(1+r) QPSK Bit Rate (bps) PSK Bit Rate (bps)
0.25 1M 2M 1M
0.5 833k 1.67M 833k
1 625k 1.25M 625k
As long as the same transmission bandwidth is available, the results would not change at a
different carrier frequency, because the shaping function h(t) and hence the zero-crossing
interval or equivalently the supported symbol rate solely depend on the available
transmission bandwidth, regardless of the carrier frequency.
b. Repeat a. for a bandwidth of 30 kHz centered about the same carrier frequency. Again,
would the results differ at a different carrier frequency?
r Baud Rate = BT/(1+r) QPSK Bit Rate (bps) PSK Bit Rate (bps)
0.25 24k 48k 24k
0.5 20k 40k 20k
1 15k 30k 15k
The results would not differ at a different carrier frequency, due to the same reasons as
stated in part a.
5-3 a. Choose an input sequence of 10 or more binary digits to be applied to a QPSK modulator
as in Fig. 5-11. Map them to the five appropriate QPSK signal pairs using Table 5-1.
Sketch the corresponding output QPSK signals using the signal representation of (5-7).
Ignore the shaping function for simplicity. Let the carrier frequency be some multiple of
1/T. Note the times at which π -radians and π / 2 -radians phase shifts occur. Correlate
these with the input bit pairs.
Choose the input sequence {sj} = {1, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 0, 1} and the bit rate R be
1bps. The input sequence can be mapped to seven QPSK (ai, bi) pairs:
(1, 1), (1, 1), (-1, 1), (-1, -1), (1, 1), (1, -1), and (-1, 1)
A phase shift of π radians occurs at t = 8 and 12 sec. when both input bits are changed at
the transition points.
A phase shift of π/2 radians occurs at t = 4, 6 and 10 sec. when only one of the input bits is
changed at the transitions.
b. Repeat a. for the OQPSK modulator of Fig. 5-15. What is the maximum phase shift in this
case?
The maximum phase shift in this case is π/2 radians.
5-4 a. Focus on QPSK. Explain, using a specified sequence of binary digits, the effect on
phase transitions of changes in the sign of first, one, and, then, both of the coefficients ai
and bi appearing in (5-7).
The effects of ai and bi on phase transitions in QPSK can be readily explained by referring
to the signal constellation diagram. If one of the coefficients ai and bi is changed, the
phasor will only swing to the adjacent quadrant, resulting in a phase change of π/2 radians.
However, if both ai and bi are changed, the phasor will swing to the diagonal quadrant,
causing a phase shift of π radians. An example of the QPSK modulation of a sequence of
binary digits can be found in the solution of question 5-3a.
bi bi bi
1 1 1
ai ai ai
-1 1 -1 1 -1 1
-1 -1 -1
QPSK signal constellation Phase shift of π/2 radians when (ai, Phase shift of π radians when (ai,
bi) changes from (1, 1) to (-1, 1) bi) changes from (1, 1) to (-1, -1)
b. Show, for the same sequence as in a., how OQPSK differs from QPSK. What is it about
this result that makes OQPSK preferable to QPSK for mobile communication?
In OQPSK, one of the sequences, say {bi}, is offset or delayed by a bit period T/2. As a
result, the transitions of {bi} only take place in the middle of each bit in {ai}, and at any
instant either ai or bi, but not both, can be changed. Hence, the maximum phase shift in
OQPSK is π/2 radians, as opposed to π radians in QPSK. In general, an abrupt phase
change of π radians results in an undesirable increase in signal transmission bandwidth
and creates problems for amplifiers and should therefore be avoided. This explains why
OQPSK is preferable to QPSK for mobile communication. An example of OQPSK
modulation can be found in the solution of question 5-3b.
c. Compare MSK with QPSK for the same sequence as in a.
In QPSK, while the carrier frequency remains constant, abrupt phase shifts of π or π/2
radians occur at the bit transition instants. On the other hand, in MSK, the phase of the
carrier is continuous, whereas the frequency is altered at bit transitions. Therefore, MSK
can also be viewed as a continuous phase (CPFSK) signal with a frequency separation
equal to half the bit rate.
5-5 a. MSK modulation is defined by (5-8) or by Fig. 5-16. Show that (5-8) may be expressed in
the equivalent forms of (5-9) and (5-10).
With reference to (5-8), consider the four cases of ai, bi:

ai = 1, bi = 1 si (t) = cos(πt/T)⋅cosω0t + sin(πt/T)⋅sinω0t = cos(ω0t – πt/T)
ai = 1, bi = -1 si (t) = cos(πt/T)⋅cosω0t – sin(πt/T)⋅sinω0t = cos(ω0t + πt/T)
ai = -1, bi = 1 si (t) = -[cos(πt/T)⋅cosω0t – sin(πt/T)⋅sinω0t] = cos(ω0t + πt/T + π)
ai = -1, bi = -1 si (t) = -[cos(πt/T)⋅cosω0t + sin(πt/T)⋅sinω0t] = cos(ω0t – πt/T + π)
From the list above, we can conclude by observation that the expression in (5-8) can be
rewritten in the equivalent forms of (5-9) and (5-10).
b. Explain, by example, the statement in the text that only one of the coefficients ai and bi
can change at a time with MSK. Why is the maximum phase change then only π/2
radians?
Same as OQPSK, in MSK one of the sequences, say {bi}, is delayed by a bit period T/2.
As a result, the transitions of {bi} only take place in the middle of each bit in {ai}.
Therefore, only one of the coefficients ai and bi can change at a time with MSK. An
example can be found in the solution of question 5-3b.
Since MSK can be viewed as an OQPSK signal with sinusoidal pulse weighting, we can
expect that it has the same maximum phase change of π/2 radians as OQPSK due to
similar reasons. However, it should be noted that with sinusoidal shaping, there is actually
no phase discontinuity in MSK. Only the signal frequency is changed according to the bit
sequences. The excess phase of an MSK signal, referenced to the unmodulated carrier
phase, is given by ±πt/T, which increases or decreases linearly during each bit period of
T/2. Therefore, at then end of each bit period, the phase of an MSK signal will either be
π/2 radians ahead or behind that of an unmodulated carrier running at a frequency of f0
during the interval.
5-6 Consider the MSK modulator shown in Fig. 5-16. Sketch the output MSK signal for a
sequence of at least 10 input bits, arbitrarily chosen. Demonstrate that the signal does have a
constant envelope. Do any phase discontinuities appear at the bit transition points? Explain.
A sketch of the output MSK signal for a 15-bit sequence can be found in the solution of
question 5-4c. From the graph, we can see that the MSK signal has a constant envelope and
there is no phase discontinuity at the bit transition points. To understand the phase continuity
property of MSK, we should refer to the expression in (5-9):
ai biπ t 0 , ai = 1
si (t ) = cos(ω0t − +θ ) ,where θ = {
T π , ai = −1
For this expression, the transitions of {ai} occur at t = (k + 1/2)T, whereas the transitions of
{bi} occur at t = kT, where k is any integer. Obviously, there is no phase change when the (ai,
bi) pair remains the same. Let us consider the two cases when one of the bits changes.
Case (i) bi changes while ai remains the same

2π
Phase Shift = ± ⋅ kT = ±2kπ or equivalently 0 radian
T
Case (ii) ai changes while bi remains the same

2π 1
Phase Shift = ± ⋅ (k + )T + (θ 2 − θ1 ) = ± (2k + 1)π ± π or equivalently 0 radian
T 2
In both cases, we have shown that there is effectively no phase change at the transition points.
Therefore, we can conclude that the phase of a MSK signal is always continuous.
5-7 a. The analysis of MSK detection following (5-10) shows that if ∆ f = R/4, as in (5-10),
successive values of ai and bi will be uniquely recovered using the FSK detector of Fig.
5-17. Carry out the analysis, filling in all of the details left out, and demonstrate the
validity of the argument used.
Let ω1 = ω0 − ∆ω and ω2 = ω0 + ∆ω , where ω0 is the angular frequency of the

unmodulated carrier, and ∆ω = π / T . Suppose cos ω1t is transmitted. Then the value of
the signal sample at decision point 1 in Fig. 5-17 is given by
T /2
2 T /2 1 T /2 1  sin 2ω1t  1 sin ω1T
O1 = ∫ cos 2 ω1tdt = ∫ (1 + cos 2ω1t )dt = t +  = +
T 0 T 0 T 2ω1  0 2 2ω1T
Since the carrier frequency f 0 is usually very high, ω1T = (2 f 0T − 1)π >> 1 . Hence,
sin ω1T
≈ 0 and O1 ≈ 1/ 2 .
2ω1T
Now consider the value of the signal sample at decision point 2:

2 T /2 1 T /2
O2 = ∫
T 0
cos ω1t ⋅ cos ω2tdt =
T ∫0
[cos(ω1 + ω2 )t + cos 2∆ωt ]dt
1  sin(ω1 + ω2 )T / 2 sin ∆ωT 
= ⋅ + 
T  ω1 + ω2 2∆ω 
sin ∆ωT
Since ω1 + ω2 >> 1 , the first term can be equated to zero and O2 ≈ . In order to
2∆ωT
have a zero output at decision point 2, we need to choose a ∆ω such that sin ∆ωT = 0 .
The smallest frequency deviation ∆ω satisfying this condition is ∆ωT = π , which
implies ∆f = 1/ 2T = R / 4 , where R is the bit rate.
In conclusion, with the FSK detector of Fig. 5-17, if a non-zero output is obtained at
decision point 1 while nothing is obtained at decision point 2, then we know a symbol of
frequency f1 is received. In that case, both ai and bi are of the same sign. On the other
hand, if a symbol of frequency f 2 is received, the decision point outputs and hence the
detection results will simply be opposite. Since only one bit can be changed at each
transition point, we can always determine the value of the new bit ai (or bi) by knowing
the value of bi (or ai) deduced previously.
b. Continuing the discussion in a. of the FSK detector of Fig. 5-17, show, as suggested in the
text, that for all possible values of ai (bi) correctly established, the detector will, an
interval T/2 later, uniquely and correctly produce the value of bi (ai) that may have been
transmitted.
To illustrate the idea of bit detection in MSK, let us consider the MSK example above.
Suppose the first bit of bi at time 0 is established to be -1, then all the subsequent bits can
be determined as follows:
Time 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Non-zero output at - 2 1 1 1 2 2 2 1 2 1 1 2 1 2
decision points 1 or 2?
ai ⋅ bi - -1 1 1 1 -1 -1 -1 1 -1 1 1 -1 1 -1
ai - 1 1 -1 -1 1 1 -1
bi -1 1 1 1 -1 1 -1 1
The binary sequence is correctly received as {0, 1, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 0, 1}, which
is just the same as the input sequence we chose in question 5-3.
5-8 Equation (5-13) represents the effect of shaping a binary pulse of width D by a gaussian filter
whose impulse response (characteristic in time) is given by (5-11) and frequency spectrum by
(5-12). Plot (5-13) as a function of time for various values of BN = BD, including specifically
the value 0.3, as well as values above and below that value. Show the effect of reducing
bandwidth B is to introduce more intersymbol interference, as indicated in the text. You will
need to use a software package, or a table, of the complementary error function erfc(x)
defined by (5-14).
In this example, the bit period is chosen to be 1. We can see that by reducing the bandwidth B,
g(t) spreads more into adjacent binary intervals, thus introducing more intersymbol
interference.
5-9 Consider π / 4 DQPSK as described in the text. The initial phase of the system is set to
start at zero. Let the incoming bit stream be 0 0 1 0 1 1 1 0 0 1 1 1, ordered from left to right.
Show the points on the constellation diagram of Fig. 5-20 through which the modulator
moves as these bits are processed, one after another.
(ai, bi) - -1, -1 1, -1 1, 1 1, -1 -1, 1 1, 1

Differential Phase Shift - -3π/4 π/4 3π/4 π/4 -π/4 3π/4
Resultant Phase 0 -3π/4 -π/2 π/4 π/2 π/4 π
6
5
4
1
3
2
5-10 A transmission bandwidth of 2 MHz is available. Nyquist rolloff shaping is used in

transmitting data.
a. Find the bit rates that may be transmitted over this channel using PSK for rolloff factors of
(1) 0.2, (2) 0.25, and (3) 0.5.
From equation (5-6a), R = BT /(1 + r ) . Therefore, the bit rates that may be supported for
the different rolloff factors are (1) 1.67Mbps, (2) 1.6 Mbps, and (3) 1.33 Mbps.
b. Nyquist rolloff shaping of 0.25 is used. It is desired to transmit at a rate of 6.4 Mbps over
this channel. Show how this may be done. Repeat for 9.6 Mbps.
With a Nyquist rolloff shaping of 0.25, the maximum baud rate is 1.6 Mbaud. In order to
transmit at a rate of 6.4 Mbps over this channel, we can map 4 bits to a symbol by means
of 16-QAM. Similarly, we can map 6 bits to a symbol through 64-QAM to support a data
rate of 9.6 Mbps.
5-11 The channel of problem 5.10 is to be used for wireless data transmission at the rate of 6.4
Mbps.
a. Estimate the maximum delay spread that may be accommodated.
From equation (5-15), the maximum delay spread that can be accommodated is given by
τ av = 1/ 2π B = 796 µsec .
b. It is desired to accommodate a delay spread of 8 µsec . OFDM is to be used. Find an

acceptable number of subcarriers, the bandwidth of each, and the resultant OFDM symbol
interval. Draw a diagram of the resultant spectrum.
The maximum bandwidth of a channel that can accommodate a delay spread of 8 µsec is
1/(2π ⋅ 8µ ) ≈ 20kHz . This corresponds to having 100 subcarriers. Assume a Nyquist
rolloff shaping of 0.25 is used. The maximum baud rate is then 1.6 Mbaud. In order to
attain a transmission rate of 6.4 Mbps, 16-QAM needs to be applied. Hence the OFDM
interval TS = 100 /1.6M = 62.5µsec .
B = 2MHz
∆f = 20kHz
f1 f2 f100 f (Hz)
c. Draw a block diagram of the OFDM transmitter of (b) if discrete Fourier Transform
processing is used for its implementation.
fc
a0 I
…
a49
6.4 Mbps 16-QAM IFFT + v(t)
…
a99 Q
(4 successive bits)
fc
Chapter 6
6-1 a. Verify the figure 832 as the capacity (number of users possible) in the analog AMPS
system. What is the corresponding figure for D-AMPS (IS-136)?
In the analog AMPS system, there are two 25 MHz bands, 824-849 MHz and 869-894
MHz, for uplink and downlink communications respectively. With each simplex channel
being 30 kHz wide, there are altogether 25M / 30k = 833 pairs of simplex channels,
supporting 833 users with duplex communication. In reality, only 832 duplex channels
are used.
D-AMPS uses the same frequency allocation plan as the analog AMPS system, however,
each band is in turn used in TDMA fashion, with repetitive frames carrying six time slots
each. Full-rate users are allocated two time slots per frame. Therefore, each frequency
channel can support three full-rate users, resulting in a system capacity of 832 × 3 =
2496.
b. Show that the European spectral allocation for GSM makes 992 channels available.
In GSM, two 25 MHz bands, 890-915 MHz uplink and 935-960 MHz downlink, are split
into 200 kHz wide bands. Since one such band in each direction is used as a guard band,
there are altogether (25M / 200k) – 1 = 124 pairs of frequency channels. With an 8-slot
TDMA frame structure and each user assigned with one time slot per frame, the system
can provide 124 × 8 = 992 duplex channels.
c. Verify the comparative capacities of the AMPS, D-AMPS, and GSM systems shown in
Table 6-2.
Table 6-2 considers a 1.25 MHz transmission band. For AMPS with a reuse factor of 7,
capacity = 1.25M / (30k × 7) ≈ 6 channels per cell. For D-AMPS, with three users
supported per TDMA frame, the capacity is tripled to 18. For GSM with a reuse factor of
4, capacity = (1.25M / 200k × 4) × 8 = 12.5.
6-2 a. Explain, in your own words, the distinction between user information bit rate and
transmission rate. In particular, show that the GSM user rate is 22.8 kbps while the
transmission rate is 270.833 kbps.
User information bit rate is the average bit rate available for each user to transmit useful
application data, like digital voice or files. Transmission rate is the raw bit rate on a
transmission channel. Since it includes both user traffic and overhead traffic for control
and signaling purposes, the transmission rate is always higher than the user information
rate.
In GSM, 148 raw data bits are transmitted in (576.92 – 30.46) = 546.46 µsec, thus the
transmission rate is 148 / 546.46µ = 270.83 kbps.
Since each user can only transmit 114 data bits per frame of (576.92µ × 8) = 4.615 msec,
and data channels only occupy 24 slots out of every 26 frames, the user rate is given by
(114 / 4.615m) × (24 / 26) = 22.8 kbps.
b. Calculate the corresponding information bit rates and transmission rates for IS-136
(D-AMPS).
For IS-136, each 40 msec frame contains 1944 bits. The transmission rate is thus 1944 /
40m = 48.6 kbps.
IS-136 adopts a 6-slot TDMA structure and in each slot carries 260 bits of user data.
Each full-rate user is assigned two time slots per frame. Therefore, the user rate is 260 ×
2 / 40m = 13 kbps.
6-3 a. Draw a diagram of a CDMA system and identify its major building blocks.
See Fig. 6-5 in the textbook.
b. What is the purpose of power control in CDMA systems?
The purpose of power control in CDMA systems is to ensure that all users in a cell have
the same level of uplink signal power received at the base station, and hence the same
SIR above a certain threshold. Since all CDMA users transmit through the same
broadband channel, their signals always interfere with each other. Without power control,
a user closer to the base station may have a higher uplink signal power, causing other
users to suffer a lower SIR and a poorer transmission link quality.
c. What is meant by soft handoff?
In soft handoff, a mobile may be in communication with two or more base stations until
the decision to connect to the new base station only is made. The same user code can be
used in any cell of a given system to which a mobile may roam. This is contrary to hard
handoff which requires the connection to the current base station be broken before
connecting to the new one.
d. Explain why CDMA offers universal reuse of channels, in which each channel can be
used in all cells.
CDMA offers universal reuse of channels because transmission orthogonality is

accomplished by assigning each user a distinct pseudorandom sequence, which is
multiplied to the information bits to convert the bit stream to a noiselike sequence of
much wider spectral width. Users each can then recover their own information bit
streams by multiplying the coded signal with their own pseudorandom sequences. Due
to the orthogonality of the codes, signals of other users using the same carrier frequency
are rejected in the decoding process. As a result, frequency channels can be reused
universally in CDMA.
6-4 Compare TDMA and CDMA schemes, identifying the advantages of each scheme over the
other.
TDMA CDMA
Voice Quality Voice quality is generally guaranteed due to Voice quality degrades as the number of
its circuit-switched nature that each active users in the system increases.
ongoing call is assigned a time slot
periodically.
Handoff Hard handoff – mobile breaks the Soft handoff – a mobile decides among two
connection with the old base station before or more base stations with which to
connecting to the new one. communicate. Advantages: (i) eliminates
“ping-pong” effect in hard handoff which
wastes network resources; (ii) smoother
user communications during handoffs.
Capacity Hard capacity – once a cell has Higher capacity and better bandwidth
accommodated the maximum number of efficiency in general. Soft capacity allows
users, new calls will be blocked and extra users to be accommodated during
existing calls may be dropped during peak period by allowing degradation in
handoffs. voice quality.
Multipath Frequency selective fading has to be Since very wideband spectrum is used,
Fading mitigated by means of channel equalization multipath echoes become so widely apart
or diversity techniques like MIMO. that they can be separately detected and
combined by a RAKE receiver to boost up
the received signal.
Security Security is not an intrinsic part of the access The coding of information bits by a private
algorithm. Extra encryption measures have pseudorandom sequence shared between
to be taken. the mobile and the base station enhances
privacy and makes cloning difficult.
Security is built-in.
Cost Lower cost in terms of base station Complicated technology results in more
equipment, space and maintenance, expensive base stations. However, fewer
especially important as cell size gets base stations are needed for the same
smaller. coverage.
Compatibility Inherently compatible with FDMA analog Incompatible with all 1G analog systems.
system; cost-effective for upgrading an
analog system to digital.
Support for The idea of frequency channel and time slot With each user sharing the whole system
Broadband assignments makes the technology bandwidth, packet-switched data
Data Service intrinsically unsuitable for supporting communication at a high bit rate is more
high-speed packet-based communication. readily supported. W-CDMA and CDMA
There are two sets of 2.5/3G TDMA 2000 are examples of 3G standards based
standards: GPRS and EDGE. on the technology.
6-5 The CDMA system IS-95 speech encoder operates at a bit rate of 9.6 kbps. Find the
spreading gain if the chip rate is 1.2288 Mchips/sec. How many chips per bit does this
represent? Superimpose a sketch of a sequence of chips on top of a sequence of bits.
Spreading gain = 1.2288 × 106 / (9.6 × 103) = 128. This represents 128 chips per bit.
The figure below shows an input binary sequence [1 0 1] spread by a pseudorandom
sequence of length 128 bits.
1 0 1
-1
1 129 257
Output Stream Sequence Number
6-6 a. Under the assumptions made in the text, show that the CDMA system IS-95, as used in
the United States, has the potential of accommodating 520 users if interference in an
isolated cell only is considered.
Assume E0/I0 = 5, information bit rate R = 10 kbps, and transmission bandwidth W =

1.25 MHz for IS-95. Then, by applying Eqn. (6-4), the number of users that can be
accommodated within this 1.25 MHz bandwidth in an isolated cell is given
by K = (1.25M /10k ) / 5 + 1 = 26 . The total number of users in the 25 MHz spectrum
allocated to cellular systems in North America would then be (25 /1.25) × 26 = 520 .
b. Show the capacity of IS-95 in the US reduces to the range of values shown in Table 6-2
if intercell interference as well as the other considerations noted in the text are taken into
account.
Assume a shadow fading model of σ = 8 dB and a propagation exponent of n = 4. For

IS-95, we choose information bit rate R = 10 kbps, and transmission bandwidth W = 1.25
MHz. From Eqn. (6-18), we have
Eb / I o = 125 /(3.38K − 1)
which has taken into account the effects of intercell interference and shadow fading. The
equation, however, has not considered the improvement in CDMA performance made
possible by silence detection, antenna sectoring and soft handoff.
According to the text, with silence detection, the resultant Eb / I o is improved by a

factor of 2.5. A further improvement of a factor of 2.4 can be achieved by using 120°
sector antennas, allowing a loss of 1 dB in antenna gain. After taking these measures into
consideration, we have
Eb / I o = 2.5 × 2.4 × 125 /(3.38 K − 1) = 750 /(3.38K − 1)
Finally, with two-cell handoff, the factor 3.38 reduces to 1.77. As a result, for a required
Eb / I o of 5, the capacity is given by
K = (750 / 5 + 1) /1.77 ≈ 85 users/cell
For a three-cell handoff, the factor 3.38 further reduces to 1.57. The capacity becomes
K = (750 / 5 + 1) /1.57 ≈ 96 users/cell
This explains the range of values for the capacity of IS-95 stated in Table 6-2.
6-7 a. Using (6-5) (or (6-7) where applicable) and (6-8) plot and compare PSK bit-error
probability as a function of Eb/I0, the ratio of signal energy per bit to interference power
spectral density, in the two cases of non-fading and fading media.
PSK Bit Error Probabilities in Fading and Non-Fading Environments

0.25
Non-Fading Environment
Fading Environment
0.2
0.15
Pe
0.1
0.05
0
0 5 10 15
Eb/I0 (dB)
PSK Bit Error Probabilities in Fading and Non-Fading Environments

0
-2
-4
-6
Pe (log)
-8
Non-Fading Environment
Fading Environment
-10
-12
-14
-16
0 5 10 15
Eb/I0 (dB)
Referring to the second plot in logarithmic scale, we see that in a non-fading

environment, the order of the PSK bit-error probability drops more rapidly as the ratio
Eb/I0 in dB increases, whereas in a fading environment the order of the bit-error
probability only drops linearly for the whole range. Therefore, the gap between the two
curves widens as the ratio Eb/I0 increases. For Eb/I0 = 10 dB, the bit-error probability of a
non-fading environment is about 4 × 10-6, which is better than the bit-error probability of
0.025 of a fading environment by nearly an order of four.
b. Using any reference on the effect of a fading environment on digital transmission (one
example appears in the text), check some of the probability of bit error entries appearing
in Table 6-1.
With reference to (6-8), for a fading environment, Eb / N 0 = 1/(4 Pe ) . For Pe = 0.01,

0.005 and 0.001, we have Eb/N0 = 14, 17 and 24 dB respectively. As it is stated in the text
that both DPSK and FSK turn out to require 3dB more signal power than the PSK case,
the corresponding values of Eb/N0 become 17, 20 and 27 dB respectively. All the
calculated values agree with the entries in Table 6-1.
6-8 This problem is devoted to establishing some of the CDMA interference expressions used in
the text.
a. Demonstrate that the total average interference power at a base station due to mobiles
outside its own cell is given by (6-13).
Assume each base station requires a received power PR which is the same throughout the
system. Then for a mobile MS1 located at a distance r1 from its base station BS1, its
controlled transmitted power is given by
PT = PR r1n 10− z1 /10
where n is the path loss exponent and z1 is the Gaussian-distributed shadow fading
random variable for the uplink path.
The transmitted power of MS1 is then received by another base station BS0 in cell S0 as
interference. This interfering power contributed by MS1 located at a distance r0 from BS0
is
r
I = PR r1n 10− z1 /10 ⋅ r0− n 10 z0 /10 = PR ( 1 ) n 10( z0 − z1 ) /10
r0
where z0 is another Gaussian shadow fading random variable for the path from MS1 to
BS0.
Assume the mobiles are uniformly distributed. Given K users per hexagonal cell of
radius R, the traffic intensity is ρ = 2 K / 3 3R 2 users/m2, and the number of mobiles in
a differential area dA(x, y) centered at point (x, y) is
2K
ρ dA( x, y ) = dA( x, y )
3 3R 2
Finally, the total average interference power at base station BS0 due to mobile outside
cell S0 is given by the integral
r 2K
I S * = E[ ∫∫ PR ( 1 ) n 10( z0 − z1 ) /10 ⋅ dA( x, y )]
0
S * r0 3 3 R 2
0
Or
2 KPR r1 ( x, y ) n
I S* = ⋅ E[10( z0 − z1 ) /10 ] ⋅ ∫∫ [ ] dA( x, y )
0
3 3R 2
S0*
r0 ( x, y )
b. Explain why two paths must be considered in evaluating the shadow fading expression
appearing in (6-13).
The two paths considered in (6-13) are: p1 – the path between the interfering mobile MS1
and its base station BS1 in cell S1, and p2 – the path between the interfering mobile MS1
and a neighboring base station BS0 at which the total average interference power is being
calculated. Shadow fading along the first path p1 affects the transmission power of MS1
due to power control in C1, whereas the shadow fading along p2 affects the interfering
power from MS1 received at BS0. Since the shadow fading conditions along the two
paths are different, their effects have to be modeled separately by two independent
random variables z0 and z1.
c. Follow the discussion in the text of the evaluation of the shadow fading expression
appearing in (6-13) and demonstrate that one obtains (6-16). Evaluate (6-16) for the
example considered in the text and show that the average outside cell interference power
is given by (6-17). Compare this result with that obtained for other values of the shadow
fading standard deviation σ.
Assume the shadow fading random variable zi is given by

zi = ah + bhi , where i = 0, 1 and a2 + b2 = 1
The random variable h represents the shadow fading in the vicinity of the transmitting
mobile and hence is common to both paths. The other random variable hi represents the
added shadow fading due to the independent propagation conditions along the two
different paths. Assume each term is equally important, then one must have a2 = b2 = 1/2.
To simplify the calculation of the expectation in (6-13), we need to define the variable
transformation
e y ≡ 10( z0 − z1 ) /10
⇒ y = ln[10( z0 − z1 ) /10 ] = ( z0 − z1 ) ⋅ (ln10) /10 = 0.23( z0 − z1 ) = 0.23b(h0 − h1 )
Since h0 and h1 are i.i.d. Gaussian random variables with zero mean and a variance of σ2,
y = 0.23b(h0 – h1) is also Gaussian with zero mean and variance
σ y2 = 0.232 b 2 ⋅ 2σ 2 = 0.0529σ 2
The shadow fading expression in (6-13) is then readily evaluated as follows:
∞ − y 2 / 2σ y2
e σ y2 / 2
∫e ⋅
( z0 − z1 ) /10
E[10 ] = E (e ) = y y
dy = e
−∞ 2πσ y2
The example in the text choose σ = 8dB. Hence, σ y2 = 0.0529 × 64 = 3.3856 and
E[10( z0 − z1 ) /10 ] = e3.3856 / 2 = 5.43 .
By numerical integration, the geometric double integral in (6-13) yields a value of 0.44
for n = 4. Combining all the results, we finally obtain the average outside cell
interference power as
I S * = PR K × 0.44 × 5.43 = 2.39 PR K
0
Now, let us consider other values of σ. For σ = 6 dB, I S * = 1.14 PR K ; for σ = 10 dB,
0
I S * = 6.2 PR K . In fact, the average outside cell interference power increases

0
exponentially with σ2. They are related by I S * = 0.44 PR K ⋅ e0.01295σ .

2
d. Use the circular-cell interferer model discussed in the text and show that the first-tier
interference power is given by (6-19b).
Rs 2π
6 KPR r1
= ∫ ∫ (r ) r1dθ dr1
n
Starting with (6-19) I1st , S *
0
π Rs2 0 0 0
Substitute r02 = (3R 2 + r12 ) − 2 3r1 R cos θ
Rs 2π
6 KPR r1n +1
We have I1st , S *
0
=
π Rs2 ∫∫
0 0 [(3R 2 + r12 ) − 2 3r1 R cos θ ]n / 2
dθ dr1
2π
dx 2π a
Choose n = 4 and using the identity ∫ = 2
0
(a + b cos x) 2
(a − b 2 )3/ 2
Rs
6 KPR 2π r15 (3R 2 + r12 )
It becomes I1st , S *
0
=
π Rs2 ∫0 [(3R 2 + r12 )2 − (2 3r1R)2 ]3/ 2 dr1
Substitute x = r1 / 3R ,
1 Rs 1 3 3
r1 = 0 → x = 0; r1 = Rs → x = ⋅ = ⋅ = 0.525
3 R 3 2π
2π x 5 (1 + x 2 ) ⋅ ( 3R )7
0.525
6 KPR
I1st , S * =
0
π Rs2 ∫
0 [(1 + x 2 ) 2 − (2 x) 2 ]3/ 2 ⋅ ( 3R)6
⋅ 3Rdx
2π x 5 (1 + x 2 ) ⋅ 3R 2
0.525
6 KPR
=
π Rs2 ∫
0
[(1 + x 2 ) 2 − (2 x) 2 ]3/ 2
dx
x 5 (1 + x 2 )
0.525
36 KPR
=
( Rs / R ) 2 ∫
0
[(1 + x 2 + 2 x )(1 + x 2 − 2 x)]3/ 2
dx
x 5 (1 + x 2 )
0.525
36 KPR
=
( Rs / R ) 2 ∫
0
[(1 + x )(1 − x)]3
dx
x 5 (1 + x 2 )
0.525
36 KPR
=
( Rs / R ) 2 ∫
0
(1 − x 2 )3
dx
Substitute u = x2, du = 2xdx

0.5252
18KPR u 2 (1 + u )
I1st , S *
0
=
( Rs / R) 2 ∫
0
(1 − u )3
du
By applying partial fraction decomposition, we have

0.5252
18KPR 4 −5 2
I1st , S *
0
=
( Rs / R) 2 ∫
0
−1 + + +
(1 − u ) (1 − u ) (1 − u )3
2
dx
0.5252
18KPR  5 1 
=  −u − 4 ln(1 − u ) − (1 − u ) + (1 − u )2 
( Rs / R) 2  0
18 KPR
= ( −3.98256 + 4) = 0.38 KPR
( Rs / R ) 2
6-9 Verify that the ratio Eb/I0 with total CDMA interference power included is given by (6-18)
for the example chosen in the text. Find the equivalent expression for a number of other
choices of the shadow fading parameter σ . (See problem 6-8c.)
From (6-18), Eb / I 0 = (W / R) /(3.38 K − 1) . Choose W = 1.25 MHz, R = 10 kbps, and K = 8,

Eb / I 0 = 4.8 , which agrees with the ratio stated in the example.
Based on the results in problem 6-8c, for σ = 6 dB, I S * = 1.14 PR K . In this case,
0
Eb / I 0 = 125 /(2.14 K − 1) . If we choose Eb / I 0 = 5 , the number of users per cell then

becomes 12.
Similarly for σ = 10 dB, I S * = 6.2 PR K and Eb / I 0 = 125 /(7.2 K − 1) . In this case, K = 3.6,
0
only around 3 to 4 users per cell can be supported.
6-10 a. The text discusses, briefly, a number of features of CDMA such as soft handoff and
automatic silence detection that improve the capacity of such systems. Which of these
features could be implemented with TDMA systems as well? Explain your conclusions.
Soft handoff is not applicable to TDMA systems because when a mobile station crosses
the cell border, unlike the case of CDMA where the same channel (i.e. spreading
sequence) can be used, it has to establish a new connection of a different frequency
channel and time slot. Hence, the original connection must be dropped first. Silence
detection is also of little value to TDMA, because fixed time slots are assigned to each
user engaging in a call. Even though the user is not speaking, communication resources
will still be occupied by the user as long as the call is ongoing. Among the three
techniques mentioned in the text, only sectoring can be implemented in TDMA. By
using directional antennas with a limited angular range of coverage at the base stations,
co-channel interference can be reduced. In this way, a smaller size of the reuse cluster
can be adopted, leading to an increase in system capacity.
b. Put the effect of these features together, as done in the text, and show how the capacity
of 84 mobiles per cell is obtained. Indicate how this number changes as the assumptions
made for the various features change as well.
Please refer to the solution of Problem 6-6b.

2919 2838 Mobile Wireless Communications - Solutions Manual CH 2-6

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2919 2838 Mobile Wireless Communications - Solutions Manual CH 2-6

Uploaded by

Copyright:

Available Formats

Mobile Wireless

Department of Electrical Engineering

 2005 All Rights Reserved

Plot of PR , db − PT ,db with respect to Distance for Orlando

Case (i) 836 MHz; winter values:

Case (ii) 836 MHz; summer values:

Case (iii) 11.2 GHz; winter values:

Case (iv) 11.2 GHz; summer values:

b. The log-normal standard deviation σ is 6 dB. An acceptable received signal is 10 mW or

Case (ii) p0 ≥ 10 mW; σ = 10 dB

Case (iii) p0 ≥ 6 mW; σ = 6 dB

Case (iv) p0 ≥ 6 mW; σ = 10 dB

With reference to the first diagram on the

With reference to the second diagram on hr

Substitute (2) into (1), we have

Substitute (4) into ∆φ ≡ 2π∆d / λ , we have ∆φ ≡ 4π ht hr / λ d .

Assume ∆φ is small and d /(d1 + d 2 ) ≈ 1

Finally, we can obtain (2-13a) by substituting the results in Steps 1 and 2:

Fig. 2-7 Received signal power as function of distance

-20 1/d4 FREE SPACE

Fig. 2-8 Effect of base station antenna height on received power

c. Do the results of Fig. 2-8 validate (2-13a)? Explain.

expectation on both sides, we have

The actual received normalized signal S R (t ) is given by (2-17):

Expanding (2-17) by trigonometry, we get

Since φk is uniformly distributed between 0 and 2π, E[cos φk ] = E[sin φk ] = 0 . Hence,

Now, consider the variances of x and y.

According to the hints, I 0 ( z ) → e z / 2π z when z 1 . This corresponds to the case

From (2-22), PR = ca 2 / 2 ⇒ a 2 = 2 PR / c (i)

Take expectation on both sides of (2-22),

Substitute (iii), (iv) and (2-25) into the above expression,

Put K ≡ A2 / 2σ R2 and cσ R2 = p /(1 + K ) , we have

σ 2 = E[( x − 0)2 ] = (b / n)2 ∑ j =1 E[( x j − 1/ 2) 2 ]

We can generate a Rayleigh-distributed random variable a from x and y by the formula

b. A different method of obtaining the Rayleigh distribution directly from a

Rewrite the expression of a, we get

Therefore, a is Rayleigh-distributed. We can obtain Rayleigh-distributed random

Relative Frequency Density of the Ricean-Distributed Random Variables Generated by Method a (n = 5)

v 100 ×103 /(60 × 60) e −1

Case (ii) ρ = 0.3 , f m = 92.59 Hz τ f = = 1.35m sec.

In a wireless medium, due to reflection, scattering, diffraction and refraction, a signal

Under a minimum mean-squared performance objective, we want to choose the tap

To obtain the optimum choice of tap gains, we find the solution of

on the simpler looking form

Multiply R-1 to both sides, we have

Choose the training sequence {sj} = {1 1 -1 1 -1 -1 -1 1 1 -1}

From equations (2-54), (2-55) and (2-57), we find that

{rj} = {1 -1 1 1 1 1 -1 -1 1 1} (third and fifth bits received in errors)

In the following two examples, we use transversal filter equalizers of 11 taps (N = 5)

A comparison of the four examples of traversal filter equalizers is summarized in the

According to Schwartz’s inequality for complex numbers,

Choose ck = ak* / nk′ and d k = g k nk′ , we have

∑a / nk′ , and the SIR on the kth diversity branch is just

Since the SIR on each branch ak / nk′

Therefore, it can be concluded that the performance of maximal-ratio combining is the

A RAKE receiver boosts up the signal reception performance by combining

3-1 a. Verify the entries in Table 3-1.

Including second-tier interferers, the expression for SIR becomes

b. Referring to Fig. 3-6, fill in the details of the derivation of (3-4).

3-3 a. Verify the entries of Table 3-3.

C D/R SIR SIRdB

Stuber’s SIR approximation:

Rappaport’s SIR approximation: