Goldsmith Solution

Chapter 1
1. In case of an accident, there is a high chance of getting lost. The transportation cost is very high each
time. However, if the infrastructure is set once, it will be very easy to use it repeatedly. Time for
wireless transmission is negligible as signals travel at the speed of light.
2. Advantages of bursty data communication
(a) Pulses are made very narrow, so multipaths are resolvable
(b) The transmission device needs to be switched on for less time.
Disadvantages
(a) Bandwidth required is very high
(b) Peak transmit power can be very high.
3. Pb = 1012
1
12
2 = 10
=
1012
2
= 5 1011 (very high)

3
4. Geo: 35,786 Km above earth RT T = 23578610

= 0.2386s
c
28000103
Meo: 8,000- 20,000 Km above earth RT T =
= 0.0533s
c
2500103
Leo: 500- 2,000 Km above earth RT T =
= 0.0033s
c
Only Leo satellites as delay = 3.3ms < 30ms
5.
6. optimum no. of data user = d
optimum no. of voice user = v
Three different cases:
Case 1: d=0, v=6
revenue = 60.80.2 = 0.96
Case 2: d=1, v=3
revenue = [prob. of having one data user](revenue of having one data user)
+ [prob. of having two data user](revenue of having two data user)
+ [prob. of having one voice user](revenue of having one voice user)
+ [prob. of having two voice user](revenue of having two voice user)
+ [prob. of having three or more voice user](revenue in this case)

2
6
6
2
2
5
0.5
$1 + 0.5 $1 +
0.8 0.2 $0.2 +
0.82 0.24 $0.4+
1
1
2
6
1
0.8 0.2 $0.2
6
2
0.8 0.2 $0.4 $0.6

2
$1.35
Case 3: d=2, v=0
revenue =2 0.5 = $1
So the best case is case 2, which is to allocate 60kHz to data and 60kHz to voice.
7.
8. 1. Hand-off becomes a big problem.
2. Inter-cell interference is very high and should be mitigated to get reasonable SINR.
3. Infrastructure cost is another problem.
9. Smaller the reuse distance, larger the number of users who can use the same system resource and so
capacity (data rate per unit bandwidth) increases.
10. (a) 100 cells, 100 users/cell 10,000 users
(b) 100 users/cell 2500 cells required
100km2
Area
2
Area/cell = 2500cells cell = .04km
(c) From Rappaport or iteration of formula, we get that 100
1
of an Erlang of traffic.
Each subscriber generates 30
Thus, each cell can support 30 89 = 2670 subscribers
Macrocell: 2670 100 267, 000 subscribers
Microcell: 6,675,000 subscribers
channels
cell
89
channels
cell
(d) Macrocell: $50 M

Microcell: $1.25 B
(e) Macrocell: $13.35 M/month 3.75 months approx 4 months to recoop
Microcell: $333.75 M/month 3.75 months approx 4 months to recoop
11. One CDPD line : 19.2Kbps
average Wimax 40M bps
number of CDPD lines 2 103
@Pb = .02
Chapter 2
1.
Pr
103
103
2
Gl
= Pt
= c/fc = 0.06
4d
2
= Pt
Pt = 4.39KW
410
= Pt
Pt = 438.65KW
4100
Attenuation is very high for high frequencies

2. d= 100m
ht = 10m
hr = 2m
delay spread = =
3. =
x+x0 l
c
= 1.33
2(x0 +xl)
p
p
(ht + hr )2 + d2 (ht hr )2 + d2
s
s
2
2
h
+
h
h
h
t
r
t
r
= d
+1
+ 1
d
d
x0 + x l =
d ht , hr , we need to keep only first order terms
s
s
1 h + h 2
1
h
h
t
r
t
r
d
+ 1
+ 1
2
d
2
d
=
2(ht + hr )
d
2 2(ht + hr )
4. Signal nulls occur when = (2n + 1)

2(x0 + x l)
= (2n + 1)
h
i
p
2 p
(ht + hr )2 + d2 (ht hr )2 + d2 = (2n + 1)
p
p
(ht + hr )2 + d2 (ht hr )2 + d2 =
(2n + 1)
2
Let m = (2n + 1)
p
p
(ht + hr )2 + d2 = m + (ht hr )2 + d2
2
square both sides
p
2
(ht + hr )2 + d2 = m2
+ (ht hr )2 + d2 + m (ht hr )2 + d2
4
2
2
x = (ht + hr ) , y = (ht hr ) , x y = 4ht hr
p
2
x = m2
+ y + m y + d2
s 4
2
1
2
2
d =
xm
y
y
m
4
s
4ht hr
(2n + 1) 2
d =
(ht hr )2 , n Z
(2n + 1)
4
5. ht = 20m
hr = 3m
fc = 2GHz = fcc = 0.15
dc = 4ht hr = 1600m = 1.6Km
This is a good radius for suburban cell radius as user density is low so cells can be kept fairly large.
Also, shadowing is less due to fewer obstacles.
6. Think of the building as a plane in R3
The length of the normal to the building from the top of Tx antenna = ht
The length of the normal to the building from the top of Rx antenna = hr
In this situation the 2 ray model is same as that analyzed in the book.
7. h(t) = 1 (t ) + 2 (t ( + 0.22s))
Gr = Gl = 1
ht = hr = 8m
fc = 900M Hz, = c/fc = 1/3
R = 1
x + x0 l
= 0.022 106 s
delay spread =
c
q
2
2 82 + d2 d
= 0.022 106 s
c
d = 16.1m
d
= = 53.67ns
c
2
Gl
1 =
= 2.71 106
4 l

2
RGr
2 =
= 1.37 106
4 x + x0
8. A program to plot the figures is shown below. The power versus distance curves and a plot of the
phase difference between the two paths is shown on the following page. From the plots it can be seen
that as Gr (gain of reflected path) is decreased, the asymptotic behavior of Pr tends toward d2 from
d4 , which makes sense since the effect of reflected path is reduced and it is more like having only a
LOS path. Also the variation of power before and around dc is reduced because the strength of the
reflected path decreases as Gr decreases. Also note that the the received power actually increases with
distance up to some point. This is because for very small distances (i.e. d = 1), the reflected path is
approximately two times the LOS path, making the phase difference very small. Since R = -1, this
causes the two paths to nearly cancel each other out. When the phase difference becomes 180 degrees,
the first local maxima is achieved. Additionally, the lengths of both paths are initially dominated by
the difference between the antenna heights (which is 35 meters). Thus, the powers of both paths are
roughly constant for small values of d, and the dominant factor is the phase difference between the
paths.
clear all;
close all;
ht=50;
hr=15;
f=900e6;
c=3e8;
lambda=c/f;
GR=[1,.316,.1,.01];
Gl=1;
R=-1;
counter=1;
figure(1);
d=[1:1:100000];
l=(d.^2+(ht-hr)^2).^.5;
r=(d.^2+(ht+hr)^2).^.5;
phd=2*pi/lambda*(r-1);
dc=4*ht*hr/lambda;
dnew=[dc:1:100000];
for counter = 1:1:4,
Gr=GR(counter);
Vec=Gl./l+R*Gr./r.*exp(phd*sqrt(-1));
Pr=(lambda/4/pi)^2*(abs(Vec)).^2;
subplot(2,2,counter);
plot(10*log10(d),10*log10(Pr)-10*log10(Pr(1)));
hold on;
plot(10*log10(dnew),-20*log10(dnew));
plot(10*log10(dnew),-40*log10(dnew));
end
hold off
Gr = 1
Gr = .316
50
10 * log10(Pr)
10 * log10(Pr)
50
0
50
100
150
20
40
Gr = .1
10 * log10(d)
20
40
G = .01
10 * rlog10(d)
60
20
40
10 * log10(d)
60
50
10 * log10(Pr)
10 * log10(Pr)
100
150
60
50
0
50
100
150
0
50
20
40
10 * log10(d)
60
20
40
10 * log10(d)
60
0
50
100
150
Phase (deg)
400
300
200
100
0
Figure 1: Problem 8
9. As indicated in the text, the power fall off with distance for the 10-ray model is d2 for relatively large
distances
p
10. The delay spread is dictated by the ray reaching last d = (500/6)2 + 102 = 83.93m
Total distance = 6d = 503.59m
0 = 503.59/c = 1.68s
L.O.S ray d = 500m
0 = 500/c = 1.67s
delay spread = 0.01s
11. fc = 900M Hz
= 1/3m
2
G = 1 radar cross
p section 20dBm = 10
log1 0 = 100
0
2
2
d=1 , s = s = (0.5d) + (0.5d) = d 0.5 = 0.5
Path loss due to scattering
"
#2
G
Pr
=
= 0.0224 = 16.498dB
Pt
(4)3/2 ss0
Path loss due to reflection (using 2 ray model)
!2
R G
2
Pr
=
= 3.52 104 = 34.54dB
Pt
s + s0
4
d = 10
Pscattering = 56.5dB
d = 100 Pscattering = 96.5dB
d = 1000 Pscattering = 136.5dB
Notice that scattered rays over long
12.
Pref lection = 54.54dB

distances result in tremendous path loss
d0
simplified
d
2 2
Gl
Pr = Pt
free space
4
d
Pr = Pt K
2
Gl
when K = 4
and d0 =
The two models are equal.
13. Pnoise = 160dBm
fc = 1GHz, d0 = 1m, K = (/4d0 )2 = 5.7 104 , = 0.3, = 4
We want SN Rrecd = 20dB = 100
Noise power is 1019

d0
P = Pt K
d
4
0.3
1017 = 10K
d
d 260.7m
14. d = distance between cells with reused freq
p = transmit power of all the mobiles

S
20dB
I uplink
(a) Min. S/I will result when main user is at A and
Interferers are at B
dA = distance
between A and base station #1 = 2km dB = distance between B and base station
#1 = 2km
h
i2
G

P
4dA
d2B
S
(dmin 1)2
=
= 100
h
i2 = 2 =
I min
4
2dA
G
2P 4dB
dmin 1 = 20km dmin = 21km since integer number of cells should be accommodated in
distance d dmin = 22km
(b)
P
d0
=k
Pu
d
h i
d0

P
k
dA
S
h i =
=
I min 2P k d0
dB
1 dB
1 dmin 1
1 dmin 1 3
=
=
= 100
2 dA
2
2
2
2
dmin = 9.27 with the same argument dmin = 10km
(c)
h i
d0

k
dA A
(d
1)4
S
h i = min
=
= 100
I min 2k d0
0.04
dB
B
dmin = 2.41km with the same argument dmin = 4km

15. fc = 900M Hz, ht = 20m, hr = 5m, d = 100m
Large urban city
small urban city
suburb
rural area
P Llargecity = 353.52dB
P Lsmallcity = 325.99dB
P Lsuburb = 207.8769dB
P Lruralarea/countryside = 70.9278dB
As seen , path loss is higher in the presence of multiple reflectors, diffractors and scatterers
Plot for Gr = 0 dB
0
10
y = 15x+8
20
30
40
y = 35x+56
50
60
1.2
1.4
1.6
1.8
2.2
2.4
2.6
2.8
Figure 2: Problem 16
16. Piecewise linear model for 2-path model. See Fig 2
P
P
17. Pr = Pt PL (d) 3i F AFi 2j P AFj
FAF =(5,10,6), PAF =(3.4,3.4)

d0
= 108 = 8dB
PL (d)K
d 0
110 = Pt 80 5 10 6 3.4 3.4
Pt = 2.2dBm
18. (a)
Pr
Pt dB
= 10 log10 K 10r log1 0 dd0

using least squares we get
10 log10 K = 29.42dB
=4
(b) PL(2Km) = 10 log10 K 10r log10 d = 161.76dB

2
(c) Receiver power can be assumed to be Gaussian with variance dB
2
X N (0, dB
)
P rob(X < 10) = P rob

19. Assume free space path loss parameters
fc = 900M Hz = 1/3m
dB = 6
SN Rrecd = 15dB
Pt = 1W
X
dB
10
<
dB
= 6.512 104
g = 3dB
Pnoise = 40dBm Precvd = 55dB
Suppose we choose a cell of radius d
dB
(d) = Precvd (due to path loss alone)
2
Gl
1.4 13
= Pt
=
4d
d2
= 10 log1 0((d))
P (Precd (d) > 55) = 0.9
Precd (d) dB
55 dB
>
= 0.9
dB
6
55 dB
= 1.282
6
dB = 47.308
(d) = 1.86 105
d = 8.68m
20. MATLAB CODE

Xc = 20;
ss = .01;
y = wgn(1,200*(1/ss));
for i = 1:length(y)
x(i) = y(i);
for j = 1:i
x(i) = x(i)+exp(-(i-j)/Xc)*y(j);
end
end
21. Outage Prob. = Prob. [received powerdB 6 T pdB ]
Tp = 10dB
(a)
outageprob. = 1 Q
T p
=1Q
5
8

5
=Q
= 26%
8
(b) = 4dB, outage prob < 1%
T p
T p
Q
> 99%
< 2.33
19.32dB
(c)
= 12dB,
T p
< 6.99 37.8dB
20
White Noise Process
Filtered Shadowing Processing
10
10
20
30
40
50
60
20
40
60
80
100
120
140
160
180
200
(d) For mitigating the effect of shadowing, we can use macroscopic diversity. The idea in macroscopic
diversity is to send the message from different base stations to achieve uncorrelated shadowing.
In this way the probability of power outage will be less because both base stations are unlikely to
experience an outage at the same time, if they are uncorrelated.
22.
2
C= 2
R
ZR
r=0
r
rQ a + b ln
dr
R
To perform integration by parts, we let du = rdr and v = Q a + b ln Rr . Then u = 12 r2 and

dv =

r
1
b
Q a + b ln
=
Q(x)|x=a+b ln(r/R)
a + b ln
= exp(k 2 /2) dr.
r
R
x
r
R
r
2
(1)
where k = a + b ln Rr . Then we get

C =
ZR
2 1 2
r R
2
1 2 1
b
r Q a + b ln
+ 2
r exp(k 2 /2) dr
2
R 2
R r=0 R
2
r
2
(2)
r=0
= Q(a) +
= Q(a) +
1
R2
1
R2
ZR
r=0
ZR
r=0
Za
1
b
r2 exp(k 2 /2) dr
r
2
1
R2 exp
2
1
exp
2
= Q(a) +
k=
2(k a)
b
(3)
b
exp(k 2 /2) dr
r
k 2 2k 2a
+
2
b
b
dk
(5)
Za
1
1
2 2
exp (k ) dk
= Q(a) + exp
2
b
2
k=
!#
"
a 2b
2 2ab
= Q(a) + exp
1Q
b2
1

2 2ab
2 ab
= Q(a) + exp
Q
2
b
b
2a
2
+ 2
b
b
(4)
(6)
(7)
(8)
(9)
Since Q(x) = 1 Q(x).

23. = 3
d0 = 1
k = 0dB
= 4dB
R = 100m
Pt = 80mW Pmin = 100dBm = 130dB
P (R) = Pt K
d0
d
= 80 109 = 70.97dB
Pmin P (R)
= 14.7575
10 log10 e
b=
= 3.2572
dB
22ab
2 2ab
2
'1
c = Q(a) + e b Q
b
a=
24. = 6
=8
P (R) = 20 + Pmin
a = 20/8 = 2.5
10 6 log10 e
= 20.3871
b=
8
c = 0.9938
2
0.7728
0.6786
0.6302
4
0.8587
0.7728
0.7170
6
0.8977
0.8255
0.7728

Since Pr (r) Pmin for all r R, the probability of non-outage is proportional to Q 1
, and thus
decreases as a function of . Therefore, C decreases as a function of . Since the average power at the
boundary of the cell is fixed, C increases with , because it forces higher transmit power, hence more
received power at r < R. Due to these forces, we have minimum coverage when = 2 and = 12. By
a similar argument, we have maximum coverage when = 6 and = 4. The same can also be seen
from this figure:
0.95
0.9
0.85
Coverage
25.
/dB
4
8
12
0.8
0.75
0.7
0.65
0.6
12
6
10
5
8
4
6
3
4
dB
The value of coverage for middle point of typical values i.e. = 4 and = 8 can be seen from the
table or the figure to be 0.7728.
Chapter 3
1. d = vt
2
r + r0 = d + 2hd
Equivalent low-pass channel impulse response is given by
c(, t) = 0 (t)ej0 (t) ( 0 (t)) + 1 (t)ej1 (t) ( 1 (t))
Gl
0 (t) = 4d
with d = vt
0 (t) = 2fc 0 (t) D0
0 (t) =Rd/c
D0 = t 2fD0 (t)dt
fD0 (t) = v cos 0 (t)
0 (t) = 0 t
R Gl
R Gl
1 (t) = 4(r+r
with d = vt
0) =
2h2
4(d+
1 (t) = 2fc 1 (t) D1

2
1 (t) =R(r + r0 )/c = (d + 2hd )/c
D1 = t 2fD1 (t)dt
fD1 (t) = v cos 1 (t)
h
1 (t) = arctan d/2
t
2. For the 2 ray model:
l
c
x + x0
1 =
c
p
p
(ht + hr )2 + d2 (ht hr )2 + d2
x + x0 l
delay spread(Tm ) =
=
c
c
when d (ht + hr )
1 2ht hr
Tm =
c d
ht = 10m, hr = 4m, d = 100m
0 =
Tm = 2.67 109 s
3. Delay for LOS component = 0 = 23 ns
Delay for First Multipath component = 1 = 48 ns
Delay for Second Multipath component = 2 = 67 ns
c = Delay for the multipath component to which the demodulator synchronizes.
Tm = max m c
m
So, when c = 0 , Tm = 44 ns. When c = 1 , Tm = 19 ns.
4. fc = 109 Hz
10
n,min = 310
8s
min fc n =
1010
3108
= 33 1
5. Use CDF strategy.

Z
Z2 Zz
(x2 +y)2
1
2 2
e
dxdy =
2 2
Fz (z)= P [x2 +y 2 z 2 ] =
x2 +y 2 z 2
z 2
r2
1
2 2 (z 0)
e
rdrd
=
1
e
2 2 2 2
df z (z) z z22
= 2 e 2 Rayleigh
dz
For Power:
z 2
z] = 1 e 2
2
1 z
fz (z)= 2 e 2 Exponential
2 2
Fz 2 (z)=P [Z
6. For Rayleigh fading channel

P r(Pr < P0 ) = 1 eP0 /2
2 2 = 80dBm, P0 = 95dBm, P r(Pr < P0 ) = 0.0311

2 2 = 80dBm, P0 = 90dBm, P r(Pr < P0 ) = 0.0952
7. For Rayleigh fading channel
Poutage = 1 eP0 /2
0.01 = 1 eP0 /Pr

Pr = 60dBm
8. 2 2 = -80dBm = 1011
Target Power P0 = -80 dBm = 1011
Avg. power in LOS component = s2 = -80dBm = 1011
105
P r[z 2 1011 ] = P r[z ]
10
Let z0 =
5
10
10
Z
=
0
z0
z
e
2
(z 2 +s2 )
2 2
I0
zs
2
dz,
z0
= 0.3457
To evaluate this, we use Matlab and I0 (x) = besseli(0,x). Sample Code is given:
clear P0 = 1e-11; s2 = 1e-11; sigma2 = (1e-11)/2; z0 =
sqrt(1e-11); ss = z0/1e7; z = [0:ss:z0]; pdf =
(z/sigma2).*exp(-(z.^2+s2)/(2*sigma2)).*besseli(0,z.*(sqrt(s2)/sigma2));
int_pr = sum(pdf)*ss;
9. CDF of Ricean distribution is
Z
FZRicean (z)
where
=
0
pRicean
(z)
Z
r
2
2z(K
+
1)
(K
+
1)z
K(K
+
1)
pRicean
exp K
I0 2z
,
(z) =
Z
Pr
Pr
Pr
z0
For the Nakagami-m approximation to Ricean distribution, we set the Nakagami m parameter to be
(K + 1)2 /(2K + 1). CDF of Nakagami-m distribution is
Z z
Nakagami-m
FZ
(z) =
pNakagami-m
(z)
Z
0
where
pNakagami-m
(z)
Z
2mm z 2m1
mz 2
=
exp
,
(m)P rm
Pr
z 0,
m 0.5
We need to plot the two CDF curves for K = 1,5,10 and P r =1 (we can choose any value for Pr as it
is the same for both the distributions and our aim is to compare them). Sample code is given:
z = [0:0.01:3]; K = 10; m = (K+1)^2/(2*K+1); Pr = 1; pdfR =
((2*z*(K+1))/Pr).*exp(-K-((K+1).*(z.^2))/Pr).*besseli(0,(2*sqrt((K*(K+1))/Pr))*z);
pdfN = ((2*m^m*z.^(2*m-1))/(gamma(m)*Pr^m)).*exp(-(m/Pr)*z.^2);
for i = 1:length(z)
cdfR(i) = 0.01*sum(pdfR(1:i));
cdfN(i) = 0.01*sum(pdfN(1:i));
end plot(z,cdfR); hold on plot(z,cdfN,b--); figure;
plot(z,pdfR); hold on plot(z,pdfN,b--);
As seen from the curves, the Nakagami-m approximation becomes better as K increases. Also, for a
fixed value of K and x, prob(<x) for x large is always greater for the Ricean distribution. This is seen
from the tail behavior of the two distributions in their pdf, where the tail of Nakagami-distribution is
always above the Ricean distribution.
10. (a) W = average received power
Zi = Shadowing over link i
Pr i = Received power in dBW, which is Gaussian with mean W, variance 2
(b)
Poutage = P [Pr,1 < T Pr,2 < T ] = P [Pr,1 < T ] P [Pr,2 < T ] since Z1 , Z2 independent
2 2

4
W T
= Q
= Q
(c)
Z
Pout
P [Pr,1 T , Pr,2 < T |Y = y] fy (y) dy
Pr,1 |Y = y ~ N W + by,a2 2 , and [Pr,1 |Y = y] [Pr,2 |Y = y]
Poutage=
2
Z
y2
W + by T
1
Q
e 22 dy
a
2
K=5
1
K = 10
1
Ricean
Nakagamim
Ricean
Nakagamim
0.9
0.9
0.9
0.8
0.8
0.8
0.7
0.7
0.7
0.6
0.6
0.5
0.5
0.5
0.4
0.4
0.4
0.3
0.3
0.3
0.2
0.2
0.2
0.1
0.1
0.1
Ricean
Nakagami m
0.6
0.5
1.5
2.5
0.5
K=1
1.5
2.5
0.5
K=5
1.5
2.5
K = 10
1.5
2
Ricean
Nakagamim
0.9
1.8
Ricean
Nakagamim
0.8
Ricean
Nakagami m
0.7
1.6
1.4
1
0.6
1.2
0.5
0.4
0.8
0.5
0.3
0.6
0.2
0.4
0.1
0.2
0.5
1.5
31
2.5
0.5
1.5
2.5
0.5
K = 10
x 10
1.8
1.6
1.4
Ricean
Nakagamim
1.2
0.8
0.6
0.4
0.2
0
3.5
4.5
5.5
Tail Behavior
Figure 1: The CDF and PDF for K = 1, 5, 10 and the Tail Behavior
1.5
2.5
let
y
=
u
Z
=
(d) Let a = b =
Z
1
W T + bu 2 u2
1
4 + by 2 y 2
Q
e 2 du =
Q
e
dy
a
a
2
2
2
1
2
, = 8, 4 = 5. With independent fading we get

2
5
Pout = Q
= 0.0708.
8
With correlated fading we get Pout = 0.1316.

Conclusion : Independent shadowing is prefarable for diversity.
11. There are many acceptable techniques for this problem. Sample code for both the stochastic technique(preferred) and the Jakes technique are included.
Jakes: Summing of appropriate sine waves
%Jakes Method
close all; clear all;
%choose N=30
N=30; M=0.5*(N/2-1); Wn(M)=0; beta(M)=0;
%We choose 1000 samples/sec
ritemp(M,2001)=0; rqtemp(M,2001)=0; rialpha(1,2001)=0; fm=[1 10
100]; Wm=2*pi*fm; for i=1:3
for n=1:1:M
for t=0:0.001:2
%Wn(i)=Wm*cos(2*pi*i/N)
Wn(n)=Wm(i)*cos(2*pi*n/N);
%beta(i)=pi*i/M
beta(n)=pi*n/M;
%ritemp(i,20001)=2*cos(beta(i))*cos(Wn(i)*t)
%rqtemp(i,20001)=2*sin(beta(i))*cos(Wn(i)*t)
ritemp(n,1000*t+1)=2*cos(beta(n))*cos(Wn(n)*t);
rqtemp(n,1000*t+1)=2*sin(beta(n))*cos(Wn(n)*t);
%Because we choose alpha=0,we get sin(alpha)=0 and cos(alpha)=1
%rialpha=(cos(Wm*t)/sqrt(2))*2*cos(alpha)=2*cos(Wm*t)/sqrt(2)
%rqalpha=(cos(Wm*t)/sqrt(2))*2*sin(alpha)=0
rialpha(1,1000*t+1)=2*cos(Wm(i)*t)/sqrt(2);
end
end
%summarize ritemp(i) and rialpha
ri=sum(ritemp)+rialpha;
%summarize rqtemp(i)
rq=sum(rqtemp);
%r=sqrt(ri^2+rq^2)
r=sqrt(ri.^2+rq.^2);
%find the envelope average
mean=sum(r)/2001;
subplot(3,1,i);
time=0:0.001:2;
%plot the figure and shift the envelope average to 0dB
plot(time,(10*log10(r)-10*log10(mean)));
titlename=[fd = int2str(fm(i)) Hz];
title(titlename);
xlabel(time(second));
ylabel(Envelope(dB));
end
fd = 1 Hz
Envelope(dB)
5
0
5
10
0.2
0.4
0.6
0.8
1
fd = 10 Hz
1.2
1.4
1.6
1.8
0.2
0.4
0.6
0.8
1
fd = 100 Hz
1.2
1.4
1.6
1.8
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
Envelope(dB)
10
0
10
20
Envelope(dB)
10
0
10
20
Stochastic: Usually two guassian R.V.s are filtered by the Doppler Spectrum and summed. Can also
just do a Rayleigh distribution with an adequate LPF, although the above technique is prefered.
function [Ts, z_dB] = rayleigh_fading(f_D, t, f_s)
%
% function [Ts, z_dB] = rayleigh_fading(f_D, t, f_s)
% generates a Rayleigh fading signal for given Doppler frequency f_D,
% during the time perios [0, t], with sampling frequency f_s >= 1000Hz.
%
% Input(s)
%
-- f_D : [Hz] [1x1 double] Doppler frequency
%
-- t
: simulation time interval length, time interval [0,t]
%
-- f_s : [Hz] sampling frequency, set to 1000 if smaller.
% Output(s)
%
-- Ts
: [Sec][1xN double] time instances for the Rayleigh signal
%
-- z_dB : [dB] [1xN double] Rayleigh fading signal
%
% Required parameters
if f_s < 1000;
f_s = 1000;
end;
% [Hz]
Minumum required sampling rate
N = ceil( t*f_s );
% Number of samples
% Ts contains the time instances at which z_dB is specified

Ts = linspace(0,t,N);
if mod( N, 2) == 1
N = N+1;
end;
f = linspace(-f_s,f_s,N);
% Use even number of samples in calculation

% [Hz]
Frequency samples used in calculation
% Generate complex Gaussian samples with line spectra in frequency domain

% Inphase :
Gfi_p = randn(2,N/2); CGfi_p = Gfi_p(1,:)+i*Gfi_p(2,:); CGfi = [
conj(fliplr(CGfi_p)) CGfi_p ];
% Quadrature :
Gfq_p = randn(2,N/2); CGfq_p = Gfq_p(1,:)+i*Gfq_p(2,:); CGfq = [
conj(fliplr(CGfq_p)) CGfq_p ];
% Generate fading spectrum, this is used to shape the Gaussian line spectra
omega_p = 1;
% this makes sure that the average received envelop can be 0dB
S_r = omega_p/4/pi./(f_D*sqrt(1-(f/f_D).^2));
% Take care of samples outside the Doppler frequency range, let them be 0
idx1 = find(f>=f_D); idx2 = find(f<=-f_D); S_r(idx1) = 0;
S_r(idx2) = 0; S_r(idx1(1)) = S_r(idx1(1)-1);
S_r(idx2(length(idx2))) = S_r(idx2(length(idx2))+1);
% Generate r_I(t) and r_Q(t) using inverse FFT:
r_I = N*ifft(CGfi.*sqrt(S_r)); r_Q = -i*N*ifft(CGfq.*sqrt(S_r));
% Finally, generate the Rayleigh distributed signal envelope
z = sqrt(abs(r_I).^2+abs(r_Q).^2); z_dB = 20*log10(z);
% Return correct number of points
z_dB = z_dB(1:length(Ts));
1 Hz
0
10
20
30
0.2
0.4
0.6
0.8
1
10Hz
1.2
1.4
1.6
1.8
0.2
0.4
0.6
0.8
1
l00 Hz
1.2
1.4
1.6
1.8
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
10
0
10
20
30
20
0
20
40
60
12. Pr = 30dBm
fD = 10Hz
P0 = 0dBm, tz =
e 1
= 0.0013s
fD 2
P0 = 15dBm, tz = 0.0072s
P0 = 25dBm, tz = 0.0264s
13. In the reader, we found the level crossing rate below a level by taking an average of the number of
times the level was crossed over a large period of time. It is easy to convince that the level crossing
rate above a level will have the same expression as eq. 3.44 in reader because to go below a level again,
we first need to go above it. There will be some discrepancy at the end points, but for a large enough
T it will not effect the result. So we have
2
LZ (above) = LZ (below) = 2fD e
And,
tZ (above) =
p(z > Z)
LZ (above)
2
p(z > Z) = 1 p(z Z) = 1 (1 e ) = e

1
tZ (above) =
2fD
The values of tZ (above) for fD = 10,50,80 Hz are 0.0224s, 0.0045s, 0.0028s respectively. Notice that as
fD increases, tZ (above) reduces.
14. Note: The problem has been solved for Ts = 10s
Pr = 10dB
fD = 80Hz
R1
R2
R3
R4
R5
R6
R7
R8
: 10dB,
: 10dB 0dB,
:
0dB 5dB,
:
5dB 10dB,
: 10dB 15dB,
: 15dB 20dB,
: 20dB 30dB,
:
30dB ,
1
2
3
4
5
6
7
8
= 9.95 103
= 0.085
= 0.176
= 0.361
= 0.325
= 0.042
= 4.54 105
= 3.72 1044
Nj level crossing rate at level Aj

q
N1 = 0,
N2 = 19.85, =
N3 = 57.38, =
N4 = 82.19, =
N5 = 73.77, =
N6 = 15.09, =
N7 = 0.03,
N8 = 0,
q 10
0.1
q 10
1
q 10
100.5
q 10
10
q 10
101.5
q 10
102
q 10
103
10
MATLAB CODE:
N = [0 19.85 57.38 82.19 73.77 15.09 .03 0];
Pi = [9.95e-3 .085 .176 .361 .325 .042 4.54e-5 3.72e-44];
T = 10e-3;
for i = 1:8
if i == 1
p(i,1) = 0;
p(i,2) = (N(i+1)*T)/Pi(i);
p(i,3) = 1-(p(i,1)+p(i,2));
elseif i == 8
p(i,1) = (N(i)*T)/Pi(i);
p(i,2) = 0;
p(i,3) = 1-(p(i,1)+p(i,2));
else
p(i,1) = (N(i)*T)/Pi(i);
p(i,2) = (N(i+1)*T)/Pi(i);
p(i,3) = 1-(p(i,1)+p(i,2));
end
end
% p =
%
%
0
%
0.0023
%
0.0033
%
0.0023
%
0.0023
%
0.0036
%
0.0066
%
0
0.0199
0.0068
0.0047
0.0020
0.0005
0.0000
0
0
15. (a)
0.9801
0.9909
0.9921
0.9957
0.9973
0.9964
0.9934
1.0000
= 70Hz
1 ( )
2 ( 0.022sec) = 49.5Hz
S(, ) =
0
else
The antenna setup is shown in Fig. 15

From the figure, the distance travelled by the LOS ray is d and the distance travelled by the first
multipath component is
s
d 2
2
+ 64
2
Given this setup, we can plot the arrival of the LOS ray and the multipath ray that bounces off
the ground on a time axis as shown in Fig. 15
So we have
s
d 2
+ 82 d = 0.022 106 3 108
2
2
2
d
4
+ 82 = 6.62 + d2 + 2d(6.6)
4
d = 16.1m
fD = v cos()/. v = fD / cos(). For the LOS ray, = 0 and for the multipath component,
= 45o . We can use either of these rays and the corresponding fD value to get v = 23.33m/s.
(b)
dc =
4ht hr
dc = 768 m. Since d dc , power fall-off is proportional to d2 .

(c) Tm = 0.022s, B 1 = 0.33s. Since Tm B 1 , we have flat fading.
16. (a) Outdoor, since delay spread 10 sec.
Consider that 10 sec d = ct = 3km difference between length of first and last path
(b) Scattering function
S(, ) = Ft [Ac (, t)]
1
1
= W
rect W
for 0 10sec
The Scattering function is plotted in Fig. 16
8m
8m
d meters
Figure 4: Antenna Setting

0.022 us
t0
t1
t0 = (d/3e8)
t1 = 2 sqrt[(d/2)^2+8^2]/3e8
Figure 5: Time Axis for Ray Arrival
(c) Avg Delay Spread =
Ac ( )d
= 5sec
Ac ( )d
v
u
u R ( T m )2 Ac ( )d
u
= 2.89sec
RMS Delay Spread = t 0
R
0
Ac ( )d
Doppler Spread =
W
2
= 50 Hz
(d) u > Coherence BW Freq. Selective Fading

Can also use T m or T m instead of Tm
1
Tm
= 105 u > = 105 kHz
(e) Rayleigh fading, since receiver power is evenly distributed relative to delay; no dominant LOS
path
(f) tR =
e
1
FD 2
with = 1, fD =
W
2
tr = .0137 sec
(g) Notice that the fade duration never becomes more than twice the average. So, if we choose our
data rate such that a single symbol spans the average fade duration, in the worst case two symbols
will span the fade duration. So our code can correct for the lost symbols and we will have error-free
transmission. So t1R = 72.94 symbols/sec
17. (a) Tm .1msec = 100sec
Bd .1Hz
Answers based on Tm or Tm are fine too. Notice, that based on the choice of either Tm , Tm or
Tm , the remaining answers will be different too.
(b) Bc T1m = 10kHz
f > 10kHz for u1 u2
(c) (t)c = 10s
(d) 3kHz < Bc Flat
30kHz > Bc Freq. Selective
S(tau,rho)
-W/2
W/2
rho
10 us
tau
Figure 6: Scattering Function
Chapter 4
1. C = B log2 1 +
C=

log2 1+ NSB
1
B
S
N0 B
As B by LHospitals rule
C=
S 1
N0 ln 2
2. B = 50 MHz
P = 10 mW
N0 = 2 109 W/Hz
N = N0 B
C = 6.87 Mbps.
Pnew = 20 mW, C = 13.15 Mbps (for x 1, log(1 + x) x)
B = 100 MHz, Notice that both the bandwidth and noise power will increase. So C = 7 Mbps.
3. Pnoise = 0.1mW
B = 20M Hz
(a) Cuser1base station = 0.933B = 18.66M bps
(b) Cuser2base station = 3.46B = 69.2M bps
P
4. (a) Ergodic Capacity (with Rcvr CSI only)= B[ 6i=1 log2 (1 + i )p(i )] = 2.8831B = 57.66 Mbps.
(b) pout = P r( < min )
Co = (1-pout )Blog2 (1 + min )
For
min > 20dB, pout = 1, Co = 0
15dB < min < 20dB, pout = .9, Co = 0.1Blog2 (1 + min ), max Co at min 20dB.
min < 5dB, pout = 0, Co = Blog2 (1 + min ), max Co at min -5dB.
Plot is shown in Fig. 1. Maximum at min = 10dB, pout =0.5 and Co = 34.59 Mbps.
5. (a) We suppose that all channel states are used
4
X 1
1
=1+
pi 0 = 0.8109
0
i
i=1
1
1
> 0 true
0 4
S(i )
1
1
=
S
0
i
3.5
x 10
Capacity (bps)
2.5
1.5
0.5
0.2
0.4
0.6
0.8
Pout
Figure 1: Capacity vs Pout
1.2322 = 1
S()
1.2232 = 2
=
1.1332
= 3
0.2332 = 4

4
X
C
i
=
log2
p(i ) = 5.2853bps/Hz
B
0
i=1
(b)
1
= E[1/]
S(i )
= i
S
= 4.2882
0.0043
S() 0.0029
=
0.4288
4.2882
= 1
= 2
= 3
= 4
C
= log2 (1 + ) = 2.4028bps/Hz
B
(c) To have pout = 0.1 or 0.01 we will have to use all the sub-channels as leaving any of these will
result in a pout of at least 0.2 truncated channel power control policy and associated spectral
efficiency are the same as the zero-outage case in part b .
To have pout that maximizes C with truncated channel inversion, we get
max
C
= 4.1462bps/Hz
B
6. (a)
pout = 0.5
SN Rrecvd
10dB
P (d)
5dB
=
=
0dB
Pnoise
10dB
w.p.
w.p.
w.p.
w.p.
0.4
0.3
0.2
0.1
Assume all channel states are used

4
X 1
1
=1+
pi 0 = 0.4283 > 0.1
0
i
i=1
not possible
Now assume only the best 3 channel states are used

3
X 1
0.9
=1+
pi 0 = 0.6742 < 1
0
i
ok!
i=1
1.3832
S()
1.1670
=
0.4832
= 1
= 2
= 3
= 4
= 10
= 3.1623
=1
= 0.1
C/B = 2.3389bps/Hz
(b) = 0.7491
C/B = log2 (1 + ) = 0.8066bps/Hz
C
(c) B max = 2.1510bps/Hz obtained by using the best 2 channel states.
With pout = 0.1 + 0.2 = 0.3
7. (a) Maximize capacity given by
S()
B log 1 +
p()d.
S
Z
C=
max
R
S(): S()p()d=S
Construct the Lagrangian function
Z
Z
S()
S()
L = B log 1 +
p()d
p()d
S
S
Taking derivative with respect to S(), (refer to discussion section notes) and setting it to zero,
we obtain,
1
1
S()
0 0
=
0
< 0
S
Now, the threshold value must satisfy
Z
0
Evaluating this with p() =
1 /10
,
10 e
1 =
=
1
1
p()d = 1
we have
Z /10
1
e
e
d
d
10 0
0
Z
1 0 /10
1
e
e
0
10 0
1
100
/10
(1)
(2)
10
1 0 /10
1
e
EXPINT(0 /10)
0
10
(3)
where EXPINT is as defined in matlab. This gives 0 = 0.7676. The power adaptation becomes
1
1
S()
0.7676 0.7676
=
0
< 0.7676
S
(b) Capacity can be computed as

Z
1
C/B =
log (/0.7676) e/10 d = 2.0649 nats/sec/Hz.
10 0.7676
Note that I computed all capacites in nats/sec/Hz. This is because I took the natural log. In
order to get the capacity values in bits/sec/Hz, the capacity numbers simply need to be divided
by natural log of 2.
(c) AWGN capacity C/B = log(1 + 10) = 2.3979 nats/sec/Hz.
(d) Capacity when only receiver knows
Z
1
C/B =
log (1 + ) e/10 d = 2.0150 nats/sec/Hz.
10 0
(e) Capacity using channel inversion is ZERO because the channel can not be inverted with finite
average power. Threshold for outage probability 0.05 is computed as
Z
1
e/10 d = 0.95
10 0
which gives 0 = 0.5129. This gives us the capacity with truncated channel inversion as
"
#
1
0.95
C/B = log 1 + 1 R 1 /10
d
10 0 e
"
#
1
= log 1 + 1
0.95
10 EXPINT(0 /10)
= 1.5463
nats/s/Hz.
(4)
(5)
(6)
(f) Channel Mean=-5 dB = 0.3162. So for perfect channel knowledge at transmitter and receiver we
compute 0 = 0.22765 which gives capacity C/B = 0.36 nats/sec/Hz.
With AWGN, C/B = log(1 + 0.3162) = 0.2748 nats/sec/Hz.
With channel known only to the receiver C/B = 0.2510 nats/sec/Hz.
Capacity with AWGN is always greater than or equal to the capacity when only the reciever
knows the channel. This can be shown using Jensens inequality. However the capacity when
the transmitter knows the channel as well and can adapt its power, can be higher than AWGN
capacity specially at low SNR. At low SNR, the knowledge of fading helps to use the low SNR
more efficiently.
8. (a) If neither transmitter nor receiver knows when the interferer is on, they must transmit assuming
worst case, i.e. as if the interferer was on all the time,
S
= 10.7Kbps.
C = B log 1 +
N0 B + I
(b) Suppose we transmit at power S1 when jammer is off and S2 when jammer is off,

S2
S1
0.75 + log 1 +
0.25
C = B max log 1 +
No B
No B + I
subject to
0.75S1 + 0.25S2 = S.
This gives S1 = 12.25mW , S2 = 3.25mW and C = 53.21Kbps.
(c) The jammer should transmit x(t) to completely cancel off the signal.
S = 10mW
N0 = .001 W/Hz
B = 10 MHz
Now we compute the SNRs as:
j =
This gives: 1 =
|1|2 103
0.001106 10106
|Hj |2 S
N0 B
= 1, 2 = .25, 3 = 4, 4 = 0.0625
To compute 0 we use the power constraint as:
X 1
1
=1
0 j +
j
First assume that 0 < 0.0625, then we have
1
4
1
1
1
=1+
+
+ +
0
1 .25 4 .0625
0 = .1798 > 0.0625
So, our assumption was wrong. Now we assume that 0.0625 < 0 < .25, then
1
1
1
3
=1+
+
+
0
1 .25 4
0 = .48 > 0.25
So, our assumption was wrong again. Next we assume that 0.25 < 0 < 1, then
2
1 1
=1+
+
0
1 4
0 = .8889 < 1
This time our assumption was right. So we get that only two sub-bands each of bandwidth 10 MHz
are used for transmission and the remaining two with lesser SNRs are left unused.
Now, we can find capacity as:
C=
X
j:j 0
B log2
j
0
This gives us, C = 23.4 Mbps.

9. Suppose transmit power is Pt , interference
power is Pint and noise power is Pnoise .
Pt
In the first strategy C/B = log2 1 + Pint +Pnoise
int
In the second strategy C/B = log2 1 + PPt P
noise
Assuming that the transmitter transmits -x[k] added to its message always, power remaining for actual
messages is Pt Pint
The first or second strategy may be better depending on
Pt
Pt Pint
Pt Pint Pnoise 0
Pint + Pnoise
Pnoise
Pt is generally greater than Pint + Pnoise , and so strategy 2 is usually better.
H(f)
0.5
0.25
fc-20
fc+20
fc
fc-10
fc+10
f (in MHz)
10. We show this for the case of a discrete fading distribution
(1 + j)2 Pj
C = log 1 +
N0 B
2
X
X
(1 + j) Pj
Pj P
L=
log 1 +
dj
N0 B
i
L
Pj
(1 + j)2 Pj
N0 B
(1 + j)2 P
letj =
N0 B
Pj
P
1
1
denote =
0
P
Pj
P
subject to the constraint
Pj
P
= 0
=
1 (1 + j)2
1
N0 B
1
1
P
j
1
1
0 j
= 1
11. S = 10mW
N0 = .001 W/Hz
B = 10 MHz
Now we compute the SNRs as:
j =
This gives: 1 =
|1|2 103
0.001106 10106
|Hj |2 S
N0 B
= 1, 2 = .25, 3 = 4, 4 = 0.0625
To compute 0 we use the power constraint as:

X 1
j
0 j
=1
+
First assume that 0 < 0.0625, then we have
4
1
1
1
1
=1+
+
+ +
0
1 .25 4 .0625
0 = .1798 > 0.0625
So, our assumption was wrong. Now we assume that 0.0625 < 0 < .25, then
3
1
1
1
=1+
+
+
0
1 .25 4
0 = .48 > 0.25
So, our assumption was wrong again. Next we assume that 0.25 < 0 < 1, then
2
1 1
=1+
+
0
1 4
0 = .8889 < 1
This time our assumption was right. So we get that only two sub-bands each of bandwidth 10 MHz
are used for transmission and the remaining two with lesser SNRs are left unused.
Now, we can find capacity as:
C=
B log2
j:j 0
j
0
This gives us, C = 23.4 Mbps.

12. For the case of a discrete number of frequency bands each with a flat frequency response, the problem
can be stated as
X
|H(fi )|2 P (fi )
log2 1 +
Pmax
N0
s.t. i P (fi )P
i
denote (fi ) =
)|2 P (f
|H(fi
N0
i)
L=
X
i
denote xi =
P (fi )
P ,
X
P (fi )
log2 1 + (fi )
+ (
P (fi ))
P
the problem is similar to problem 10

x?i =
1
1
0 (fi )
where 0 is found from the constraints
X 1
1
1
1
= 1 and
0i
0 (fi )
0 (fi )
i
13. (a) C=13.98Mbps
MATLAB
Gammabar = [1 .5 .125];
ss = .001;
P = 30e-3;
N0 = .001e-6;
Bc = 4e6;
Pnoise = N0*Bc;
hsquare = [ss:ss:10*max(Gammabar)];
gamma = hsquare*(P/Pnoise);
for i = 1:length(Gammabar)
pgamma(i,:) = (1/Gammabar(i))*exp(-hsquare/Gammabar(i));
end
gamma0v = [1:.01:2];
for j = 1:length(gamma0v)
gamma0 = gamma0v(j);
sumP(j) = 0;
a = gamma.*(gamma>gamma0);
[b,c] = max(a>0);
gammac = a(find(a));
pgammac = pgamma(i,c:length(gamma));
Pj_by_P = (1/gamma0)-(1./gammac);
sumP(j) = sumP(j) + sum(Pj_by_P.*pgammac)*ss;
end
end
[b,c] = min(abs((sumP-1)));
gamma0ch = gamma0v(c);
C = 0;
a = gamma.*(gamma>gamma0ch);
[b,c] = max(a>0);
pgammac = pgamma(i,c:length(gamma));
C = C + Bc*ss*sum(log2(gammac/gamma0ch).*pgammac);
end
(b) C=13.27Mbps
MATLAB
Gammabarv = [1 .5 .125];
ss = .001;
Pt = 30e-3;
N0 = .001e-6;
Bc = 4e6;
Pnoise = N0*Bc;
P = Pt/3;
for k = 1:length(Gammabarv)
Gammabar = Gammabarv(k);
hsquare = [ss:ss:10*Gammabar];
gamma = hsquare*(P/Pnoise);
pgamma = (1/Gammabar)*exp(-hsquare/Gammabar);
gamma0v = [.01:.01:1];
for j = 1:length(gamma0v)
gamma0 = gamma0v(j);
a = gamma.*(gamma>gamma0);
[b,c] = max(a>0);
pgammac = pgamma(c:length(gamma));
Pj_by_P = (1/gamma0)-(1./gammac);
sumP(j) = sum(Pj_by_P.*pgammac)*ss;
end
[b,c] = min(abs((sumP-1)));
gamma0ch = gamma0v(c);
a = gamma.*(gamma>gamma0ch);
[b,c] = max(a>0);
pgammac = pgamma(c:length(gamma));
C(k) = Bc*ss*sum(log2(gammac/gamma0ch).*pgammac);
end Ctot = sum(C);
Chapter 5
P
1. si (t) = Pk sik k (t)
sj (t) = k sjk k (t)
where {k (t)} forms an orthonormal basis on the interval [0,T]
!2
Z T
Z T X
X
[si (t) sj (t)]2 dt =
sim m (t)
sjm m (t) dt
0
!2
(sim sjm ) m (t)
dt
Notice all the cross terms will integrate to 0 due to orthonormal property. So we get
Z TX
=
(sim sjm )2 m (t)m (t)dt
0
(sim sjm )2 = ||si sj ||2
2. Consider
Z
0
2t
2t
1
sin
+ cos
1 =
T
T
T

1
2t
2t
2 =
sin
cos
T
T
T
Z T
Z T
1 (t)2 (t) = 0,
21 (t) = 1,
22 (t) = 1
0
3.
s0m (t)
M
1 X
= sm (t)
si (t)
M
i=1
Z
0
=
0
Z
s02
m (t)dt
=
0
!2
M
1 X
sm (t)
si (t) dt
M
i=1
si s are orthonormal so the cross terms integrate to 0 and we get

Z T
Z T
1 2
1
(M 1)
0 =
s2m (t)dt
sm (t)dt =
=
M
M
M
0
0
!
!
Z T
M
M
1 X
1 X
< s0m (t), s0n (t) >=

sm (t)
si (t)
sn (t)
si (t) dt =
M
M
M
0
i=1
i=1
4. (a)
Z
< f1 (t), f2 (t) >=
f1 (t)f3 (t)dt = 0
Z
< f2 (t), f3 (t) >=
f1 (t)f2 (t)dt = 0
Z
< f1 (t), f3 (t) >=
f2 (t)f3 (t)dt = 0
f1 (t), f2 (t), f3 (t) are orthogonal

(b)
a = 2,
x(t) = af1 (t) + bf2 (t) + cf3 (t)

1
1
1
0 t 1 : x(t) =
a + b + c = 1
2
2
2
1
1
1
1 t 2 : x(t) =
a+ b c=1
2
2
2
1
1
1
2 t 3 : x(t) = a + b + c = 1
2
2
2
1
1
1
3 t 4 : x(t) = a + b c = 3
2
2
2
b = 2, c = 2
x(t) = 2f1 (t) + 2f2 (t) 2f3 (t)
5. (a) A set of orthonormal basis functions is

In this set the given waveforms can be written as
s1 = [1 2 -1 -1]
s2 = [1 -1 1 -1]
s3 = [-2 1 1 1]
s4 = [1 -2 -2 2]
now we can see using Matlab or otherwise that the dimensionality is 4
(b) done in part a
(c)
||s1 s2 ||2 = 14
||s1 s3 ||2 = 22
||s1 s4 ||2 = 27
||s2 s3 ||2 = 14
||s2 s4 ||2 = 19
||s3 s4 ||2 = 31
||s1 ||2 = 10
||s2 ||2 = 4
||s3 ||2 = 6
||s4 ||2 = 13
The minimum distance between any pair is
14
6. From 5.28 we have

m
b = mi corresponding to sbi = arg maxsi p(|si )p(si )
max L(si ) = p(|si )p(si )
si
max l(si ) = logL(si ) = log p(|si ) + log p(si )

si
N
N
1 X
= max log(N0 )
(j sij )2 + log p(si )
si
2
N
0
|
{z
}
j=1
constant
t
4
Figure 1: Problem 5
Ac
-A
S
Ac
- Ac
S
4
Figure 2: Problem 6
= max
si
= min
si
1
||j sij ||2 + log p(si )
N0
1
||j sij ||2 + log 1/p(si )
N0
s1 = (Ac , 0)
s2 = (0, Ac )
s3 = (Ac , 0)
s4 = (0, Ac )
p(s1 ) = p(s3 ) = 0.2

p(s2 ) = p(s4 ) = 0.3
1
2 1
2
2
zi = x < | ||x si || + log 1/p(si ) <
||x sj || + log 1/p(sj )i 6= j
N0
N0
which can be further solved using Matlab or otherwise far a given value of Ac and N0
P
7. nr (t) = n(t) N
j=1 nj j (t)
j = sij + nj
We know N span the signal space. Suppose we add (M-N) additional basis vectors so that 1 . . . M
span the noise space . This can always be done for some M(may be infinite). Also MN
n(t) =
M
X
nk k (t) wherei form an orthonormal set
k=1
then
M
N
X
X
np p (tp ) (sij + nj )
E[nr (tk )rj ] = E
nk k (tk )
k=1
p=1
Since the signal is always independent of noise and white noise components in the orthogonal directions
are independent too, we have
E[nr (tk )rj ] = E[n2j j (tk )] E[n2j j (tk )] = 0 , for j = 1 . . . N
2.5
1.5
0.5
g(t)
g(Tt)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Figure 3: Problem 9a
8. Suppose sm (t) is transmitted and Nt is noise added. If the k th filter is hk (t), the output of the k th
filter is
Z
yk (t) =
(sm ( ) + N )hk (t )d
sampling at time T gives

Z
yk (T ) =
Denote noise contribution as k =
sm ( )hk (T )d +
N hk (T
N hk (T )d
)d
E[k ] = 0
Z
N0
|hk (T )|2 d
2k =
2
R
2
Signal energy = sm ( )hk (T )d

2
s
(
)h
(T
)d
m
k
R
N0
2
2 |hk (T )| d
SN R =
use Cauchy-Schwartz inequality to get upper bound on SNR as

Z
2
|sm ( )|2 d
SN R
N0
with equality iff
opt
hopt
k (T ) = sm ( ) hk (t) = sm (T t)
which is the required result for matched filter

q
9. (a) g(t) = T2 0 t T
q
g(T t) = T2 0 t T
plotted for T=1 , integral value = 2/T = 2
0.9
0.8
0.7
g(t)
g(Tt)
0.6
0.5
0.4
0.3
0.2
0.1
0.1
0.2
0.3
0.4
0.5
t
0.6
0.7
0.8
0.9
0.9
Figure 4: Problem 9b
1.8
g(t)
g(Tt)
1.6
1.4
1.2
0.8
0.6
0.4
0.2
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Figure 5: Problem 9c
(b) g(t) = sinc(t) 0 t T
g(T t) = sinc(T t) 0 t T
plotted for T=1 , integral value = 0.2470
(c) g(t) =
2 t2 /2
0tT
e
2 (T t)2 /2
t) == e
g(T
0tT
plotted for T=1 , integral value = 0.009
MATTLAB CODE
T = 1; alpha = 1;
t = [0:.01:T];
%% Part a)
g = repmat(sqrt(2/T),1,length(t));
gm=repmat(sqrt(2/T),1,length(t));
int_a = sum(g.*gm)*.01;
plot(t,g,b--); hold on; plot(t,gm,b:);
%% Part b)
figure;
g = sinc(t);
gm = sinc(T-t);
int_b = sum(g.*gm)*.01;
plot(t,g,b--);
hold on;
plot(t,gm,b:);
%% Part c)
figure; g = (sqrt(pi)/alpha)*exp(-((pi)^2*t.^2)/alpha^2);
gm=(sqrt(pi)/alpha)*exp(-((pi)^2*(T-t).^2)/alpha^2);;
int_c=sum(g.*gm)*.01; plot(t,g,b--);
hold on;
plot(t,gm,b:);
RT
10. For Fig 5.4 k = 0 ( )k ( )d
RT
RT
For Fig 5. k = 0 ( )k (T (T ))d = 0 ( )k ( )d which is the same as above.
P P
dij
11. (5.40) gives 14 4i=1 4j=1,j6=i Q 2N
= 4.1 109
0
(5.43) gives (4 1)Q dmin

= 2.3 108
2N2 0
dmin
(5.44) gives (41)

exp 4N
= 1.9 107
dmin
dmin
(5.45) gives Mdmin Q

= 2Q
= 1.5 108
2N
2N
0
MATLAB CODE
Ac = 4;
s(1,:) = [Ac 0];
s(2,:) = [0 2*Ac];
s(3,:) = [0 -2*Ac];
s(4,:) = [-Ac 0];
sume = 0; for i = 1:4
for j = 1:4
if j ~= i
d(i) = norm(s(i,:)-s(j,:));
sume = sume+Q(d(i)/sqrt(2));
end
end
end E1 = .25*sume;
dmin = min(d);
E2 = 3*Q(dmin/sqrt(2));
E3 = (3/sqrt(pi))*exp(-dmin^2/4);
E4=2*Q(dmin/sqrt(2));
Ac
-A
S
Ac
- Ac
S
4
4-PSK
d min =
8-PSK
2
d min =
12.
Z
1 (t) =
( ) cos(2fc + )g(t (T ))d

[si ( ) + n( )] cos(2fc + )g(t (T ))d
si ( ) = si1 g( ) cos(2fc + 0 ) + si2 g( ) sin(2fc + 0 )

Z
si1
[cos(4fc + + 0 ) + cos( 0 )]g( )g(t (T ))d +
1 (t) =
2
Z
si2
[cos(4fc + + 0 ) cos( 0 )]g( )g(t (T ))d +
2
Z
n( ) cos(2fc + )g(t (T ))d
where 0 =
Similrly we can find 2 (t). Notice that terms involving fc will integrate to 0 approximately as fc T 1
13. For 4PSK dmin = p2 4P SK = 1

1
For 8PSK dmin = + 2 cos(/4) 8P SK = 1cos(/4)
= 3.4142
extra energy factor = 3.4142 = 5.33dB
14. For square QAM constellations, it is easy to derive that
!
d2min 2l
dmin 1 l
sl =
(2 1) = p
(4 1)
6
(2) 3
sl
sl+1
4l
3
44l
= 4sl
3
(16 QAM) For MQAM, sl (l = 2) = 2.5d2min

(4 PAM) For MPAM, sl (l = 2) = 1.25d2min
(16 PSK) For MPSK, sl (l = 2) =
d2min
2(1cos(/8))
= 6.5685d2min
15. M points are separated by an angle 2

M
If || > 21 2
M , we will go into the decision region for another adjacent symbol and so will make a
detection error
16. Gray encoding of bit sequence to phase transitions:
We first draw the figure and write down bit sequences for each phase in a way that exactly 1 bit changes
between nearest neighbors. We get the following table
Bit Sequence
000
001
010
011
100
101
110
111
Phase transition
0
/4
3/4
/2
/4
/2
3/4
The resulting encoding of the given sequence is as follows:

Bit Sequence
101
110
100
101
110
Mapped Symbol
s(k-1) = Aej/4
s(k) = Aej/4
s(k + 1) = Aej3/4
s(k + 2) = Aej/2
s(k + 3) = A
s(k + 4) = Aej
17. (a) a = 0.7071A

b = 1.366A
(b) A2 = r2 2 2 cos 4
r = 1.3066A
(c) Avg power of 8PSK = r2 = 1.7071A2
Avg power of 8 QAM = 1.1830A2
The 8QAM constellation has a lower average power by a factor of 1.443 (1.593 dB)
(d) See Fig 10
(e) We have 3 bits per symbol symbol rate = 30 Msymbols/sec
2
011
010
001
000
110
1
100
111
101
A
b
a=A
8-PSK AND 8-QAM

111
011
110
100
010
101
000
001
Figure 10: Problem 17d
S2
S3
S2
S3
S1
S1
S4
S4

S2
00
S3
01
00
S2
10
S3 01
S1
10
11
S4
11
S1
S4

18. (a) one set is on the axis and the other is /4 offset. If the current symbol uses a point of , the
next symbol must come from and vice versa
(b) See Fig 12
(c) 0 1 00 10 01 11 10 01 01 Assume we start from the points
s3 s02 s1 s03 s4 s01 s3 s03
(d) See Fig 13
b1 b0
Phase change from previous symbol
00
01
11
10
/4
3/4
3/4
/4
had phase
Given, last symbol of

Given, last symbol of
/4
00
/4
had phase
Assuming we started from
-3
-3
01
points
S
2
00
S
1
10
S
3
01
S
1
S
1
11
10
S
2
01
S
4
01
Initial
symbols
Figure 13: Problem 18d

19.
2 cos(2fi t) cos(2fj t)dt = 0
|0
Z T
2 cos (2(fi + fj )t) +
2 cos (2(fi fj )t)
0
{z
} |
{z
}
A
A 0 , as fi + fj 1
B = sin (2(fi fj )T )
is 0 first time for 2(fi fj )T = |fi fj | = 0.5T
MATLAB CODE
gamma_dB = [0:.01:60];
gamma = 10.^(gamma_dB/10);
vBdT = .01;
x = 2*pi*vBdT;
rho_C = besselj(0,x);
Pb_bar=.5*((1+gamma*(1-rho_C))./(1+gamma));
semilogy(gamma_dB,Pb_bar);
hold on;
vBdT = .001;
x = 2*pi*vBdT;
Pb_bar= .5*((1+gamma*(1-rho_C))./(1+gamma));
semilogy(gamma_dB,Pb_bar,b:);
vBdT = .0001;
x = 2*pi*vBdT;
Pb_bar=.5*((1+gamma*(1-rho_C))./(1+gamma));
semilogy(gamma_dB,Pb_bar,b--);
20.
p(kT ) =
p0 = p(0) k = 0
0
k=
6 0
. . . (a)
10
BDT = .01
BDT = .001
BDT = .0001
10
10
Pbbar
10
10
10
10
10
10
20
30
40
50
60
bar in dB

Z
P (f )ej2f t df
p(t) =
Z
p(kT ) =
p(kT ) =
=
P (f )ej2f kT df
Z
P (f )ej2f kT df
m= (2m1)/2T
Z 1/2T
P f+
m= 1/2T
1/2T X
Z
=
(2m+1)/2T
1/2T
1/2T
m j2f kT
e
df
T
m j2f kT
P f+
e
df
T
m=
Q(f )ej2f kT df
p(kT ) =
...1
1/2T
Q(f) is periodic with period 1/T and therefore it can be expanded in terms of Fourier coefficients
Q(f ) =
qn ej2f nT
n=
Z 1/2T
where qn = T
Q(f )ej2f nT df
1/2T
Compare 1 and 2 to get

qn = T p(nT )
a translates to
qn =
But this means that Q(f ) = p0 T or

21. Gaussian pulse is given as
l= P
p0 T
0
n=0
n 6= 0
(f + l/T ) = p0 T
2 t2 /2
g(t) =
e
Notice that g(t) never goes to 0 except at

Nyquist criterion
g0 k = 0
g(kT ) =
0 k 6= 0
...2
cannot be satisfied for any finite T.
Chapter 6
1. (a) For sinc pulse, B =
1
2Ts
Ts =
1
2B
= 5 105 s
(b) SN R = NP0bB = 10
Since 4-QAM is multilevel signalling
2Es
1
s
SN R = NP0bB = N0EBT
=
BT
=
s
N0 B
2
s
Es
SNR per symbol = N
=
5
0
Eb
SNR per bit = N
=
2.5
(a symbol has 2 bits in 4QAM)
0
Es
(c) SNR per symbol remains the same as before = N
=5
0
SNR per bit is halved as now there are 4 bits in a symbol
Eb
N0
= 1.25
2. p0 = 0.3, p1 = 0.7
(a)
dmin
Pe = P r(0 detected, 1 sent 1 sent)p(1 sent) + P r(1 detected, 0 sent 0 sent)p(0 sent)
dmin
dmin
dmin
+ 0.3Q
=Q
= 0.7Q
2N0
2N0
2N0
= 2A
s
2
2A
= Q
N0
(b)
p(m
= 0|m = 1)p(m = 1) = p(m
= 1|m = 0)p(m = 0)
A+a
Aa
0.7Q q = 0.3Q q , a > 0
N0
2
N0
2
Solving gives us a for a given A and N0

(c)
p(m
= 0|m = 1)p(m = 1) = p(m
= 1|m = 0)p(m = 0)
B
A
0.7Q q = 0.3Q q , a > 0
N0
2
Clearly A > B, for a given A we can find B

2
Eb
A
(d) Take N
=N
= 10
0
0
In part a) Pe = 3.87 106
In part b) a=0.0203 Pe = 3.53 106
In part c) B=0.9587 Pe = 5.42 106
Clearly part (b) is the best way to decode.
MATLAB CODE:
A = 1;
N0 = .1;
a = [0:.00001:1];
t1 = .7*Q(A/sqrt(N0/2));
N0
2
t2=.3*Q(a/sqrt(N0/2));
diff = abs(t1-t2);
[c,d] = min(diff);
a(d)
c
3. s(t) = g(t) cos 2fc t
r = r cos
where r is the signal after the sampler if there was no phase offset. Once again, the threshold that
minimizes Pe is 0 as (cos ) acts as a scaling factor for both +1 and -1 levels. Pe however increases
as numerator is reduced due to multiplication by cos
dmin cos
Pe = Q
2N0
4.
Z
A2c
Tb
Z
2
A2c
cos 2fc tdt =
Tb
1 + cos 4fc t
2
T
b sin(4fc Tb )
= A2c +
8fc
2
{z
}
|
0 as fc 1
A2c Tb
=1
2
x(t) = 1 + n(t)
Let prob 1 sent =p1 and prob 0 sent =p0
1
[1.p1 + 0.p0 ] +
6
1
[0.p1 + 1.p0 ]
6
1
1
[p1 + p0 ] =
6
6
Pe =
2
2
[0.p1 + 0.p0 ] + [0.p1 + 0.p0 ] +
6
6
( p1 + p0 = 1 always )
dmin
2N0
5. We will use the approximation Pe (average number of nearest neighbors).Q

where number of nearest neighbors = total number of points taht share decision boundary
(a) 12 inner points have 5 neighbors
4 outer points have 3 neighbors
avg number
= 4.5
of neighbors
2a
Pe = 4.5Q
2N0
2a
=
3Q
(b) 16QAM, Pe = 4 1 14 Q 2a
2N0
2N0
2a
2a
23+32
Q 2N = 2.4Q 2N
(c) Pe
5
0
0
3a
14+43+42
3a
=
2.67Q
(d) Pe
Q
9
2N
2N
0
6.
Ps,exact = 1
!!2
r
3 s
2( M 1)
1
Q
M 1
M
Figure 1: Problem 5
Ps,approx
r
!
4( M 1)
3 s
=
Q
M 1
M
approximation is better for high SNRs as then the multiplication factor is not important and Pe is
dictated by the coefficient of the Q function which are same.
MATLAB CODE:
gamma_db = [0:.01:25];
gamma = 10.^(gamma_db/10);
M = 16;
Ps_exact=1-exp(2*log((1-((2*(sqrt(M)-1))/(sqrt(M)))*Q(sqrt((3*gamma)/(M-1))))));
Ps_approx = ((4*(sqrt(M)-1))/sqrt(M))*Q(sqrt((3*gamma)/(M-1)));
semilogy(gamma_db, Ps_exact);
hold on
semilogy(gamma_db,Ps_approx,b:);
7. See figure. The approximation error decreases with SNR because the approximate formula is based
on nearest neighbor approximation which becomes more realistic at higher SNR. The nearest neighbor
bound over-estimates the error rate because it over-counts the probability that the transmitted signal
is mistaken for something other than its nearest neighbors. At high SNR, this is very unlikely and this
over-counting becomes negligible.
8. (a)
Z
Ix (a) =
eat
dt
x2 + t2
since the integral converges we can interchange integral and derivative for a0
Ix (a)
a
Ix (a)
x Ix (a)
a
2
teat
dt
x2 + t2
(x2 + t2 )eat
dt =
x2 + t2
Z
=
e
0
at2
1
dt =
2
Problem 2 Symbol Error Probability for QPSK
10
100
Ps
10
200
Approximation
Exact Formula
10
300
10
10
15
20
25
30
Problem 2 Symbol Error Probability for QPSK
10
10
Approximation
Exact Formula
10
10
5
SNR(dB)
10
Figure 2: Problem 7
(b) Let Ix (a) = y, we get
1
y x y =
2
comparing with
0
y 0 + P (a)y = Q(a)
2
P (a) = x
R
I.F. = e
P (u)u
= ex
2a
x2 a
1
Q(a) =
2
y =
solving we get
y =
x2 u
e
du
a
ax2
e erf c(x a)
2x
(c)
2x ax2
2x
2
e
erf c(x a) = Ix (a) eax =
a=1
Z
2
2x ax2 eat
erf c(x) =
e
dt
x2 + t2
0
Z
2 /2 x2 /sin2
=
e
d
0
Z
1
1 /2 x2 /2sin2
Q(x) = erf c(x/ 2) =
e
d
2
0
Z
0
eat
dt
x2 + t2
9. P = 100W
N0 = 4W, SN R =25
Pe = Q( 2) = Q( 50) = 7.687 1013

data requires Pe 106
voice requires Pe 103
so it can be used for both.
with fading Pe = 41 = 0.01
b
So the system cant be used for data at all. It can be used for very low quality voice.
10. Ts = 15sec
1
at 1mph Tc = B1d = v/
= 0.74s Ts
outage probability is a good measure.
at 10 mph Tc = 0.074s Ts outage probability is a good measure.
at 100 mph Tc = 0.0074s = 7400s > 15s outage or outage combined with average prob of error can
be a good measure.
11.
Z
M (s) =
Z
=
es p()d
1
es e/ d
1
1 s
12. (a) When there is path loss alone, d = 1002 + 5002 = 100 6 103
Pe = 12 eb b = 13.1224
P
14
N0 B = 13.1224 P = 1.3122 10
h i2
P
G
4.8488W
Pt = 4d
(b)
x = 1.3122 1014 = 138.82dB
P,dB N (P , 8), dB = 8
P (P,dB x)
P,dB P
x P
P
8
8
x P
Q
8
x P
8
P = 128.5672dB = 1.39 1013
= 0.9
= 0.9
= 0.9
= 1.2816
13. (a) Law of Cosines:
c2 = a2 + b2p
2ab cos C with a,b = Es , c = dmin , C = = 22.5
c = dmin = 2Es (1 cos 22.5) = .39 Es
Can also use formula fromreader
q
dmin 2
(b) Ps = m Q m s = 2Q
= 2Q( .076s )
2No
m = 2, m = .076
(c) Pe =
=
Ps (s )f (s )ds
m Q( m s )f (s )ds
Using alternative Q form
1
R2
gs
m
=
1 + (sin
d with g = 2m
2
)
0h
q
q
i
gs
.038s
= 2m 1 1+g
=
1
1+.038s =
s
the integral
(d) Pd =
1
.076s ,
where we have used an integral table to evaluate
Ps
4
(e) BPSK: P b = 41 = 103 , b = 250, 16PSK: From above get s = 3289.5

b
Penalty = 3289.5
250 = 11.2dB
Also will accept b (16P SK) = 822 = 5.2dB
14.
Z
Pb =
But from 6.65

M (s) =
Pb =
For M = 4,
= 10
Pb ()p()d
1
Pb () = e
2
Pb =
1
2
Z
0
1
eb p ()d = M
2
s m
1
m
1
m
1+
2
m
P b = 3.33 103
15. %Script used to plot the average probability of bit error for BPSK modulation in
%Nakagami fading m = 1, 2, 4.
%Initializations
avg_SNR = [0:0.1:20]; gamma_b_bar = 10.^(avg_SNR/10); m = [1 2 4];
line = [-k, -r, -b]
for i = 1:size(m,2)
for j = 1:size(gamma_b_bar, 2)
Pb_bar(i,j) = (1/pi)*quad8(nakag_MGF,0,pi/2,[],[],gamma_b_bar(j),m(i),1);
end
figure(1);
semilogy(avg_SNR, Pb_bar(i,:), line(i));
hold on;
end
xlabel(Average SNR ( gamma_b ) in dB); ylabel(Average bit error
probability ( P_b ) ); title(Plots of P_b for BPSK modulation in
Nagakami fading for m = 1, 2, 4); legend(m = 1, m = 2, m =
4);
function out = nakag_MGF(phi, gamma_b_bar, m, g);

%This function calculates the m-Nakagami MGF function for the specified values of phi.
%phi can be a vector. Gamma_b_bar is the average SNR per bit, m is the Nakagami parameter
%and g is given by Pb(gamma_b) = aQ(sqrt(2*g*gamma_b)).
out = (1 + gamma_b_bar./(m*(sin(phi).^2)) ).^(-m);
SNR = 10dB
M
1
2
4
BER
2.33102
5.53103
1.03103
16. For DPSK in Rayleigh fading, Pb = 21b b = 500

No B = 3 1012 mW Ptarget = b N0 B = 1.5 109 mW = -88.24 dBm
Now, consider shadowing:
P
Pr
Pout = P [Pr < Ptarget ] = P [ < Ptarget Pr ] = target
Ptarget Pr
3.73 108
1 (.01) = 2.327 =
Pr = 74.28 dBm =
d = 1372.4 m
mW = Pt
2
4d
17. (a)
1 =
2 =
0 w.p. 1/3
30 w.p. 2/3
5 w.p. 1/2
10 w.p. 1/2
In MRC, = 1 +2 . So,
10
=
35
40
w.p.
w.p.
w.p.
w.p.
1/6
1/6
1/3
1/3
(b) Optimal Strategy is water-filling with power adaptation:

S()
=
S
1
0
1 , 0
0
< 0
Notice that we will denote by only hereon to lighten notation. We first assume 0 < 5,
4
X
1
1
pi = 1
0 i
i=1
X pi
1
=1+
0
i
i=1
0 = 0.9365 < 5
So we found the correct value of 0 .
C=B
4
X
log2
i=1
i
0
pi
C = 451.91 Kbps
(c) Without, receiver knowledge, the capacity of the channel is given by:
C=B
4
X
log2 (1 + i )pi
i=1
C = 451.66 Kbps
Notice that we have denote by to lighten notation.
18. (a)
s(k) = s(k 1)
z(k 1) = gk1 s(k 1) + n(k 1)
z(k) = gk s(k) + n(k)
From equation 5.63 , the input to the phase comparator is
z(k)z ? (k 1) = gk g( k 1)? s(k)s? (k 1) + gk s(k 1)n?k1 +
g( k 1)? s? (k 1)nk + nk n?k1
but s(k) = s(k 1)
s(k)s? (k 1) = |sk |2 = 1
(normalized)
(b)
n
ek = s?k1 nk
n
ek1 = s?k1 nk1
?
?
ze = gk gk1
+ gk n
e?k1 + gk1
n
ek
p1 p2
A
B
x (s) =
=
+
(s p1 )(s p2 )
s p1 s p2
p1 p2
A = (s p1 )x (s)|s=p1 =
p1 p2
p1 p2
B = (s p2 )x (s)|s=p2 =
p2 p1
(c) Relevant part of the pdf
p1 p2
(p2 p1 )(s p2 )
p1 p2
1
p1 p2
1
px (x) =
L
=
ep2 x
(p2 p1 )
(s p2 )
(p2 p1 )
x (s) =
(d)
p1 p2
Pb = prob(x < 0) =
(p2 p1 )
ep2 x dx =
,x < 0
p1
p2 p1
(e)
p2 p1 =
1
1
b + 1
+
=
2N0 [ b (1 c ) + 1] 2N0 [ b (1 + c ) + 1]
N0 [ b (1 c ) + 1][ b (1 + c ) + 1]
Pb =
b (1 c ) + 1
2( b + 1)
(f) c = 1
Pb =
1
2( b + 1)
19. b 0 to 60dB
c = J0 (2BD T ) with BD T = 0.01, 0.001, 0.0001
where J0 is 0 order Bessel function of 1st kind.
1 1 + b (1 c )
Pb =
2
1 + b
when BD T = 0.01, floor can be seen about b = 40dB
when BD T = 0.001, floor can be seen about b = 60dB
when BD T = 0.0001, floor can be seen between b = 0 to 60dB
20. Data rate = 40 Kbps
Since DQPSK has 2 bits per symbol. Ts =
2
40103
= 5 105 sec
DQPSK
2
Gaussian Doppler power spectrum, c = e(BD T )
BD = 80Hz
Rician fading K = 2
c = 0.9998
s
"
#
"
#
(c / 2)2
(2 2)K/2
1
1
exp
= 2.138 105
P f loor =
2
1 (c / 2)2
1 c / 2
21. ISI:
Formula based approach:
Pf loor =
Tm
Ts
Since its Rayleigh fading, we can assume that Tm Tm = 100ns

Pf loor 104
which gives us
Tm 2
104
Ts
T
Ts p m = 10sec
Pf loor
So, Ts 10s. Tb 5s. Rb 200 Kbps.
Thumb - Rule approach:

t = 100 nsec will determine ISI. As long as Ts T , ISI will be negligible. Let Ts 10 T . Then
2bits 1 symbols
R symbol
= 2Mbps
Ts
sec
Doppler:
BD = 80 Hz
v
u 2
u c / 2
1
1 t
2
2
1 c / 2
Pf loor = 104
c 0.9999
But c for uniform scattering is J0 (2BD Ts ), so
c = J0 (2BD Ts ) = 1 (fD Ts )2 0.9999
Ts 39.79s
Tb 19.89s. Rb 50.26 Kbps.
Combining the two, we have 50.26 Kbps Rb 200 Kbps (or 2 Mbps).
22. From figure 6.5
with Pb = 103 ,
BPSK
d = Tm /Ts ,
Tm = 3s
d = 5 102
Ts = 60sec
R = 1/Ts = 16.7Kbps
QPSK
d = 4 102
Ts = 75sec
R = 2/Ts = 26.7Kbps
MSK
d = 3 102
Ts = 100sec
R = 2/Ts = 20Kbps
Chapter 7
1. Ps = 103
QPSK, Ps = 2Q( s ) 103 , s 0 = 10.8276.

Pout (0 ) =
M
Y
1e
i=1
1 = 10, 2 = 31.6228, 3 = 100.

M = 1

0
Pout = 1 e 1 = 0.6613
M = 2

0
0
1 e 2 = 0.1917
Pout = 1 e 1
M = 3

0
0
0
Pout = 1 e 1
1 e 2
1 e 3 = 0.0197
/ M 1 e/
2. p () = M
1e
= 10dB = 10
as we increase M, the mass in the pdf keeps on shifting to higher values of and so we have higher
values of and hence lower probability of error.
MATLAB CODE
gamma = [0:.1:60];
gamma_bar = 10;
M = [1 2 4 8 10];
fori=1:length(M)
pgamma(i,:) = (M(i)/gamma_bar)*(1-exp(-gamma/gamma_bar)).^...
(M(i)-1).*(exp(-gamma/gamma_bar));
end
3.
Z
Pb =
=
=
=
=
1
e p ()d
2
0
Z
iM 1
1 M h
/
e
1e
e/ d
2
0
Z
iM 1
M (1+1/) h
e
1 e/
d
2 0
M 1
M X
M 1
(1)n e(1+1/) d
n
2
n=0
M
1
1
M X
M 1
(1)n
= desired expression
n
2
1+n+
n=0
0.1
M=1
M=2
M=4
M=8
M = 10
0.09
0.08
0.07
p ()
0.06
0.05
0.04
0.03
0.02
0.01
10
20
30
40
50
60
Figure 1: Problem 2
4.
p () =
P r{2 < , 1 < }

<
P r{ 1 } + P r{2 < , 1 < } >
If the distribution is iid this reduces to
P1 ()P2 ( )
<
p () =
P r{ 1 } + P1 ()P2 ( ) >
5.
Z
Pb =
1
e p ()d
2
1 eT / 1 er /
p () =
2 eT / 1 er /
Pb =
=
< T
> T
Z T
Z
1
1
/
T /
r /
e/ e d
e
e d +
1e
2e
2
2
0
T
1
T T /
T /
+e
e
1e
2( + 1)
6.
Pb
1
2(+1)
no diversity
SC(M=2)
SSC
M 1
m
7. See
MATLAB CODE:
gammab_dB = [0:.1:20];
gammab = 10.^(gammab_dB/10);
M= 2;
P b (20dB)
0.0050
0.0076
0.0129
9.7 105
2.7 104
M PM 1
m
2 m=0 (1)
1+m+
1
T / + eT eT /
1
e
2(+1)
As SNR increases SSC approaches SC
P b (10dB)
0.0455
10
M=2
M=3
M=4
1
10
Pb,avg (DPSK)
10
10
10
10
10
10
10
12
14
16
18
20
avg
Figure 2: Problem 7
for j = 1:length(gammab)
Pbs(j) = 0
for m = 0:M-1
f = factorial(M-1)/(factorial(m)*factorial(M-1-m));
Pbs(j) = Pbs(j) + (M/2)*((-1)^m)*f*(1/(1+m+gammab(j)));
end
end
semilogy(gammab_dB,Pbs,b--)
hold on
M = 3;
Pbs(j) = 0
for m = 0:M-1
end
end
semilogy(gammab_dB,Pbs,b-.);
hold on
M = 4;
Pbs(j) = 0
for m = 0:M-1
end
end
semilogy(gammab_dB,Pbs,b:);
hold on
8.
1
=
N0
P
M
i=1 ai i
PM 2
i=1 ai
2
1
N0
a2i
P
P
a2i
i2
P
=
i2
N0
Where the inequality above follows from Cauchy-Schwartz condition. Equality holds if ai = ci where
c is a constant
9. (a) i = 10 dB = 10, 1 i N
N = 1, = 10, M = 4
Pb = .2e
1.5 (M1)
= .2e15/3 = 0.0013.
(b) In MRC, = 1 + 2 + . . . + N .
So = 10N
Pb = .2e
1.5 (M 1)
= .2e5N 106
N 2.4412
So, take N = 3, Pb = 6.12 108 106 .
10. Denote N (x) =
2
1 ex /2
2
, Q0 (x) = N (x)
Z
Pb =
Q() = 0,
Z
2
0
p
Q( 2)dP ()
P (0) = 0
p
p
1
d
2
1
Q( 2) = N ( 2) = e
d
2
2
2
Z
1
1
e P ()d
Pb =
2
2
0
M
X
(/)k1
P () = 1 e/
(k 1)!
k=1
1
1
1
e d =
1
2
2
2
0
"
k1 #
Z
M
M
X
1
1 1 / X (/)k1
1
1+ 1
1/2
e
d =
e
d
e
2
(k 1)!
(k 1)! 2 0
2
k=1
k=1
1/2
1
Denote A = 1 +
m
Z
M
1
X
1 1
/A2 1/2
=
e
d
m! 2 pi 0
m=0
let /A2 = u
2 m
Z
1/2
uA
1 1
u u
A2 du
=
e
m! 2 pi 0
A
m=0
M
1
X
A
2m 1 A2m A
=
+
m
2
22m m
m=1
M
1
X
1A
2m 1 A2m+1
=
m
2
22m m
M
1
X
Pb
m=1
11.
1
2
DenoteN (x) = ex /2
2
Q0 (x) = [1 (x)]0 = N (x)

Z p
Z
Pb =
Q( 2)dP () =
0
Z
0
1
1
e d =
2
2
0
Z
1
1
e e2/ d =
2
2
0
r
r
1 1
/
2
e
e
1 2Q
d =
2
2
1
1
e P ()d
2
2

1
1
1

=
2
2
1
q
2 1 + 2
1
1
2 B A
(1)
(2)
(3)
1
2
where A = 1 + , B = 1 +
s
"
#
1
1
overall P b =
1 1
2
(1 + )2
12.
1
2
no diversity
two
two
two
two
branch
branch
branch
branch
SC
SSC
EGC
MRC
R
R
R
R
Pq
b
h
i
b
1 1+
b
Q( 2)p d
Q( 2)p d
Q( 2)p d
Q( 2)p d
P b (10dB)
P b (20dB)
0.0233
0.0025
0.0030
0.0057
0.0021
0.0016
3.67 105
1.186 104
2.45 105
0.84 105
As the branch SNR increases the performance of all diversity combining schemes approaches the same.
MATLAB CODE:
gammatv = [.01:.1:10];
gammab = 100;
gamma = [0:.01:50*gammab];
for i = 1:length(gammatv)
gammat = gammatv(i);
gamma1 = [0:.01:gammat];
gamma2 = [gammat+.01:.01:50*gammab];
tointeg1 = Q(sqrt(2*gamma1)).*((1/gammab)*(1-exp(-gammat/gammab)).*exp(-gamma1/gammab));
tointeg2 = Q(sqrt(2*gamma2)).*((1/gammab)*(2-exp(-gammat/gammab)).*exp(-gamma2/gammab));
anssum(i) = sum(tointeg1)*.01+sum(tointeg2)*.01;
end
13. gammab_dB = [10];
Gamma=sqrt(gammab./(gammab+1));
pb_mrc =(((1-Gamma)/2).^2).*(((1+Gamma)/2).^0+2*((1+Gamma)/2).^1);
pb_egc = .5*(1-sqrt(1-(1./(1+gammab)).^2));
10
MRC
EGC
dB penalty ~ .5 dB
Pb()
10
10
10
10
10
12
14
16
18
20
14. 103 = Pb = Q( 2b ) 4.75, = 10

P
(0 /)k1
MRC Pout = 1 e0 / M
k=1 (k1)! = 0.0827
ECG Pout = 1 e2R R eR (1 2Q( 2R )) = 0.1041 > Pout,M RC

15. P b,M RC = 0.0016 < 0.0021P b,EGC
16. If each branch has = 10dB Rayleigh
/(/2)
2
= overall recvd SNR = 1 +
e(/2)2 0
2
BPSK
Z p
Pb =
Q( 2)p d = 0.0055
0
17. p() where 0 p()ex d =

we will use MGF approach
1
Pb =
0.01
/2
2i=1 Mi
1
sin2
d
=
=
18.
Pb =
Nakagami-2 fading
1
2
Z
1 /2
(0.01 sin )2 d
0
(0.01)2
= 0.0025
4
3 X
2
1+ m
l+m
;
m
2
m=0
r
=
1+

2
1
= 1+
sin2
2 sin2
3
Z
1 /2
1
Pb =
M 2
d, = 101.5 = 5.12 109
0
sin
MATLAB CODE:
gammab = 10^(1.5);
Gamma = sqrt(gammab./(gammab+1));
sumf = 0;
for m = 0:2
f = factorial(2+m)/(factorial(2)*factorial(m));
sumf = sumf+f*((1+Gamma)/2)^m;
end
pb_rayleigh = ((1-Gamma)/2)^3*sumf;
phi = [0.001:.001:pi/2];
sumvec = (1+(gammab./(2*(sin(phi).^2)))).^(-6);
pb_nakagami = (1/pi)*sum(sumvec)*.001;
19.
1
Pb =
/2
1+
2 sin2
2
1+
sin2
gammab_dB = [5:.1:20];
gammabvec = 10.^(gammab_dB/10);
for i = 1:length(gammabvec)
gammab = gammabvec(i);
phi = [0.001:.001:pi/2];
sumvec = ((1+(gammab./(2*(sin(phi).^2)))).^(-2)).*((1+...
(gammab./(1*(sin(phi).^2)))).^(-1));
pb_nakagami(i) = (1/pi)*sum(sumvec)*.001;
end
2
10
10
Pbavg
10
10
10
10
10
15
20
avg (dB)
20.

2 p
Pb = Q
2b (3) sin
3
8

= 2/3, g = 3 sin2
8

1
g
g
M 2
= 1+
sin
sin2
M
Z
/2
g
Pb =
1+
d
0
sin2
MATLAB CODE:
M = [1 2 4 8];
alpha = 2/3; g = 3*sin(pi/8)^2;
gammab_dB = [5:.1:20];
for k = 1:length(M)
phi = [0.001:.001:pi/2];
sumvec = ((1+((g*gammab)./(1*(sin(phi).^2)))).^(-M(k)));
pb_nakagami(k,i) = (alpha/pi)*sum(sumvec)*.001;
end
end
0
10
Pb
avg
10
10
10
15
10
10
15
avg (dB)
20
21.
Q(z) =
Q2 (z) =
Ps (s ) =
z2
exp 2
d , z > 0
sin
0
Z /4
z2
exp
d , z > 0
2 sin2
0
Z /2
1
gs
1
exp 2
d
sin
M
0
Z /4
1 2
gs
1
d
exp 2
sin
M
0
Z
Z
Ps =
=
But = 1 + 2 + . . . + M = i
=
22. Rayleigh: Ms (s) = (1 s s )1
Rician: Ms (s) =
MPSK
1+k
1+ks s
exp
ks s
1+ks s
Z
Ps =
/2
Ps ( )p ( )d
Z /2 Z
1
g
1
exp
p ()d d
sin2
M
0
0
2 Z /4 Z
1
g
1
exp
p ()d d
sin2
M
0
0
Z /2
1
g
M
1
d
i=1 Mi 2
sin
M
0
2 Z /4
g
4
1
M
1
i=1 Mi 2
d
sin
M
0
4
(M 1)/M
M s
g
sin2
d no diversity
Three branch diversity
g = sin
Ps =
16
(M 1)/M
1+
g
sin2
2
(1 + k) sin2
k s g
exp
d
(1 + k) sin2 + g s
(1 + k) sin2 + g s
= 0.1670
MQAM:
Formula derived in previous problem with g =
P s = 0.0553
MATLAB CODE:
gammab_dB = 10;
K = 2;
1.5
161
1.5
15
g = sin(pi/16)^2;
phi = [0.001:.001:pi*(15/16)];
sumvec=((1+((g*gammab)./(sin(phi).^2))).^(-1)).*((((...
(1+K)*sin(phi).^2)./((1+K)*sin(phi).^2+...
g*gammab)).*exp(-(K*gammab*g)./((1+K)*sin(phi).^2+g*gammab))).^2);
pb_mrc_psk = (1/pi)*sum(sumvec)*.001;
g = 1.5/(16-1);
phi1 = [0.001:.001:pi/2];
phi2 = [0.001:.001:pi/4];
sumvec1=((1+((g*gammab)./(sin(phi1).^2))).^...
(-1)).*(((((1+K)*sin(phi1).^2)./((1+K)*...
sin(phi1).^2+g*gammab)).*exp(-(K*gammab*g)./((...
1+K)*sin(phi1).^2+g*gammab))).^2);
sumvec2=((1+((g*gammab)./(sin(phi2).^2))).^(-1)).*((((...
(1+K)*sin(phi2).^2)./((1+K)*sin(phi2).^2+...
g*gammab)).*exp(-(K*gammab*g)./((1+K)*sin(phi2).^2+g*gammab))).^2);
pb_mrc_qam = (4/pi)*(1-(1/sqrt(16)))*sum(sumvec1)*.001 - ...
(4/pi)*(1-(1/sqrt(16)))^2*sum(sumvec2)*.001;
0
10
10
10
10
Ps
avg
10
10
10
10
10
10
10
15
23. MATLAB CODE:
M = [1 2 4 8];
alpha = 2/3;
g = 1.5/(16-1);
gammab_dB = [5:.1:20];
for k = 1:length(M)
phi1 = [0.001:.001:pi/2];
20
phi2 = [0.001:.001:pi/4];
sumvec1 = ((1+((g*gammab)./(1*(sin(phi1).^2)))).^(-M(k)));
sumvec2 = ((1+((g*gammab)./(1*(sin(phi2).^2)))).^(-M(k)));
pb_mrc_qam(k,i) = (4/pi)*(1-(1/sqrt(16)))*sum(sumvec1)*.001 - ...
(4/pi)*(1-(1/sqrt(16)))^2*sum(sumvec2)*.001;
end
end
Chapter 8
1. (3,1) code
(a) parity bit indicates if number of 1s is even or odd
even number of 1s parity = 0
odd number of 1s parity = 1
0 0 0
0 0 1
0 1 0
0 1 1
1 0 0
1 0 1
1 1 0
1 1 1
0
1
1
0
1
0
0
1
(b) dmin = 2
2. (a) see Matlab
(b) see Matlab
(c)
7
X
ci gi = 0
i=1
gi = ith column of G
c2 + c4 + c5 = 0
c1 + c3 + c5 = 0
c2 + c3 + c6 = 0
c1 + c2 + c7 = 0
c1
c2
c3
c4
c5
c6
c7
=1
=1
=1
=1
=0
=0
=0
c1
c2
c3
c4
c5
c6
c7
=0
=1
=0
=1
=0
=1
=1
1
1
T
H =
1
0
0
0
0
1
0
1
0
1
1
c1
c2
c3
c4
c5
c6
c7
=0
=1
=1
=0
=1
=0
=1
0
1
0
1
(d) R= 1 1 0 1 0 1 1
S = RH T = 1 0 0
(e) see Matlab
(f) A row reduced echelon form can be obtained by row operations and column permutations. Clearly
we get that as
1 0 0 0 0 1 1
0 1 0 0 1 1 0
0 0 1 0 1 0 1
0 0 0 1 1 1 1
See Figure 1
c4
c0=u0
u
u3
u2
u1
u0
c1=u1
c2=u2
c5
c3=u3
c6
Figure 1: Problem 2f
MATLAB
G = [0 1 0 1 1 0 0; 1 0 1 0 1 0 0 ; 0 1 1 0 0 1 0; 1 1 0 0 0 0 1];
b = 0:15;
a = dec2bin(b,4);
for i = 1:length(b)
u = a(i,:);
c(i,:) = zeros(1,length(G));
for j = 1:length(u)
h(j,:) = G(j,:)*u(j);
c(i,:) = mod(c(i,:)+h(j,:),2);
end
end
% c =
%
%
%
%
%
%
%
0
1
0
1
1
0
0
1
1
0
0
1
0
0
1
1
1
1
0
0
0
0
0
0
0
0
0
0
1
1
0
0
1
1
0
0
0
1
0
1
0
1
%
%
%
%
%
%
%
%
%
%
1
0
0
1
0
1
1
0
1
0
1
0
1
0
0
1
1
0
0
1
0
0
0
0
1
1
1
1
0
0
0
0
1
1
1
1
1
1
1
1
1
1
1
1
1
1
0
0
0
0
1
1
0
0
1
1
0
0
1
1
0
1
0
1
0
1
0
1
0
1
0
1
0
0
1
1
d = c(2:15,:); dmin = min(sum(d));

% dmin =
%
%
3
% H =
%
%
%
%
1
0
0
1
1
1
1
0
1
1
1
0
0
0
1
R = [1 1 0 1 0 1 1]; S = mod((R*H),2)
% S =
%
%
count = 1; for i = 1:length(c)

d = c(i,:);
if sum(d) == dmin
e(count,:) = c(i,:);
count = count+1;
end
end
% e =
%
%
%
%
%
%
%
%
1
0
1
0
0
0
1
1
1
0
0
1
0
0
0
1
1
0
0
1
0
0
0
0
0
1
1
1
3. t = 1 , 1 error can be corrected.

dmin 1 = 2 errors can be detected
0
0
1
1
1
0
0
0
1
0
1
0
0
1
1
0
0
1
0
1
0
4.
g(x) = 1 + x + x4
c1 (x) = 1 + x + x3 + x7 c1 (x)/g(x) = 1 + x3 , remainder = 0 valid
c2 (x) = 1 + x3 + x5 + x6 c2 (x)/g(x) = 1 + x + x2 , remainder = x4 not valid
P
i
u(x) = 10
i=0 ci x
4
x u(x)
p(x)
g(x) = q(x) + g(x) (degree of p(x) should be less than 4)
say u(x) = 1 (for example)
for a given u(x), the systematic form of codeword is generated in three steps
1 u(x)xnk = 1.x1511 = x4
x4
p(x) = 1 + x
2 1+x+x
4 q(x) = 1,
4
3 c(x) = x u(x) + p(x) = 1 + x + x4 (c(x) = codeword in systematic form)
5. (a)
1
0
G=
0
0
in systematic form
0
1
0
0
1
0
1
0
1
1
0
1
0
1
1
0
0
0
0
1
0
0
1
1
Gsys
1
0
=
0
0
H37
1 1 1 0 1 0 0
= 0 1 1 1 0 1 0
1 1 0 1 0 0 1
(b)
0
1
0
0
0
0
1
0
0
0
0
1
1
1
1
0
0
1
1
1
1
1
0
1
(c) Crec = 1 0 1 0 0 1 1
S = CH T = 0 0 0 another valid codeword corresponding to input [1 0 1 0]
(d) All possible received codewords to give error 0 are all possible codewords.
h
i
P
n1
n+1
1
(1 + x)n + n(1 + x) 2 (1 x) 2
6. N (x) = ni=0 Ni xi = n+1
Ni =number of codewords of weight i
(a) (7,4)
n=7
code7 block length
1
3 (1 x)4 = 1 + 7x3 + 7x4 + x7
(1
+
x)
+
n(1
+
x)
8
number of codewords of weight 0 = 1
number of codewords of weight 7 =1
(b) r =10dB
p = Eb /N0 Ec /N0 = 40/7
p = Q( 80/7) = 3.6162 104 From 8.38:
P
Pe 1j=2 6 [4p(1 p)]wj /2 = 3.9951 104
7. Hamming code, m=7
n = 2m 1 = 127
k = 2m 1 m = 120
dmin = 3
Pe = Pb (2k 1) [4p(1 p)]dmin /2 1
p = 2Q
r
!

Es
2s sin
= 2Q
2
sin
= 0.0962
M
N0
8
For uncoded modulation

2
p= Q
3
!

Eb
2 3 sin
= 0.0870
N0
8
for this sceme, unmodulated transmission is better.

8. There is negative coding gain below 2.8dB
snr_dB = 0:.1:20;
snr = 10.^(snr_dB/10);
k = 2;
for i = 1:length(snr)
snr_c = (2/5)*snr(i);
p = .5*erfc(sqrt(2*snr_c)/sqrt(2));
Pe(i) = (2^k-1)*(4*p*(1-p))^(3/2);
Pe_uncoded(i) = .5*erfc(sqrt(2*snr(i))/sqrt(2));
end
plot(snr_dB,Pe)
hold on;
plot(snr_dB,Pe_uncoded,r)
1.5
Coded
Uncoded
0.5
10
12
14
16
18
20
Figure 2: Problem 8
9. (7,4) Hamming code
Gc = 10 log10 [Rc dmin k ln (2/b )] = 2.1128dB
Rc = 4/7
dmin = 3
k=4
b = 15dB = 101 .5
10. At high SNR, the performance is almost the same.
MATLAB
clear;
snr_dB = 0:.1:10;
snr = 10.^(snr_dB/10);
k = 12;
n = 24;
dmin = 8;
for i = 1:length(snr)
snr_c = (k/n)*snr(i);
p = .5*erfc(sqrt(2*snr_c)/sqrt(2));
Pe_hdd(i) = (2^k-1)*(4*p*(1-p))^(dmin/2);
Pe_sdd(i) = (2^k-1)*.5*erfc(sqrt(2*snr(i)*(k/n)*dmin)/sqrt(2));
Pe_uncoded(i) = .5*erfc(sqrt(2*snr(i))/sqrt(2));
end
plot(snr_dB,Pe_hdd,b--)
hold on;
plot(snr_dB,Pe_sdd,b:)
% plot(snr_dB,Pe_uncoded,b-.)
11. Extended
12), dmin = 8, t = 3, b = 10dB = 10
p Gray code
(24,
p
p
p=Q
2Ec /N0 == Q
2Rc Eb /N0 = Q
2 1/2 10 = 7.82 104
2.62 1010 Pe 1.073 107
350
HDD
SDD
300
250
200
150
100
50
12. k=3
K=4
8PSK modulation
N= 2k 1 = 7
t = b5(N K)c = b5(3)c = 1
dmin = N K + 1 = 4
Ps 3.49 1011
k1
Pb = 22k 1 Ps = 1.99 1011
10
13. Pe < 2k
2dmin 1
1 dmin
= 1.35 108
4Rc b
dmin
Eb
= 15dB Rc N
= 15dB Rc b = 15dB = 101.5
0
(23,12) code, dmin = 7
Ec
N0
14. Choose a (5,2) linear block code with dmin = 3 as in problem 8

Rc = 2/5
b = 10
dmin = 3
Now, from (8.81)
fading,
with Rayleigh
2d
1
d
min
min
1
Pe < 2k < 2k
= 0.0098 102 bit errors do not occur mostly and any depth
4Rc b
dmin
is ok as long as delay is less than 5msec.
ndTs 5msec
d 30
we use an interleaver with n columns and d rows.
15. U = [ 0 1 1 0 1 0 1 1 0 1]
C = 000 111 101 001 100 010 100 101 001 100
see Figure 4
0/000
00
00
0/011
1/111
01
01
1/110
0/010
10
10
1/101
0/001
11
1/110
11
16. (a) S =S2 S3
see Figure 3:16a
(b) p=103
Path metric = log (p(R/C))
2 X
3
X
=
log P (Rij /Cij )
i=0 j=1
= 3 log p + 5 log(1 p) = 20.73
0/000
S0 =00
0/011
1/110
S1 =01
1/101
0/110
S2 =10
1/000
0/101
1/011
S3 =11
(c) We consider the path that goes from S0 to S2 to S1 to S0 . This path is at the minimum hamming
distance to the all 0 path.
C1 = 110 110 011
Path metric = log (p(R/C))
2 X
3
X
=
log P (Rij /Cij )
i=0 j=1
= 5 log p + 4 log(1 p) = 34.54
17. (a) see figure 6

2
D
1/011
d=11
D
1
0/000
0/101
1/000
1
2
D
1/110
c=10
a=00
b=01
0/110
e=00
0/011
2
D
1/101

(b) see figure 7
Xc = JN D2 Xa + JN D2 Xb
Xb = JD2 Xc + JD2 Xd
Xd = JN Xc + JN D2 Xd
Xe = JD2 Xb
2
JN D
d=11
2
JN
JD
2
a=00
JD
JD
c=10
b=01
e=00
JN D
2
JN D

Solving we get
J(D, N, J) =
Xe
J 3 N D6 (1 JN D2 + JN )
=
Xa
(1 JN D2 ) J 2 N D4 (1 JN D2 + JN )
(c) dmin = 6
(d) b = 10
Rc = 1/3
T (D, N ) =
Pb <
N D6 (1N D2 +N )
(1N D2 )N D4 (1N D2 +N )
dT (D,N )
|N =1,D=eb Rc = 6.19
dN
109
(e) p = Q 2b Rc = 0.0049
)
Pb < dT (D,N
|N =1,D=4p(1p) = 2.2883 105
dN
eSDD
Coding gain = 10 log10 PPeHDD
= 35.678
18. Tc = 10ms
Rs = 50, 000sps
N Ts > Tc
N Tc /Ts = 500
delay = N(N-1)Ts 5sec
Memory associated = N (N21) = 1.2475 105
19. (a) see figure 8
dmin = 2d16QAN (for parallel transitions)

Gc =
d2min
2 = 1.365 > 1
d08P SK
(b) see figure 9

coded bits
d0,16QAM
0/00
S0 =00
0/11
1/11
S1 =01
coset 1
00
coset 2
01
coset 3
10
coset 4
11
1/00
0/01
S2 =10
1/10
0/10
S3 =11
1/01
00
dmin
10
01
uncoded bits
coset1
11

each branch corresponds to 4 parallel transitions
coded bits pick the coset
uncoded bits pick a point in the coset
(c) Minimum distance error event through the trellis would normally be S0 S2 S1 S0 .
However since parallel transitions are also a part of the trellis, minimum distance error event will
be another point of a coset being decoded which is at dmin
(d) dmin = 2d016QAM = 1.264
20. (a) For C1
n = 22 1 = 3
k = 22 1 2 = 1
dmin = 3
Pe < (4p(1 p))3/2
For C2
For C3
Pe < 4 (4p(1 p))3/2

Pe < 11 (4p(1 p))3/2
(b) Ts = 10s
RC1 = 1/3 Data rate = 3 105 bps RC2 = 4/7 Data rate = 1.75 105 bps RC3 = 11/15
Data rate = 1.36 105 bps
(c) (25,16) code
21. For low priority bits use (7,4) Hamming code with BPSK
Pb = 2.199 104
For high priority bits use (23,12) Golay code with BPSK
Pb = 9 109
Chapter 9
1.
1
Ps =
g s 1
1+
d
sin2
g = sin2
P s = P b log2 8 = 3P b = 3 10
= 1119
Rb
3Rs
=
=3
B
B
2. Pout = P ( < 0 ) = 1 e0 /100 = 0.1
0 = 10.536
1
= 1/E0 [1/] = R 1
= 56.3125 = 17.506dB
0 P ()
For 4-QAM, SNR =10.35dB
16-QAM, SNR = 17.35dB
Maximum constellation size is 16-QAM
3. For QPSK
Pb = Q

= 103
= 10.35dB = 10.85
Settin = 1/E0 [1/] and solving for 0
0 = 0.0056
Pout = 1 e0 /100 = 5.6 105
4. (a) M=4
Pb = Q( ) = 103
= 10.35dB
M=16
= 17.35dB
< 10.35dB
no transmission
10.35dB < < 17.35dB 4-QAM
> 17.35dB
16-QAM
0 = 10.35dB
(b)
P (4QAM ) = P (10.35dB < < 17.35dB)
Z 17.35dB
1 /100
e
d
=
10.35dB 100
= 0.3164
P (16QAM ) = P ( > 17.35dB)
= e17.5dB/100
= 0.5809
Rb = 0.3164 2 + 0.5809 4 = 2.9564bps/Hz
(c)
R 10
Pe =
5. For BPSK
P ()Q( )d
R 0
= 0.0242
0 P ()d
p
Pb = Q( 2) = 103
= 4.77 = 6.78dB
For QPSK
Pb = Q( ) = 103
= 4.77 = 10.35dB
For 8PSK
Pb 0.67Q
2 sin(/8) = 103
= 14.78dB
> 14.78dB
10.35dB < < 14.78dB
6.78dB < < 10.35dB
3.78dB < < 6.78dB
2.78dB < < 3.78dB
1.78dB < < 2.78dB
< 1.78dB
no code 8PSK
no code QPSK
no code BPSK
1st code BPSK
2nd code BPSK
3rd code BPSK
no transmission
For Rayleigh fading

R = P ( > 14.78dB).3 + P (10.35dB < < 14.78dB).2
+P (6.78dB < < 10.35dB).1 + P (3.78dB < < 6.78dB)
1
1
+P (2.78dB < < 3.78dB) + P (1.78dB < < 2.78dB)
3
4
= 2.6061bps/Hz
6.
Z
J=
Z
KS()
log2 1 +
P ()d S()P ()d
S
J
=0
S()
1
KS()
S
1+
1
S()
0
=
0
S
K
=0
S
1
K
0 /K
< 0 /K
S()
1
1
K
M () = 1 + K
= 1 + K
=
0 K
0
S
M () = /K
log2 M () = log2 (/K )
7. (a) P e =
R
0
1 / 1.5/(M 1)
2e
d
e
1
2
(b) Using (a)

Numerically, M=2
BP SK
Specral efficiency = 1bps/Hz
P e = 0.025
(c) In Fig 9.3 , Average SNR = 20dB

Spectral efficiency = 4bps/Hz
Adaptive modulation has higher spectral efficiency.
8. (a)
Pb .2e
M () = 1 +
1.5
(M
1)
S()
S
1.5 S()
S()
= 1 + K
ln(5Pb ) S
S
where
K=
1.5
ln(5Pb )
We maximize spectral efficiency by maximizing:
Z
S()
E[log2 M ()] = log2 1 + K
p()d
s
subject to:
Z
S()p()d = S
This gives the water-filling solution for optimal power adaptation as:
1
0
1
S()
0 K K
=
0
< K0
S
This gives that the optimal rate adaptation is
M () =
,
K
where K = 0 /K.
To find the cut-off, we use the average power constraint equation as
Z
1
1
1 1
e d = 1
K

K K
For = 20 dB = 100 and Pb = 104 , Using Matlab we get K = 0.1973, K = 4.3, 0 = 0.8486.
(b)
R
= E[log2 M ()] =
B
S()
1
p()d =
log2
log2 1 + K
e d
K
S
K
Using Matlab we get 3.7681 bits/sec/Hz

(c)
M () = 1 + K
S()
S
For truncated channel inversion, the power adaptation is given as

S()
0
=
0 o.w.
S
where
=
1
= R
E0 [1/]
0
1 1
d
e
The value of 0 is chosen so as to maximize spectral efficiency i.e.

R
= max log2 (1 + K)p( > 0 )
0
B
1
= max log2 1 + K
p( > 0 )
0
E0 [1/]
Using Matlab we get,
R
B
= 3.3628 at 0 = 23.18 and pout = 0.2069.
9. Assume a target BER of 103

(a)
S()
=
S
1
0
1
K
0 /K
< 0 /K
Try 4 channels
0 =
1
K
1
1
i P (i )
+1
= 0.6491
1.5
K =
= 0.2831
ln(5Pb )
= 0 /K = 2.2928
1 = 100.5 > 0.6491
we should use 4 channels

optimal power and rate adaptation
1
1
S()
=
0.6491
0.2831
S
Instant Rate
log2 M () = log2 (/2.2928)
(b) Average spectral density
Pi log2 (/2.2928) = 2.457bps/Hz
(c) Truncated channel inversion

First we use 4 channels
4
X
1
Pi = 0.1530
i
i=1
R
K
0.2831
= log2 1 +
= log2 1 +
= 1.5111
B
E(1/)
0.1530
E (1/) =
Second, we use 3 channels

4
X
1
E (1/) =
Pi = 0.0265
i
i=1
R
K
= log2 1 +
0.6 = 2.1278
B
E(1/)
Now, we use 2 channels
4
X
1
Pi = 0.0065
i
i=1
R
K
= log2 1 +
0.4 = 2.1910
B
E(1/)
E (1/) =
Finally, we use 1 channel
we use 2 channels.
Power control policy
R
K
= log2 1 +
0.2 = 2.0936
B
E(1/)
1
S()
153.84
=
=
,
E(1/)
= 15dB, 20dB
R
= 2.191bps/Hz
B
This is less than waterfilling
10. = 100, Pb = 103
Constellation restricted to M = {0, 2, 4, 16, 64, 256}
R
1
= max log2 b 1 + K
cM p( > 0 )
0
B
E0 [1/]
Define
x= 1+K
1
E0 [1/]
Using Matlab:
For the unrestricted case, we get, (R/B)max = 3.76 at 0 = 20.25 and x(0 ) = 24.37. As for the
continuous case, the R/B curve is a strictly concave function of 0 , we know that the best choices of
M are either 16 or 64 as 16 x(0 ) = 24.37 64.
For x = 16, we get 0 = 9.4 and corresponding (R/B) as 3.64
For x = 64, we get 0 = 60.5 and corresponding (R/B) as 3.27.
So, we choose M = 16 and corresponding spectral efficiency is 3.36 bits/sec/Hz.
11. (a)
< 20
20 < < 40
40 < < 160
160 < < 640
640 < <
From table 9.2,
= 20dB, fD = 100Hz
no transmission
M=2
M=4
M=16
M=64
fc
D = 80Hz is 4/5 of fD = 100Hz
D
b = fc
= 54 j
fD
= [0.737 0.301 1.06 2.28 3.84]5/4 ms

= [0.9213 0.3762 1.3250 2.85 1.7875] ms
number of symbols = /Ts

Spectral efficiency:
R
B
= P (20 < < 40)0.1 + P (40 < < 160)0.2

+P (160 < < 640)0.4 + P (640 < < )0.8
= 1.8995bps/Hz
(b)
Pb =
s
!Z ?
p
K Mj
2( Mj 1)
3(Mj 1)
p
Q
P ()d = 2.25 104
? M 1
K(Mj 1)
Mj log2 Mj
K
j
N
1
X
j=1
the exact BER is better than the target BER

12. (a) = 100
B = 30 KHz
fD = 80 Hz
Pb = 103
=
Z
Pb =
1
1
(5Pb ) p()d =
5
1.5
0.5
1
1
(5Pb ) p()d = 0.0017
5
where the last integral was evaluated using Matlab.

Another, more analytical way to solve this is a method proposed by William Wu, a student enrolled
for the course as:
Z
Pb u .2
1.5
0.5
(5Pb )1/ d
This integral cannot be evaluated in closed form, but it can be expressed in terms of a special
function; namely, the exponential integral.
RK
Consider the integral S = K12 a1/x dx, where a is a real positive number. Then a = ep for some
p R. Rewriting the integral in terms of p, and then integrating by parts, we have
Z
K2
S=
a
K1
1/x
K2
dx =
p/x
dx =
K1
p/x
K2 Z
x
K1
K2
xep/x
K1
p
dx
x2
K2 Z K2 p
p/x
e x +
ep/x dx.
x
K1
| K1 {z
}
With regards to the second integral , let u = p/x. Then x = up , and dx =
p
du,
u2
yielding
K4
(u)e
p
du = p
u2
K3
K4
K3
eu
du
u
p
p
where K3 = K
and K4 = K
. Applying the data provided in this problem, p u 5.298, K3 u
1
2
10.596, and K4 u 3.532. Reversing the orientation of integration yields
K3
eu
du = p(E1 (K4 ) E1 (K3 ))
u
K4
R t
where E1 is the exponential integral function E1 (x) , x e t dt. Conclusively, using K1 =
0.5, K2 = 1.5, and Pb = 103 , we have
=p
K2
p/x
Pb u (1/5) e x + p(E1 (K4 ) E1 (K3 )) u 1.674 103

K1
where E1 (x) was evaluated using the expint() function in MATLAB.

(b)
(t) = (t ) + (t)
(t))
1.5(t) S(
Pb ((t), (t)) 0.2exp
M (
(t)) 1 S
(t)
= 0.2(5Pb ) (t)
= = 0.2(5Pb )
where
=
1
+
(1)
(2)
(3)
(t )
(t)
(t)
(t)
IF the fading process is stationary t does not matter. In general (t ) and (t) are not
independent of (t).
In further calculation we drop t as process stationary
Z Z
Pb
0 0.2(5Pb ) p(, )d
d
0
Notice that integrates only from 0 as the transmitter which has knowledge of only does not
transmit if is less than 0 , where as can vary over its entire range.
Z Z
Pb
0 0.2(5Pb ) p(
|)d
p()d
0
where we have used Bayes rule to relate the joint distribution to the marginal distribution as
p(, ) = p(
|)p()
Now using
1
=
we get
Z
Pb
0 0.2(5Pb )
1
+
p (( + )|) d( + ) p()d
where if {0 , } and {0, }, then ( + ) = {0, }.

Now IF, (t )| and | are independent then we can write
p(( + )|) = p(|) p(|)
Further, IF (t )| only depends on p( ), we only need the distribution of p( ) and p( |).
13. Suppose target Pe = 103
Kc = KGc = 3dB 0.2831 = 0.5662
1
1
S()
0 Kc 0 /Kc
=
0
< 0 /Kc
S
Z
S()
= 1 0 = 0.93
S
1
1
S()
0.93 0.5562 1.67
=
0
< 1.67
S
Z
R
p()d = 5.1bps/Hz
=
log2
B
Kc
Kc
14.
Pb
4
Q
log2 M
4
q
log2 M M6
1
C1 =
!
3
M 1
3
exp
2(M 1)
4
q
log2 M M6
1
C2 = 1.5
C3 = 1
C4 = 1
15.
S()
=
K()p()d +
P ()d
S
0
0
Z
S()
1
C2 S()
p()d +
=
log2 C4
p()d
C3
ln(b /C1 ) S
S
0
Z
1/ ln 2
J
C2
0
+
S()
C
2
S()
C4 ln(pb /C1 ) S ln(Pb /C1 )S S
(
1
3(ln12)S K
S() 0, K() 0
S()
=
0
else
S
16. (a) = 30 dB = 1000

Pb = 107
c1
c2
c3
1st Bound 0.05 6
1.9
2nd Bound 0.2
7
1.9
3rd Bound 0.25 8 1.94
We use eq. 9.46 in the reader
c4
1
-1
0

1
c4
Pb
= ln
K
c2
c1
c2

K=
c4 ln Pc1b
For the 1st bound:

K = 0.4572
S()
=
S
1
0 K
1
K
> 0
o.w.
(4)
K is known to be 0.4572, 0 is found using the constraint
Z
1
1
p()d = 1
0 K
K
0
Using Matlab, we get 0 = 2.16
For the 2nd bound:
K = -0.4825
S()
=
S
01K
0
1
K
> 0
o.w.
(5)
K is known to be -0.4825, 0 is found using the constraint
Z
1
1
p()d = 1
0 K
K
0
For the 3rd bound:
K =
S()
=
S
K0 is
ln(Pb /c1 )
c2
K0
0
> 0
o.w.
= 1.8415, 0 is found using the constraint
Z
K0
p()d = 1
0
0
(6)
S(r) / Avg(S) vs r
2
Bound 1
Bound 2
Bound 3
1.8
1.6
1.4
S(r) / Avg(S)
1.2
0.8
0.6
0.4
0.2
10
15
r (dB)
20
25
30
Figure 1: Problem 16, S()/S

k(r) vs r
5
Bound 1
Bound 2
Bound 3
4.5
3.5
k(r) (bps/Hz)
2.5
1.5
0.5
10
15
r (dB)
20
25
30
Figure 2: Problem 16, k()

(b)
(
k() =
1
c3

log2
0
o.w.
(7)
The plots for S()/S and k() are given below:

For bound 1, the cutoff is dictated by the power constraint, where as for bounds 2 and 3, where
K is either negative or infinite, the cutoff is dictated by the rate positivity constraint.
(c) By looking at the curves we notice that for Low SNR
1st Bound: Water-filling
2nd Bound: Channel-Inversion
3rd Bound: Constant Power
For high SNRs, all the three bounds have the same power adaptation.
The rates are almost the same for all the three bounds.
17. M = {M0 = 0, . . . MN 1 } where M0 is no data transmission
kj = log2 Mj , j > 0
k0 = 0
kj assigned to Rj = [j 1, j ) , j = 0, . . . N 1
1 = 0
N 1 =
"
Pb () c1 exp
Setting target value Pb = P ()
ln
L=
kj
L
j
=0
c1 S()
S
2c3 kj c4
c4 )
j=1
N
1 Z j
X
p()d +
j1
2c3 kj c4
ln(Pb /c1 ) c3 kj
(2
c2
S()
=
S
N
1
X
Pb
c1
c1 S()
S
j=1
j1
h(kj )
h(kj )
p()d 1
0j N 2
L
h(k1 )
= k1 p(0 )
p(0 ) = 0
0
0
0 =
h(k1 )
k1
for some constant =
for j > 0
h(kj )
h(kj+1 )
L
= kj p(j ) kj+1 p(j ) +
p(j )
p(j ) = 0
0
j
j
j =
h(kj+1 ) h(kj )
1j N 2
kj+1 kj
For some = where is found from the average power constraint as:
N
1 Z j
X
j1
j=1
h(kj )
p()d = 1
18.
J=
N
1
X
j=1
Z
kj
j1
N
1
X
p()d + 1
Z
kj
j=1
j1
(Pb () P b )p()d + 2
J
=0
S()
Notice that Pb () is a function of S()
1
Pb ()
p() + 2 p() = 0 , j1 j
S()
S()p()d S
0
Pb ()
2
=
S()
kj 1
"
Pb () = c1 exp
c1 S()
S
2c3 kj c4
c2
f (kj )
Pb ()
2
= Pb () c3 k S =
Pb () =
j
S()
kj 1
kj
2
c4
where =
j1 j
S2
c2 1
f (kj )
c2 S()/S
= c1 exp
kj
f (kj )
f (kj )
c2 S()/S
= ln
f (kj )
c1 kj
where
Sj ()
S
h
i
f (k )
f (k )
= cj2 ln c1 kj j
S() = Sj () , j1 j
J
=0
j
0j N 2
(kj kj+1 ) + 1 [kj (Pb (j ) P b ) kj+1 (Pb (j ) P b )] + 2 [Sj (j ) Sj+1 (j )] = 0

Pb (j ) = P b
1
2 Sj+1 (j ) Sj (j )
1 1
(kj+1 kj )
Substituting for Pb (j ) and Sj (j ) we et an equation between j , 1 and 2

h
i
h
i
f (kj+1 )
f (kj+1 )
f (kj )
f (kj )
ln
ln
c1 j kj+1
j c2
c1 j kj
f (kj )
2 j c2
1
= Pb
j kj
1 1
(kj+1 kj )
2
where = cS
2 1
Notice that 1 and 2 are found using the constraints
N
1 Z j
X
j1
j=1
and
N
1
X
j=1
Z
kj
fi ()
p()d = 1
S
j1
(Pb (j ) P b )p()d = 0
19. = 20dB = 100

Pb = 104
c1 = 0.05, c2 = 6, c3 = 1.9, c4 = 1
M = {0, 2, 4, 8, 16}
1 = 0
(N 1) = 4 =
N = 5.
h(kj )
S()
=
,
S
where
h(kj ) =
j1 j , , j = 0, 1, 2, 3, 4
ln(Pb /c1 ) c3 kj
2
c4 , , j = 0, 1, 2, 3, 4
c2
kj = log2 Mj , j = 0, 1, 2, 3, 4
Region boundaries are 0 , 1 , 2 and 4 given by
0 =
j =
Find using:
h(k1 )
k1
h(kj+1 ) h(kj )
, j = 1, 2, 3
kj+1 kj
N
1 Z j
X
j=1
j1
h(kj )
p()d = 1
Using Matlab, we find that = 0.7

Once we know the region boundaries (as we know ), we find the spectral efficiency as:
R
B
N
1 Z j
X
j1
j=1
N
1
X
kj
j=1
N
1
X
kj p()d
p()d
kj p(j)
p(j) =
(9)
j1
j=1
where
(8)
p()d
j1
Using Matlab we find that R/B = 3.002 bits/sec/Hz.

20. From 9.84 and 9.85
S() = Sj () , j1 j
(10)
where
f (kj ) f (kj )
Sj ()
ln
c1 kj c2
S
Pbj () =
f (kj )
kj
, j = 1, . . . N 1
j1 j
,j = 1...N 1
Solving approximately
j1 =
Solve for and using
N
1 Z j
X
j=1
and
f (kj )
kj
N
1
X
j=1
Z
kj
j1
,j = 1...N 1
Sj ()
p()d = 1
S
j1
(Pbj () P b )p()d = 0
using Matlab we get one solution as

= 3 = 4 106
optimal region boundaries 0, 8.1964, 19.3932, 50.9842, 144.7588,
Pout = 0.0787 ASE = 2.5810 bits/sec/Hz
MATLAB
M = [0 2 4 8 16];
N = 5; k = log2(M);
k(1) = 0;
c1 = .05;
c2 = 6;
c3 = 1.9;
c4 = 1;
Pb_bar = 1e-4;
gamma_bar = 100;
f = (2.^(c3*k))-c4;
lambda1 = 1e4; ss = .1; ss1 = .0001; ss2 = .1; count1 = 1; for
lambda = ss1:ss1:ss1
count2 = 1;
for rho = 1:ss2:2
gamma_bnd = [f./k]*rho;
gamma_bnd(1) = 0;
gamma_bnd(6) = 100*gamma_bar;
for i = 1:N
a = [gamma_bnd(i):ss/10:gamma_bnd(i+1)];
loi(i) = length(a);
end
mloi = max(loi);
gamma = zeros(5,mloi);
for i = 1:N
gamma(i,:) = [a zeros(1,mloi-length(a))];
end
sum_power(count1,count2) = -1;
for i = 1:N-1
S_by_S_bar = -log((lambda*f(i+1))./(c1*gamma(i+1,1:loi(i))*k(...
i+1))).*(f(i+1)./(gamma(i+1,1:loi(i))*c2));
if min(S_by_S_bar)<0
stop;
end
p_gamma = (1/gamma_bar)*exp(-gamma(i+1,1:loi(i))/gamma_bar);
sum_int(i) = (ss/10)*sum(S_by_S_bar.*p_gamma);
sum_power(count1,count2) = sum_power(count1,count2)+sum_int(i);
end
sum_Pb(count1,count2) = 0;
for i = 1:N-1
Pb_gamma = repmat(Pb_bar-(1/lambda1), 1, length(p_gamma));
Pb_int(i) = (ss/10)*k(i+1)*sum((Pb_gamma-Pb_bar).*p_gamma);
sum_Pb(count1,count2) = sum_Pb(count1,count2)+Pb_int(i);
end
count2 = count2+1;
end
count1 = count1+1;
end
PART b
M = [0 2 4 8 16];
N = 5;
k = log2(M);
k(1) = 0;
c1 = .05;
c2 = 6;
c3 = 1.9;
c4 = 1;
Pb_bar = 1e-4;
gamma_bar = 100;
f = (2.^(c3*k))-c4;
lambda1 = 1e4; ss = .1;
count2 = 1;
for rho = 15:1:20
gamma_bnd = [f./k]*rho;
gamma_bnd(1) = 0;
gamma_bnd(6) = 100*gamma_bar;
for i = 1:N
loi(i) = length(a);
end
mloi = max(loi);
gamma = zeros(5,mloi);
for i = 1:N
gamma(i,:) = [a zeros(1,mloi-length(a))];
end
sum_Pb(count2) = 0;
for i = 1:N-1
Pb_gamma = repmat(Pb_bar-(1/lambda1), 1, length(p_gamma));
Pb_int(i) = (ss/10)*k(i+1)*sum((Pb_gamma-Pb_bar).*p_gamma);
sum_Pb(count2) = sum_Pb(count2)+Pb_int(i);
end
count2 = count2+1;
end
6
1.4
x 10
1.2
Pb()
0.8
0.6
0.4
0.2
0
2
10
10
10
10
10
(dB)
21. C1 = 0.05 C2 = 6 C3 = 1.9 C4 = 1
K=
Pb () =
C2
C1 /P b 1
c1
c2 S()S
2c3 k() c4
+1
10
10
2.5
S()/Sbar
1.5
0.5
0
2
10
10
10
10
10
10
(dB)
S()
=
S
1
0
C4
K
C4 0 /K
else
1
log2 (K/0 )
c3
Z
R
K
=
log2
p()d
B
0
C4 0 /K
k() =
Matlab gives
b = 0.92,
MATLAB CODE:
c1 = .05;
c2 = 6;
c3 = 1.9;
c4 = 1;
Pb_bar = 1e-3;
K = c2/((c1/Pb_bar)-1);
meu_psi_db = 20;
sigma_psi_db = 8;
sigma = 10^(sigma_psi_db/10);
meu = 10^(meu_psi_db/10);
Pout = 1016 ,
R/B = 1.964bits/sec/Hz
10
S = sqrt(log(exp(log(sigma^2)-2*log(meu))+1));
M = (2*log(meu)-S^2)/2;
gamma0_bar = [.9:.01:1];
ss1 = .1;
for k1 = 1:length(gamma0_bar)
a = [(c4/K)*gamma0_bar(k1):ss1:1000*meu];
b = (1./(S*sqrt(2*pi)*a)).*exp(-(log(a)-M).^2/(2*S^2));
S_by_S_bar = (1/gamma0_bar(k1)) - ((c4/K)./a);
sum_int(k1) = ss1*sum(S_by_S_bar.*b);
end
[m,n] = min(abs(sum_int-1));
gamma0_bar_chosen = gamma0_bar(n);
a = [(c4/K)*gamma0_bar_chosen:ss1:100*meu];
b = (1./(S*sqrt(2*pi)*a)).*exp(-(log(a)-M).^2/(2*S^2));
S_by_S_bar = (1/gamma0_bar_chosen) - ((c4/K)./a);
kgamma = (1/c3)*log2(K*(a/gamma0_bar_chosen));
ASE = sum(b.*kgamma)*ss1;
Pout = 1-(sum(b)*ss1);
22. Notice that from 9.91 to 9.97 remains unchanged
(9.97)
We maximize spectral efficiency by maximizing
Z
1
KS()
log2 c4 +
E[k()] =
p()d
c3
S
0
subject to
S()
p()d = 1
S
c2
c1 /P b 1
for c4 < 0, K =
we get the solution similar to 9.54 as
S()
=
S
1
0
c4
K
4
0 c
K
o.w.
where 0 is determined by power constraint

Notice that it is like channel inversion as c4 is negative
Also, optimal rate adaptation is
1
c4
k() =
log2
c3
0
k(bar)
10
20
30
40
50
60
70
bar
23. Similar to the previous problem, we get the optimal adaptation as
(
K0
0
S()
0
=
0
o.w.
S
where K0 = c1 /Pc2b 1 and 0 is found from power constraint.
Notice that it is an on-off power transmission scheme.
Optimal rate adaptation is given as

1
k() =
log2
c3
0
80
90
100
1.4
1.2
S(bar)/Sbar
0.8
0.6
0.4
0.2
10
20
30
40
50
60
bar
70
80
90
100
Chapter 10
1. (a)
(AAH )T
= (AH )T .AT
T
= (AT ) AT
= AAH
(AAH )H
For AAH ,
= AAH
= , i.e. eigen-values are real

AAH = QQH
(b) X H AAH X = (X H A)(X H A)H = kX H Ak 0

AAH is positive semidefinite.
(c) IM + AAH = IM + QQH = Q(I + )QH
AH positive semidefinite i 0i
1 + i > 0i
IM + AAH positive definite
(d)
det[IM + AAH ] = det[IM + QQH ]
= det[Q(IM + M )QH ]
= det[IM + M ]
Rank(A)
= i=1
(1 + i )
e Q
eH ]
det[IN + AH A] = det[IN + Q
e N + N )Q
eH ]
= det[Q(I
= det[IN + N ]
Rank(A)
= i=1
(1 + i )
AAH and AH A have the same eigen-value

det[IM + AAH ] = det[IN + AH A]
2. H = U V T
0.4793
0.8685 0.1298
U = 0.5896 0.4272 0.6855
0.6508 0.2513
0.7164
1.7034
0
0
0 0.7152
0
=
0
0 0.1302
0.3458
0.6849
0.4263
0.5708
0.2191
0.0708
V =
0.7116 0.6109
0.0145
0.2198
0.3311 0.9017
3. H = U V T
Let
1 0
U = 0 1
0 0
1 0
V = 0 1
0 0
1 0
0 2
1 0 0
H= 0 2 0
0 0 0
4. Check the rank of each matrix
rank(HI ) = 3
multiplexing gain = 3
rank(H2 ) = 4
multiplexing gain = 4
5.
C=
Constraint
Vi =
RH
X
i=1
i
log2 1 +
Mt
i = constant
1
1
=0
i
Mt ln 2 (1 + i ) Mt ln 2 (1 + i )
Mt
Mt
i = j
when all RH singular values are equal, this capacity is maximized.
6. (a) Any method
to
.13 .08
D = .05 .09
.23 .13
show
H U V is acceptable. For example:
.11
H H
.14 where : dij = ij
Hij 100
.10
(b) precoding filter M = V 1

shaping filter F = U 1
.5195 .3460 .7813

F = .0251 .9078 .4188
.8540 .2373
.4629
.2407 .8894 .3887

M = .4727 .2423 .8473
.8478 .3876
.3622
Thus Y = F(H)MX + FN = U U VV X + U N
= X + U N
(c)
Pi
P
1
1
1
1
o i for i > o , 0
2
P
i = Nio B
= 94.5 for i = 1,
Assume 2 > 0 > 3 since
else
6.86 for i = 2, .68 for i = 3
3 = .68 is clearly too small for data transmission
P Pi
= 1 20 11 12 = 1 0 = 1.73
P1
P2
P = .5676
P = .4324
C = B log2 1 + 1 PP1 + log2 1 + 2 PP2

= 775.9 kbps
P
(d) With equal weight beamforming, the beamforming vector is given by c = 1 [1 1 1]. The SNR
(3)
is then given by:

SN R =
cH H H Hc
= (.78)(100) = 78.
N0 B
(1)
This gives a capacity of 630.35 kbps. The SNR achieved with beamforming is smaller than the
best channel in part (c). If we had chosen c to equal the eigenvector corresponding to the best
eigenvalue, then the SNR with beamforming would be equal to the largest SNR in part(c). The
beamforming SNR for the given c is greater than the two smallest eigenvalues in part(c) because
the channel matrix has one large eigenvalue and two very small eigenvalues.
7. C = max B log2 det[IM + HRX H H ]
RX : T (RX ) = If the channel is known to the transmitter, it will perform an SVD decomposition of
H as
H = U V
HRX H H = (U V )RX (U V )H
By Hadamards inequality we have that for A <nn
det(A) ni=1 Aii
with equality iff A is diagonal.
We choose RX to be diagonal, say = then
det(IM R + HRX H H ) = det(I + 2 )
C = Pmax Bi log2 (1 + i i )
i
where i are the singular values.
8. The capacity of the channel is found by the decomposition of the channel into RH parallel channels,
where RH is the rank of the channel matric H.
X
C= P
max
B log2 (1 + i i )
i :
where i are the RH non-zero singular values of the channel matrix H and is the SNR constraint.
i = i
Then the optimal power allocation is given as
1
Pi
0
=
0
P
1
i
i 0
i < 0
for some cut-off value 0 . The resulting capacity is given as

X
C=
B log2 (i /0 )
i:i 0
(2)
For
1
1
H=
1
1
1 1
1
1 1 1
1
1
1
1
1 1
RH = 3, 1 = 80, 2 = 40, 3 = 40. We first assume that 0 is less than the minimum i which is 40.
0 =
1+
3
P3
1
i=1 i
which gives 0 = 2.8236 < mini i , hence the assumption was correct.
C
= 12.4732 bits/sec/Hz
B
For
1
1
1 1
1
1 1
1
H=
1 1
1
1
1 1 1 1
RH = 4, 1 = 40, 2 = 40, 3 = 40, 4 = 40. We first assume that 0 is less than the minimum i
which is 40.
4
0 =
P4 1
1 + i=1 i
which gives 0 = 3.6780 < mini i , hence the assumption was correct.
C
= 13.7720 bits/sec/Hz
B
h11
.
9. H =
.
.
hMr 1
. . . h1Mt
...
.
...
.
...
.
. . . hMr Mt M
r Mt
Denote G = HH T
1
Gii =
lim
Mt Mt
1
lim
[hi1 . . . hiMt ]
Mt Mt
hi1
.
.
.
hiMt
Mt
1 X
khij k2
Mt Mt
lim
j=1
= Ej khij k
= 2
= 1 i
lim
Mt ,i6=j
1
Gij
Mt
1
=
lim
[hi1 . . . hiMt ]
Mt Mt
hj1
.
.
.
hjMt
Mt
1 X
=
lim
hik hjk
Mt Mt
k=1
= Ek hik hjk
= Ek (hik )Ek (hjk )
= 0 i, j, i 6= j
1
lim
HH T = IM
M M
i
h
HH T
= B log2 det [IM + IM ]
lim B log2 det IM +
M
M
= B log2 [1 + ] det IM
= M B log2 [1 + ]
10. We find the capacity by randomly generating 103 channel instantiations and then averaging over it.
We assume that distribution is uniform over the instantiations.
MATLAB CODE
clear;
clc;
Mt = 1;
Mr = 1;
rho_dB = [0:25];
rho = 10.^(rho_dB/10);
for k = 1:length(rho)
for i = 1:100
H = wgn(Mr,Mt,0,dBW,complex);
[F, L, M] = svd(H);
for j = 1:min(Mt,Mr)
sigma(j) = L(j,j);
end
sigma_used = sigma(1:rank(H));
gamma = rho(k)*sigma_used;
%% Now we do water filling\
gammatemp = gamma;
gammatemp1 = gammatemp;
gamma0 = 1e3;
while gamma0 > gammatemp1(length(gammatemp1));
gamma0 = length(gammatemp1)/(1+sum(1./gammatemp1));
gammatemp = gammatemp(1:length(gammatemp)-1);
end
C(i) = sum(log2(gammatemp1./gamma0));
end
Cergodic(k) = mean(C);
end
25
Mt = Mr = 3
Mt = 2 Mr =3
Mt = Mr = 2
Mt = 2 Mr =1
Mt = Mr = 1
20
Cergodic
15
10
10
15
20
25
(dB)
11. We find the capacity by randomly generating 104 channel instantiations and then averaging over it.
We assume that distribution is uniform over the instantiations.
MATLAB CODE
clear;
clc;
Mt = 1;
Mr = 1;
rho_dB = [0:30];
rho = 10.^(rho_dB/10);
for i = 1:1000
[F, L, M] = svd(H);
sigma(j) = L(j,j);
end
C(i) = sum(log2(1+gamma/Mt));
end
Cout(k) = mean(C);
pout = sum(C<Cout(k))/length(C);
while pout > .01
Cout(k) = Cout(k)-.1;
end
if Cout(k)<0;
Cout(k) = 0;
end
end
25
Mt = Mr = 3
Mt = 2 Mr =3
Mt = Mr = 2
Mt = 2 Mr =1
Mt = Mr = 1
20
Coutage
15
10
10
15
(dB)
20
25
30
12.
?
P (u n < X) = P
M
r
X
!
ui ni < X
i=1
Mr
X
ui P (ni < X)
i=1
= P (ni < X)
the statistics of u? n are the same as the statistics of each of these elements
13.
x = ku? Hvxk
= ku? Hvk2 kxk2
H
= v H H H u? u? Hvkxk2
= v H H H Hvkxk2
= v H QH Qvkxk2
max kxk2
with equality when u, v are the principal left and right singular vectors of the channel matrix H
SN Rmax = max
14.
kxk2
= max
N
0.1 0.5 0.9

H = 0.3 0.2 0.6
0.1 0.3 0.7
When both the transmitter and the receiver know the channel, for beamforming, u and v correspond
to the principal singular vectors (or the singular vectors corresponding to the maximum singular value
of H). Notice that the singular values of H are the square root of the eigen values of HH H (Wishart
Matrix).
Using Matlab, we get that the maximum singular value of H is 1.4480 and the singular vectors corresponding to this value are
0.7101
uopt = 0.4641
0.5294
and
0.1818
= 0.4190
0.8896
vopt
T v
It is easy to check that uTopt uopt = 1 and vopt
opt = 1 and that
uTopt H vopt = 1.4480

Since, during beamforming from eq. 10.17 in reader,
y = (uT Hv)x + uT n
and for a given transmit SNR of , the received SNR is given as

SNRrcvd = (uTopt Hvopt )2
since, uopt has norm 1, noise power is not increased. For, = 1, SNR is simply (1.4480)2 = 2.0968.
When the channel is not known to the transmitter, it allocates equal power to all the antennas and so
the precoding vector (or the optimal weights) at the transmitter is given as

1
1
1
v2 =
3 1
Define
h = Hv2
So, eq. 10.17 in the reader can be written as
y = (uT h)x + uT n
To maximize SNR we need to find a u2 of norm 1 such that (u0 h) is maximized.
Using Matlab, we get that the maximum singular value of h is 1.2477 and the singular vector corresponding to this value is
0.6941
u2 = 0.5090
0.5090
It is easy to check that uT2 u2 = 1 and that
uT2 h = 1.2477
Alternatively, from MRC concept we know that:
0.6941
= 0.5090
u2 =
||h||
0.5090
hH
where ||h|| is the L2 norm of h.

For a given transmit SNR of , the received SNR is given as
SNRrcvd = (uT2 Hv2 )2
since, u2 has norm 1, noise power is not increased. For, = 1, SNR is simply (1.2477)2 = 1.5567.
15. (a) = 10 dB = 10
Pe = d
So to have Pe 103 , we should have d 3, or at least d = 3. Solving the equation that relates
diversity gain d to multiplexing gain r at high SNRs we get
d = (Mr r)(Mt r)
3 = (8 r)(4 r)
Solving for r we get
r = 3.35 or 8.64
We have that r min{Mr , Mt }, so r 4 and so r = 3.35. But we know that r has to be an integer.
So, we take the nearest integer which is smaller than the calculated value of r, which gives us r=3 .
No credits for this part:
If we are allowed to assume that equations 10.23 and 10.24 hold at finite SNRs too and we are
given that we can use base 2 for logarithms, we can find the data rate as
R = r log2 () = 9.96 bits/s/Hz
(b) With, r = 3, we can find d as
d = (Mr r)(Mt r) = (8 3)(4 3) = 5
For this value of d,
Pe = d = 105
16.
According to SVD of h
= 1.242
C/B = log2 (1 + ) = log2 (1 + 1.2422 .10) = 4.038bps/Hz
17.
H=
.3 .5
.7 .2
.5946 .8041
.8041 .5946
.8713
0
0
.3328
.8507 .5757
.5757 .8507
P = 10mW
N0 = 109 W/Hz
B = 100 KHz
(a) When H is known both at the transmitter and at the receiver, the transmitter will use the optimal
precoding filter and the receiver will use the optimal shaping filter to decompose the MIMO channel into 2 parallel channels. We can then do water-filling over the two parallel channels available
to get capacity.
Finding the i s
1 =
21 P
= 75.92
N0 B
2 =
22 P
= 11.08
N0 B
Finding 0
Now, we have to find the cutoff value 0 . First assume that 0 is less than both 1 and 2 . Then
1
1
1
1
=1
0 1
0 2
2
1
1
=1+
+
0
1 2
1
0 =
= 1.81
1
1 + 1 + 12
which is less than both 1 and 2 values so our assumption was correct.
Finding capacity
Now we can use the capacity expression as
C=
2
X
B log2
i=1
i
0
= 800 Kbps
(b) Total is
Essentially we have two parallel channels after the precoding filter and the shaping filter are used
at the transmitter and receiver respectively.
M () = 1 + K
S()
S
Finding K
1.5
= .283
ln(5Pb )
K=
K = 0 /K.
Finding 0 or K
We now find the cut-off 0 . First assume that 0 < {1 , 2 }. Notice that 1 and 2 have already
been calculated in part (a) as 1 = 75.92 and 2 = 11.08.
1
1
0 1 K
1
1
0 2 K
=1
2
1
1
=1+
+
0
1 K
2 K
1
= 1.4649
0 =
1
1
1
1 + K 1 + 2
which is less than both 1 and 2 values, so our assumption was correct.
K = 0 /K = 5.1742
Finding Rate R
Therefore the total rate, R is given as
R = B log2
1
K
+ log2
2
K
R = B4.97
This gives that R=497.36 Kbps (Obviously less than ergodic capacity).
(c) Since now we use beamforming to get diversity only, the transmitter and the receiver use the
principal left and right eigen vectors of the Wishart Matrix HHH .
Once this is done the SNR at the combiner output is simply max , where max is the maximum
eigen value of the Wishart Matrix HHH and is NP0 B
Finding s
As given in the question, max is 0.7592 and was calculated to be 100. So we get that s = 75.92 .
Finding Pb
When using BPSK, s = b . Now we can use the expression for Pb for BPSK
0
Using the approx. given in the Ques.
Pb = Q
2b = Q 2 75.92 =
3.4 1035 Using Matlab
Credit is given for either value.
Finding Rate R
Since we are using BPSK and are given that B = 1/Tb , we get the rate using BPSK to be
R=100 Kbps .
Comparing with previous part
Comparing with part (b), we can see that the rate R decreases by 397.36 Kbps and the Pb improves as Pb is now 3.4 1035 0 whereas earlier it was 103 .
(d) Therefore we see that we can tradeoff rate for robustness of the system. If we are willing to decrease the rate at which we transmit, we can get more diversity advantage i.e. one strong channel
which gives a much less value of Pb .
18. (a) clear;
clc;
Mt = 4;
Mr = Mt;
rho_dB = [0:20];
rho = 10.^(rho_dB/10);
for i = 1:1000
[F, L, M] = svd(H);
sigma(j) = L(j,j);
end
%% Now we do water filling\
gammatemp = gamma;
gamma0 = 1e3;
while gamma0 > gammatemp1(length(gammatemp1));
gamma0 = length(gammatemp1)/(1+sum(1./gammatemp1));
gammatemp = gammatemp(1:length(gammatemp)-1);
end
C(i) = sum(log2(gammatemp1./gamma0));
end
Cergodic(k) = mean(C);
end
(b) clear;
clc;
Mt = 4;
Mr = Mt;
rho_dB = [0:20];
rho = 10.^(rho_dB/10);
for i = 1:1000
[F, L, M] = svd(H);
sigma(j) = L(j,j);
end
end
Cout(k) = mean(C);
end
19. using Matlab we get Cout = 7.8320
MATLAB CODE
clear;
clc;
Mt = 4;
Mr = Mt;
rho_dB = 10;
rho = 10.^(rho_dB/10);
for i = 1:1000
25
Cergodic Mt = Mr = 1
C M =M =1
out t
r
Cergodic Mt = Mr = 4
Cout Mt = Mr = 4
20
15
10
10
12
14
16
18
20
(dB)
[F, L, M] = svd(H);
sigma(j) = L(j,j);
end
end
Cout(k) = mean(C);
while pout > .1
Cout(k) = Cout(k)-.01;
end
if Cout(k)<0;
Cout(k) = 0;
end
end
20. As increases, the span of cdf becomes narrower and so capacity starts converging to a single number.
MATLAB CODE
clear;
clc;
Mt = 8;
Mr = Mt;
rho_dB = 10;
rho = 10.^(rho_dB/10);
for i =
H =
[F,
for
1:1000
wgn(Mr,Mt,0,dBW,complex);
L, M] = svd(H);
j = 1:min(Mt,Mr)
sigma(j) = L(j,j);
end
gamma = rho*sigma_used;
end
[f,x] = ecdf(C);
Empirical CDFs of Capacity w/o Tx CSI
1
0.9
M=4
M=6
M=8
0.8
0.7
FX(x)
0.6
0.5
0.4
0.3
0.2
0.1
0
6
10
11
12
x
13
14
15
16
17
18
Chapter 11
1. See Fig 1
2B = 100 KHz
fc-B
fc+B
fc = 100 MHz
Figure 1: Band of interest.
B = 50 KHz, fc = 100 MHz
Heq (f ) =
1
=f
H(f )
Noise PSD = N0 W/Hz. Using this we get

Z
Noise Power =
=
fc +B
N0 |Heq (f )|2 df
fc B
Z fc +B
N0
f 2 df
fc B
3 (fc +B)
= N0
f
3
(1)
(2)
(3)
(fc B)
N0
(fc + B)3 (fc B)3
3
= 1021 N0 W
(4)
(5)
Without the equalizer, the noise power will be 2BN0 = 105 N0 W. As seen from the noise power values,
there is tremendous noise enhancement and so the equalizer will not improve system performance.
2. (a) For the first channel:
ISI power over a bit time = A2 Tb /Tb = A2 For the 2nd channel:
R (n+1)Tb t/T
2 P
m dt = 2e1/2 A2
e
ISI power over a bit time = ATb
n=1 nTb
S(t)
T /2
m
h (t)
1
T
m
h (t)
2
(b) No ISI: pulse interval = 11/2s = 5.5s
Data rate = 1/5.5s = 181.8Kbps
If baseband signal =100KHz: pulse width = 10s
Data rate = 2/10s + 10s = 100Kbps
3. (a)
h(t) =
e
0
t0
o.w.
(6)
= 6 sec
Heq (f ) =
Z
H(f ) =
1
H(f )
t
e ej2f t dt
(7)
1
+ j2f
Hence,
Heq (f ) =
(b)
RB
Sx (f )|H(f )|2 |Heq (f )|2 df

RB
2
B N0 |Heq (f )| df
RB
2
B Sx (f )|H(f )| df
2BN0
SNReq
=
SNRISI
1
+ j2f
(8)
h (t)
1
10us
X (t)
1us
h (t)*X(t)
1
12us
Assume Sx (f ) = S, B f B

2BS 2
N0 2B
+ 83 B 3
2
RB
S B
|H(f )|2 df
2BN0
2B
2
1
+ 43 B 2
2
1.617 106
= 0.9364 = 0.28 dB
(c)
h [n] = 1 + e
Ts
[n 1] + e
Ts
2T s
[n 2] + ...
2Ts
3Ts
H(z) = 1 + e
z 1 + e
z 2 + e z 3 + ...

n
X
Ts
z
1
=
e z 1 =
=
Ts
Ts 1
ze
1e
z
n=0
Heq (z) =
1
H(z)+N0 .
Now, we need to use some approximation to come up with the filter tap
Ts
coefficient values. If we assume N0 u 0 (the zero-forcing assumption), we get Heq (z) = 1e z 1 .

1
Thus, a two tap filter is sufficient. For Ts = 30
ms, we have a0 =1, a1 = 0.0039 as the tap
coefficient values. Any other reasonable way is also accepted.
4. i = ci where {ci } is the inverse Z- transform of 1/F(z)
Show that this choice of tap weights minimizes
2
1
N
N
(
z
+
.
.
.
+
z
)
N
N
F (z)
. . . (1)
at z = ej
If F(z) is of length 2 and monic, say F (z) = 1 a1 z then
1
= 1 + a1 z 1 + a21 z 2 + . . .
F (z)
where c1 = a1 , c2 = a21 , a1 < 1
It is easy to see that the coefficients become smaller and smaller. So if we had the opportunity to
cancel any (2N+1) coefficients we will cancel the ones that are closest to z 0 . Hence we get that i = ci
minimizes (1). The result can be similarly proved for length of F(z) greate than 2 or non-monic.
5. (a) Heq (f ) =
1
H(f )
for ZF equalizer
2
0.5
Heq (f ) =
fc 20M Hz f < fc 10M Hz

fc 10M Hz f < fc
fc f < fc + 10M Hz
fc + 10M Hz f < fc + 20M Hz
o.w.
(9)
(b) S=10mW Signal power

N = N0 [12 10M Hz + 22 10M Hz + 0.52 10M Hz + 42 10M Hz] = 0.2125mW
SN R = 47.0588 = 16.73dB
(c) Ts = 0.0125sec
1.5SN R
Pb 0.2e1.5/M 1 or M 1 + ln(5P
b)
for Pb = 103 M 14.3228 (M 4 thus using the formula is reasonable)
R = logT2sM = 307.2193M bps
(d) We use M=4 non overlapping subchannels, each with B=10MHz bandwidth
1: fc 20M Hz f < fc 10M Hz 1 = 1
2: fc 10M Hz f < fc 2 = 0.5
3: fc f < fc + 10M Hz 3 = 2
4: fc + 10M Hz f < fc + 20M Hz 4 = 0.25
P 2
Power optimization: i = N0 Bi for i = 1, 2, 3, 4
1 = 1000 2 = 250 3 = 4000 4 = 62.5
for Pb = 103 K = 0.2831
Pi
=
P
1
0
1
Ki
i 0 /K
i 0 /K
We can see that all subchannels will be used and

P1 = 2.6523 P2 = 2.5464 P3 = 2.6788 P4 = 2.1225
and
thus R = 2B
P4
i=1 log(Ki /0 )
0 = 3.7207
= 419.9711M bps
6. (a)
F{f (t)} =
FZ (f ) =
|f | < 1/T
o.w.
1 X
n
F f+
TS n=
Ts
= 1
folded spectrum of f(t) is flat.
T
0
(10)
(b)
yk = y(kT + t0 )
X
=
Xi f (kT + t0 iT )
i=
N
X
Xi f ((k i)T + t0 )
i=N
N
+k
X
= Xk sinc(t0 ) +
Xi f ((k i)T + t0 )
i=N +k,i6=k
{z
ISI
(c)
ISI =
N
+k
X
Xi
i=N +k,i6=k
sin ((k i) + t0 /T )
(k i) + t0 /T
N
X
= sin(t0 /T )
i=N,i6=0
= sin(t0 /T )
=
7. gm (t) = g ? (t) = g(t) = sinc(t/TS ), |t| < Ts

1
Noise whitening filter : G? (1/z
?)
1
1
+
t0 /T i t0 /T i
i=1
N
X
2
sin(t0 /T )
Thus, ISI as N
N
X
1
t0 /T i
n=1
n2
n
t20 /T 2
8. Jmin = 1
Jmin
j= cj fj
T
=1
2
B(z) = C(z)F (z)

F (z)F ? (z 1 )
=
F (z)F ? (z 1 ) + N0
X(z)
=
X(z) + N0
I
1
B(z)
b0 =
dz
2j
z
I
1
X(z)
=
dz
2j
z[X(z) + N0 ]
Z /T
X(ejT )
T
=
d
2 /T X(ejT ) + N0
/T
/T
X(ejT )
d
X(ejT ) + N0
T
2
/T
/T
Z /T
T
2
Z
= Ts
=
N0
jT
X(e )
T 1
/T
0.5Ts
0.5Ts
+ N0
N0
d
2
n= |H( + 2n/T )| + N0
N0
df
F (f ) + N0
9.
VW J
J
J
,...,
w0
wN
wT Mv w? 2<{Vd w? } + 1
J
= 2Mv wT 2Vd
w
J
= 0 2Mv wT = 2Vd
w
1 H
wopt = MvT
Vd
10.
Z
Jmin
N0
F (f ) + N0
N0
F (f ) + N0
Jmin
Ts
0
0.5Ts
0.5Ts
N0
df
F (f ) + N0
Jmin 0
N0
=1
N0
Z 0.5Ts
Ts
1df = 1
0 Jmin 1
0.5Ts
11.
F (f ) =
=
1 X
n
F f+
Ts n=
Ts

1 X
j4 f + Tn
j2 f + Tn
s
s
+ 0.3e
1 + 0.5e
Ts n=
MMSE equalizer :
Z
Jmin = Ts
DF equalizer :
0.5/Ts
Jmin = exp Ts
0.5/Ts
N0
df
F (f ) + N0
0.5/Ts
N0
ln
F (f ) + N0
0.5/Ts
)
df
12. (a) G(f) is a sinc(), so theoretically infinite. But 2/T is also acceptable (Null to Null bandwidth)
(b) T is more likely since T = 109 sec
As long as > Tb , get ISI and so, frequency selective fading
(c) Require Tb = Tm + T R =
1
Tm +T
= 49997.5bps
1
for ZF equalizer
(d) Heq (z) = F (z)
1
Heq (z) = d0 +d1 z 1
+d2 z 2
Long division yields the first 2 taps as
w0 = 1/0
w1 = 1 /02
13. (a)
2
1
Hzf (f ) =
= 3
H(f )
0 f 10KHz
10KHz f 20KHz
20KHz f 30KHz
30KHz f 40KHz
40KHz f 50KHz
(b) The noise spectrum at the output of the filter is given by N (f ) = N0 |Heq (f )|2 , and the noise
power is given by the integral of N (f ) from -50 kHz to 50 kHz:
Z
50kHz
N=
N (f )df
f =50kHz
= 2N0
50kHz
f =0kHz
|Heq (f )|2 df
= 2N0 (1 + 4 + 9 + 16 + 25)(10kHz)
= 1.1mW
0
(c) The noise spectrum at the output of the filter is given by N (f ) = (H(fN)+)
2 , and the noise power
is given by the integral of N (f ) from -50 kHz to 50 kHz. For = .5 we get
N = 2N0 (.44 + 1 + 1.44 + 1.78 + 2.04))(10kHz) = 0.134 mW

For = 1 we get
N = 2N0 (.25 + .44 + .56 + .64 + .69))(10kHz) = 0.0516 mW
(d) As increases, the frequency response Heq (f ) decreases for all f . Thus, the noise power decreases,
but the signal power decreases as well. The factor should be chosen to balance maximizing the
SNR and minimizing distortion, which also depends on the spectrum of the input signal (which
is not given here).
(e) As , the noise power goes to 0 because Heq (f ) 0 for all f . However, the signal power
also goes to zero.
14. The equalizer must be retrained because the channel de-correlates. In fact it has to be retrained at
least every channel correlation time.
Benefits of training
(a) Use detected data to adjust the equalizer coefficients. Can work without training information
(b) eliminate ISI.
15. N = 4
LMS-DFE: 2N +1 operations/iteration 9 operations/iteration
RLS: 2.5(N )2 + 4.5N operations/iteration 58 operations/iteration
Each iteration, one bit sent. The bit time is different for LMS-DFE/RLS, Tb (LMS-DFE) <Tb (RLS).
But time to convergence is faster for RLS.
Case 1: fD = 100 Hz (tc ) 10 msec, must retrain every 5 msec.
LMS-DFE: R =
RLS: R =
107
9
107 50 bits
58 - 5 msec
1000 bits
5 msecs
= 911 Kbps
= 162 Kbps
Case2: fD = 1000 Hz retrain every 0.5 msec

RLMS-DFE = 0 bps
RRLS = 72.4 Kbps
16. In the adaptive method, we start with some initial value of tap coefficients W0 and then use the
steepest descent method
Wk+1 = WK GK . . . (1)
where is some small positive number and GK is the gradient of MSE = E|dk dk |2 is RWk p
(Notice that 11.37 was a solution of gradient =0 , RW = p)
GK = RWk p = E[k Yk ? ]
where Yk = [yk+L . . . yKL ]T and k = dk dk

Approximately (1) can be rewritten as
Wk+1 = Wk + k Yk ?
Chapter 12
1. (a) i = cos (2j/TN t + j )

To form a set of orthonormal basis on [0, TN ]
RT
We need, 0 N j k dt = 0
Z
0
TN
Z
j k dt =
TN
cos (2j/TN t + j ) cos (2k/TN t + k ) dt
TN
1
1
cos (2(j + k)/TN t + j + k ) + cos (2(j k)/TN t + j k ) dt
2
2
0
TN
1
=
[sin (2(j + k) + j + k ) sin (j + k )]
2 2(j + k)
TN
1
+
[sin (2(j k) + j k ) sin (j k )]
2 2(j k)
= 0
j and k are integers

The minimum separation for sub-carriers cos (2j/TN t + j )is 1/TN for any j
(b) If j = 0j
Z
0
TN
j k dt =
1
TN
1
TN
sin 2(j + k) +
sin 2(j k) = 0
2 2(j + k)
2 2(j k)
2(j + k) = l1 2(j k) = l2
j and k are multiples of 1/2
The minimum separation:1/2TN
l1 , l2 Z
2. (a) TN = 1/BN = 10Tm = 10/Bc = 10/10KHz = 1ms

(b) B =
N (1++)
= 128(1+1.5+0.1)
= 333KHz
TN
1ms
N ++
= 128+1.5+0.1
= 129.6KHz
TN
1ms
(c) B =
The total bandwidth using overlapping carriers is less than half of the non overlapping bandwidth.
3.
x3 [n] = x2 [n] x1 [n]
=
X3 [k] =
=
N
1
X
m=0
N
1
X
x2 [m]x1 [(n m)N ]

kn
x3 [n]N
n=0
" 1
N
1 N
X
X
n=0
m=0
#
kn
x2 [m]x1 [(n m)N ] N
interchange order of summation

X3 [k] =
=
N
1
X
m=0
N
1
X
x2 [m]
"N 1
X
#
kn
x1 [(n m)N ]N
n=0
km
x2 [m]X1 [k]N
m=0
= X1 [k]
N
1
X
km
x2 [m]N
m=0
= X1 [k]X2 [k]
circular convolution of discrete-time sequences leads to multiplication of their DFTs.
4. (a) For FDM, the number of subchannels =
1
10 sec
B
(f )c
=5
Ts = 10 sec, R =
= 0.1 Mbps
h q
i
(b) Pbn = 12 1 b bn+1

where bn = 1000/n
n
n
1
2
3
4
5
bn
1000
500
333
250
200
Pbn
2.5 104
5 104
7.5 104
103
1.25 103
P
BER after decoding = 5i=3 Pr [i channels in error] = 3.5 109
The total date rate of the system is the same as the data rate of any of the subcarriers (since they
all have the same bits transmitted over them) R = 0.1 Mbps
(c) Since it is not specified which equation to use for calculation of SNR, all answers
based on any correct equation in the reader are being given full credit. What is
given below is just one way to do the problem. Your answer can be totally different
but we still give credit for it.
For BPSK in Rayleigh fading:
r
Pb =
1
2
1+
For P b 103 , we get SNRmin = 248.75.
For higher order QAMs we use the equation given in the reader:
s
!
M
0.5M s
Ps =
1
2
1 + 0.5M s
which gives
M
Pb
2 log2 M
0.5M s
1 + 0.5M s
For 4-QAM (since its a rectangular constellation), from Table 6.1 we have M = 1 and M = 1, so
we get SNRmin = 249.25. However using the exact equation 6.81, we get that for 4-QAM, SNRmin
= 453, which makes more sense as it is much greater than that required for BPSK. Hence we will
use this value.
For 8-QAM (since its a non-rectangular constellation), from Table 6.1 we have M = 4 and
M = 3/7, so we get SNRmin = 1552, which is higher than the SNR on any branch and so should
not be used.
n sn Max(M )
Pb
1 1000
4
0.454 103
2 500
4
0.906 103
Hence:
3 333
2
0.748 103
4 250
2
0.997 103
5 200
0
1.25 103 with BPSK
R = 0.1
n log2 (M )
= 0.6 Mbps
5. (a) If the baseband bandwidth is 100 Khz, then at the carrier frequency they have bandwidth of
B = 200 KHz.
For flat fading we need the coherence bandwidth to be much greater than the bandwidth of the
signal. Therefore Bc 10B = 2 MHz.
For independent fading, we want the channel between two carriers to be uncorrelated. That is,
we want Bc < B = 200 KHz.
If the fading between the different channels is correlated, they will all tend to have fades at the
same instants. Therefore, coding over sub-channels will not work because all channels will tend
to fade at same time and the code will not be able to correct all these errors.
(b) We have
BER .2e1.5(M 1)
which means that
M 1+
1.5
= 1 + 0.283
ln 5BER
For the first sub-channel, this gives M 4.5672, which means we use 4-QAM.
For the second sub-channel, M 8.108, which means we use 8-QAM.
For the third sub-channel, M 18.856, which means we use 16-QAM.
Therefore, at each symbol time we will transmit 9bits.
We have that Ts = 1/B = 10 s, means we transmit at 900 Kbps total.
(c) To achieve the same data rate, we will need 3bits/symbol per sub-channel, that is 8-QAM constellation.
To achieve this, we need a minimum SNR of 13.93 dB per sub-channel.
In the first sub-channel we must increase the power by 2.93dB, giving a transmit power of Pt1 =
196.33 mW.
For the second sub-channel we need 0.07 dB less power, that is Pt1 = 98.40 mW.
For the third sub-channel we need 4.07 dB less, that is Pt2 = 39.17 mW.
The total transmit power is now, Pt = 333.90 mW. Therefore we need to increase the transmit
power by 33.90 mW with respect to case b).
6. Tc = 20s
Bc = 1/Tc = 50KHz
BN = Bc /2 = 25KHz
B = N BN = 8 25KHz = 200KHz
SN R = 20dB, target BER = 103
For MQAM Pe 0.2e1.5/M 1
M = 1 + 1.5/ln 5Pe = 29.31

M = 16
BN
R = N RN = N log2 M/TN = N log2 M 1+
= 400Kbps
7.
yN 1
yN 2
.
.
.
y0
h0
0
...
...
...
0
h1
h0
...
...
...
...
...
...
...
...
...
...
h
h1
...
...
...
...
0
h
...
...
...
...
...
...
...
...
...
h0
h0
0
...
...
...
0
...
h0
...
...
...
...
h
...
...
...
...
...
0
h1
...
...
...
...
...
h
...
...
...
...
0
0
...
...
h
h0
0
0
0
0
h2
h1
...
0
...
...
h3
h2
0
...
...
...
...
...
0
0
...
...
h2
h1
...
0
...
...
...
...
0
0
0
0
...
...
h0
0
...
...
h2
h1
...
h0
...
...
h3
h2
h
...
...
...
...
...
0
h1
...
...
h2
h1
...
h
...
...
...
...
...
...
...
...
...
...
0
0
...
...
h
h0
...
...
...
...
...
...
xN 1
.
.
.
.
x0
xN 1
.
.
.
.
x0
...
...
...
...
...
...
0
0
...
...
...
h
xN 1
.
x0
x1
.
x
xN 1
.
.
.
.
x0
N 1
.
.
.
.
0
N 1
.
.
.
.
0
8. DFT:
X[i] =
N 1
1 X
in
x[n]N
N n=0
X[0] =
N 1
1 X
x[n]1
N n=0
N 1
1 X
n
x[n]N
N n=0
......
......
N 1
1 X
(N 1)n
X[N 1] =
x[n]N
N n=0
1
1
1
... ... 1
N 1
1
2
. . . . . . N
N
N
1
...
... ... ...
X(n) = . . . . . .
N
.
.
.
.
.
.
.
.
.
... ... ...
2(N 1)
(N 1)2
N 1
1
N
N
. . . . . . N
X[1] =
.x(n)
N 1
.
.
.
.
0
9. (a) First row of Q = [1 1 . . . 1] P

HQ(1, :)T = H[1 1 . . . 1]T = i=0 hi [1 1 . . . 1]TP
First row of Q is an eigenvector of H with 0 = i=0 hi
N 1
2
(b) Second row of Q = 1 N N
. . . . . . N
N 1 T
2
HQ(2, :)T = H 1 N N
. . . . . . N
=
i
hi N
1 N
2
N
N 1
. . . . . . N
i=0
i
hi N
Q(2, :)T
i=0
P
i
second row of Q is an eigenvector of H with 1 = i=0 hi N
h
i
2(k1)
(N 1)(k1)
k1
(c) For k th row of Q = 1 N
N
. . . . . . N
h
iT
2(k1)
(N 1)(k1)
k1
HQ(k, :)T = H 1 N
N
. . . . . . N
=
=
X
i=0
i(k1)
hi N
i(k1)
hi N
2(k1)
k1
1 N
N
(N 1)(k1)
. . . . . . N
Q(k, :)T
i=0
k, k th row of Q is an eigenvector of H with eigenvalue k =
i(k1)
i=0 hi N
10. x
e[n] = |0 . . {z
. . . . 0} x0 . . . . . . xN 1
For n N
y[n] = x
e[n] ? h[n]
X
=
h[k]e
x[n k]
k=1
=
=
X
k=1
h[k]x[n k]
h[k]x[n k]N
k=1
= x[n] h[n]
For n <
y[n] = x
e[n] ? h[n] + x
e[n + N ] ? h[n + N ]
X
X
=
h[k]e
x[n k] +
h[k]e
x[N + n k]
=
k=1
h[k]x[n k] +
k=1
k=n+1
X
k=n+1
h[k]x[n k]N
k=1
= x[n] h[n]
h[k]x[n k]N
iT
11. (a)
H=
.7 .5 .3 0 0 0 0 0 0 0
0 .7 .5 .3 0 0 0 0 0 0
0 0 .7 .5 .3 0 0 0 0 0
0 0 0 .7 .5 .3 0 0 0 0
0 0 0 0 .7 .5 .3 0 0 0
0 0 0 0 0 .7 .5 .3 0 0
0 0 0 0 0 0 .7 .5 .3 0
0 0 0 0 0 0 0 .7 .5 .3
Y = Hx +
(b)
e =
H
.7 .5 .3 0 0 0 0 0
0 .7 .5 .3 0 0 0 0
0 0 .7 .5 .3 0 0 0
0 0 0 .7 .5 .3 0 0
0 0 0 0 .7 .5 .3 0
0 0 0 0 0 .7 .5 .3
.3 0 0 0 0 0 .7 .5
.5 .3 0 0 0 0 0 .7
e = M M H
H
(c) The flat fading channel gains are the diagonal elements of the matrix
MATLAB CODE
clear all;
H=[ .7 .5 .3 0 0 0 0 0
0 .7 .5 .3 0 0 0 0
0 0 .7 .5 .3 0 0 0
0 0 0 .7 .5 .3 0 0
0 0 0 0 .7 .5 .3 0
0 0 0 0 0 .7 .5 .3
.3 0 0 0 0 0 .7 .5
.5 .3 0 0 0 0 0 .7 ]
[V,D] = eig(H)
V*D*V-H
12. (a) = 4
The VC system doesnt require a cyclic prefix to make the subchannels orthogonal.
(b)
y255
y254
.
.
.
y0
1
0
...
...
...
0
0.6
1
...
...
...
...
0.7
0.6
...
...
...
...
0.3
0.7
...
...
...
...
0.2
0.3
...
...
...
...
0
0.2
...
...
...
1
0
0
...
...
...
0.6
0
...
...
...
...
0.7
...
...
...
...
...
0.3
0
0
...
...
...
0.2
x255
.
.
.
.
x4
255
.
.
.
.
0
H = U V H
Singular values (using svd in Matlab)
(c) use svd in Matlab
1/2bit
1 sub-carrrier symbol
1 coded bit
coded bit sub-carrrier symbol
4106 sec
3/4bit
6 coded bit
4106 sec
13. Rmin = 52sub-carriers

Rmax = 52sub-carriers
= 6.5M bps
= 58.5M bps
14.
48
1/2bit
1 coded bit
sub-carriers
2
4 106 sec
3/4bit
6 coded bit
48
sub-carriers
+
2
4 106 sec
= 7.5M bps
R =
15. x(t) = sinc
t
T
cos( t
T )
1 2t
T
Th

T
T
X() =
1
sin
2
2 ||
i || T (1 )
T
T (1 ) ||
||
T (1
T (1
+ )
+ )
max |x(t)|2 = s2 (t)|t=0 = 1

t
Et |X(t)|2 =
=
1
E |X()|2
2T
Z (1+)
T
1
T
T
1 2
2
1 sin
T
(1 )
|
||
|2 d +
2T
2
2
T
2T
T
(1)
T
= 1 0.5
P AR =
=0
=1 ,
=2
maxt |x(t)|2
1
=
Et [|X(t)|2 ]
1 0.5
1
2 This pulse shape is less sensitive to timing errors.
P AR =
16.
lim =
=
ICIi =
TN 1 ej2(+m)
lim
0
j2( + m)
TN 1 ej2
lim
0
j2m
TN (j2)
j2m
TN
m
X
|Im |2
m6=i
X TN 2
m6=i
=
C0
X 1 2
m
m6=i
X 1 2
=
m
m6=i
(TN )2
Chapter 13
1. xi =
RT
0
x(t)s(t)dt =
PN
2
j=1 sij
+ Ij sij )
(see [1] for details)
(a)
E[xi /si (t)] = Es ( to show )
si (t) =
sij j (t)
By linearity of expectation,
N
N
X
X
E[(
s2ij + Ij sij )/si (t)] = E[(
s2ij /si (t)] + E[Ij sij )/si (t)]
j=1
(1)
j=1
sij are all zero mean and variance Es /N

Also, the interfering signal is independent of the transmitted signal. Thus, the equation (1) above,
evaluates to
Es
N + E[Ij ]0 = Es
N
Notice that the knowledge of si is needed to get the correlation to work out. If si is not known,
correlation will do the right thing only 1/M of the time as we see in the next part.
(b)
E[xi ] =
N
X
E[s2ij ]
j=1
N
X
E[s2ij /sij ]p(sij )
j=1
Es
M
(c)
V ar[xi /si (t)] = E[x2i /si (t)] E[xi /si (t)]
= E[x2i /si (t)]
X
= E
Ik Il sik sil
k,l
i2k E[s2ik ] =
Es
Ej
N
Es X 2
Ik
N
k
(d) As in part b,
E E
V ar[xi ] = NsMj
(e) SIR =
E 2 [xi ]
V ar[xi ]
(As correlator gives non-zero expected output only 1/M of the time)
(Es /M )2
Es Ej
NM
Es N
Ej M
2. Matlab
fc = 100e6;
Ts = 1e-6;
ss = 100;
t = [Ts/ss:Ts/ss:2*Ts];
s = [ones(1,100) -1*ones(1,100)]; sc = [ones(1,10) -1*ones(1,10)];
for i = 1:9
sc = [sc ones(1,10) -1*ones(1,10)];
end
car = cos(2*pi*fc*t);
x = s.*sc.*car;
1
0.8
0.6
0.4
x(t)
0.2
0.2
0.4
0.6
0.8
0.2
0.4
0.6
0.8
1.2
1.4
1.6
1.8
2
6
x 10
Figure 1: Problem 2
3. = 10s
(a) No fading if hop rate greater than 1/e
c <
1
1
Rc =
= hoprate >
= 100KHz
c
e
(b) c = 50s
s = 0.5ms
Since number of reflected paths get averaged over one symbol duration, we have flat fading.
(c) c = 50s
s = 0.5s
System has severe frequency-selective fading.
4. n0 (t) = n(t)sc (t)

(a)
n0 ( ) = E[n0 (t)Fn0 (t)]
(Real n(t))
= E[(n(t)sc (t))F(n(t)sc (t))]

Z
= E
n( )sc ( )n(t + )sc (t + )d
Z
=
n ( )c ( )d
[ P SD = F(n0 ( )) ]
(b) If c ( ) = ( ) n0 ( ) = n ( )
(c) If n(t)is AWGN then n ( ) = N20 ( )
n0 ( ) = N20 c ( )
As N , n0 ( ) = N20 ( ) (same as n ( ))
5. sc (t) is real and periodic. c (t) is periodic with the same period.
c (t) =
c (t) =
Let x = t +
Z
1
sc ( )sc (t + )d
T T
Z
1
sc ( )sc (t + )d
T T
integral over any interval of length T
Z
1
sc (x + t)sc (x)dx
T T
= c (t)
c (t) =
To prove that maximum is at t = 0,

Z
Z
c (t) = sc ( )sc (t + )d s2c ( )d = c (0)
Hence maximum is at 0.
6.
1
T
Z
0
sc (t 0 )sc (t 1 )dt =
Let
x = t 0
1
T
Z
T
sc (t 0 )sc (t 1 )dt
Z
1
=
sc (x)sc (x + 0 1 )dx
T T
= c (0 1 ) = c (1 0 )
7. sc (t) is periodic with period T

sc (t T ) = sc (t)
c (t) =
c (t + T ) =
Z
1 T
sc ( )sc (t + )d
T 0
Z
1 T
sc ( )sc (t + T + )d
T 0
But sc (t + T + ) = sc (t + )
c (t + T ) =
1
T
Z
0
sc ( )sc (t + )d = c (t)
8. We are given in (13.19) that
(
c ( ) =
)
1 | |(1+1/N
Tc
1/N
| | Tc
| | Tc
periodic with period

N Tc = Ts
one period
T_c
NTc = Ts (N+1)Tc
Figure 2: Problem 8
1
N +1
1
1
(t)
sinc2 (f ) (f )
1+
N
N
N
N
1
1
1
1
2
1+
(t/Tc )
Tc (1 + )sinc (f Tc ) (f Tc )
N
N
N
N
make periodic with period N Tc
X
X
1
m
(t kN Tc )
S(f )|f = NmT (f
)
c
N
T
N
T
c
c
m=
k=
sc (f )
X
N + 1
m
2 m
N 2 sinc ( n )(f Ts )
m=
1
(m)
2
N
| {z }
neglect for large N
MATLAB
clear;
Ts = 1e-6;
spread = 1000;
ss = 10;
f = [-1*spread*(1/Ts): ss/Ts: spread*(1/Ts)];
N = 100;
m = -1*spread:ss:spread;
for i = 1:length(m)
Psc(i) = ((N+1)/N^2)*(sinc(m(i)/N))^2 - (1/N^2)*(m(i)==0);
end
plot(f*Ts,Psc,bo)
0.01
0.009
0.008
0.007
Psc(f)
0.006
0.005
0.004
0.003
0.002
0.001
0
1000
800
600
400
200
200
400
600
800
1000
f (in MHz)
Figure 3: Problem 8
9. Both m-sequences and random binary spreading sequences have balanced run-length and shift properties.
First notice that all these properties are trivially true for a random binary sequence (rbs)
1. balanced: since a rbs is generated by coin tosses using a fair coin E[heads] = E[00 s] = N/2 =
E[10 s] = E[tails] after N coin tosses.
2. run-length:
prob of having (single 1 or single 0 in a row) = 1/2
prob of having (2 1s or 2 0s in a row) = 1/4
prob of having (k 1s or k 0s in a row) = 1/2k

3. Shift: Comparing a rbs with a shifted version of itself is same as comparing two independent coin
tosses, so prob(2 independent coin tosses result in same outcome) = 1/2 = prob(2 independent coin
tosses result in different outcome)
half of the positions will be same and the remaining half will be different
Now we show the same for m-sequences
1. balanced: Consider a shift register generator of length r. The shift register passes through all
possible nonzero states. Of these states 2r1 = 12 (N + 1) have a one in the right-most position, and
2r1 1 have a zero in the right-most position. Thus there is one more one than zeros in the output
sequence.
2. run-length: Consider a shift register of length r. There can be no run of ones having length l r
since this would require that the all-one shift register state be followed by another all-ones state. This
cannot occur since each shift register state occurs once and only once during N cycles. Thus there is a
single run of r consecutive ones and this run is preceded by a zero and followed by a zero.
A run of r-1 ones must be preceded and followed by a zero. This requires that the shift register state
which is r-1 ones followed by a 0 be followed immediately by the state which is a 0 followed by r-1
ones. These two states are also passed through in the the generation of the run of r ones, where they
are separated by the all-ones state. Since each state occurs only once, there can be no run of r-1 ones.
A run of r-1 zeros must be preceded by and followed by 1s. Thus the shift register must pass through
the state which is a 1 followed by r-1 zeros. This state occurs only once so there is a single run of r-1
zeros.
Now consider a run of k ones where 1 k r 1. Each run of k ones must be preceded by and
followed by a 0. Thus the shift register must pass through the state which is a 0 followed by k ones
followed by a 0, with r-k-2 remaining positions taking arbitrary values. There are 2rk2 possible ways
to complete these remaining positions in the shift register, so there are 2rk2 runs of k ones. Similarly,
there are 2rk2 runs of k zeros.
3. Shift: We first prove a property known as shift and add. This says that the modulo-2 sum of an
m-sequence and any phase shift of the same sequence is another phase of the same m-sequence.
Proof: Consider a shift register. Since different initial conditions result in a different phase of the same
sequence, two phases b(D) and b(D) of the same sequence can be written as b(D)=a(D)/g(D) and
b(D)=a(D)/g(D), where a(D) and a(D) are distinct initial conditions. The modulo-2 sum b(D) +
b(D) =[a(D)+a(D)]/g(D) = a(D)/g(D). Since the modulo-2 sum o any two distinct initial conditions
is a third distinct initial condition, a(D)/g(D) =b(D) is a third distinct phase of the original sequence
b(D).
Now we prove the claim by contradiction. Suppose they do not match and mismatch at exactly half
the locations. Then their sum will have more 1s and 0s. sum is also an m-sequence which is supposed
to have equal 1s and 0s; thus we have a contradiction.
10. h(t) = 0 (t 0 ) + 1 (t 1 )
neglecting noise, the signal input to the synchronizer is given as
z(t) = 0 x(t 0 )sc (t 0 ) cos2 (2fc (t 0 )) + 1 x(t 1 )sc (t 1 ) cos(2fc (t 1 )) cos(2fc (t 0 ))
We have assumed that the demodulator got synchronized to the first multipath
Assuming x(t) remains the same xk over an interval [0,T] we have
w( ) =
1
T
Z
0
0 xk sc (t 0 )sc (t ) cos2 (2fc (t 0 ))dt
Z
1 T
+
1 xk sc (t 1 )sc (t ) cos(2fc (t 1 )) cos(2fc (t 0 ))dt
T 0
0.5xk
0.5xk
0 c ( 0 ) +
1 c ( 1 ) cos(2fc (0 1 ))
T
T
For binary transmission xk = 1
So Costa loop will try to maximize |w( )| and will synchronize to the first component if 0 is higher
or the second component of 1 cos(2fc (0 1 )) is higher.
A similar analysis can be done if demodulator synchronizes to the other multipath component. However,
notice that the demodulator will also synchronize to the multipath component for which 1 is greater.
Essentially, both the demodulator and the Costa loop will synchronize to the first multipath component
if 0 is higher and to the second multipath component otherwise.
11. Pb = 106
For DPSK, we know that the BER is given by 12 eb , which for a maximum BER of 106 , gives a value
of minimum receive SNR of min = 13.12.
We assume that the noise statistics are the same after de-spreading and that the possible interference
is modeled as white noise.
We have that
SNR(after-despreading) = K2c ( )SNR(before-despreading)
For the first branch, c (0) = 1, for the second branch c (Tc /4) = 0.733 and for the third branch
c (Tc /3) = 0.644. Therefore, the SNRs after de-spreading are: SNR1 = 150, SNR2 = 80.667, SNR3 =
62.296.
With selection combining, the outage probability is given by
Pout = (1 e
min
1
)(1 e
min
2
)(1 e
min
3
) = 0.24%
12. (a) E(rP ) = E(r1 ) + E(r2 ) = 9 + 92 = 11.9438

(b) p(rP ) = p(r1 )Fp(r2 )
a=1/6. See Fig 4
1/6
6(1+
2 )
6+12
12+ 6
12(1+
2 )

1
2
exprb rb = 8.5172
1
62
height of = 16 8.51727.9625
= 0.0471
(c) Pb =
Pout = 12 (0.0471)(8.5172 7.9625) = 0.0131
1/6
12
13. (a) For the multipaths to be attenuated by 1/N , we need 1 Tc and 1 Tb Tc . Similarly, we
need 2 Tc and 2 Tb Tc .
(b) Instantaneous BER in DPSK is given by Pb = 21 eb . For Pb = 103 , this corresponds to b = 6.21.
While the channel has average power , the outage probability in Rayleigh fading is given by
Pout = 1 e0 / . Note that 0 is Rayleigh distributed with average power 5 with probability
.5 and the channel is Rayleigh distributed with average power 10 with probabilty .5. This is not
equivalent to Rayleigh fading with average power 7.5. While the channel has average power 5
the outage probability is 1 e6.21/5 , and while the channel has average power 10 the outage
probability is 1 e6.21/10 . The overall outage probability is given by
Pout = .5(1 e6.21/5 + 1 e6.21/10 ) = .587
(c) The outage probability when using selection combining is the product of each of outage probability
of each branch. For branch 2, the outage probability is .5(1 + 1 e6.21/20 ) = .633. For branch
3, the outage probability is .75(1 e6.21/5 ) + 0.25(1 e6.21/10 ) = 0.649. Therefore, the overall
outage probability is given by:
Pout = Pout,1 Pout,2 Pout,3 = (0.587)(0.633)(0.649) = 0.241
(d) Again, the outage probability of a 2-branch RAKE with selection combining is given by the product
of the outage probability of each branch. Therefore, we should select the two branches with the
smallest outage probabilities, i.e. branches 0 and 1. The corresponding outage probability is
Pout = Pout,1 Pout,2 = (0.587)(0.633) = 0.37
14. Following along the same lines as the previous question, we know that outage on the ith branch is
given as:
(
1
w.p. 0.25
Pout =
min
1 e i
w.p. 0.75
For DPSK, we know that the BER is given by 12 eb , which for a maximum BER of 103, gives a
value of minimum receive SNR of min = 6.21. Therefore, the outage probability will be
6.21
6.21
6.21
Pout = 0.25 + 0.75 1 e 20
0.25 + 0.75 1 e 10
0.25 + 0.75 1 e 6.67
6.21
6.21
0.25 + 0.75 1 e 5
0.25 + 0.75 1 e 4
= 12.49%
For the case where there is always a multipath in each bin with average SNR of 20,
6.21 5
Pout = 1 e 20
= 0.14%
Clearly the outage probability is much smaller in the second case.
15. (a)
i =
1
Tb
Tb
sc (t iTc )
N
X
alphaj (t jTc )sc (t)dt
j=0
Z Tb
1
=
j
sc (t iTc )sc (t jTc )dt
Tb 0
j=0
i forith carrier
= j ij =
0 fori 6= j
N
X
thus i = i for all i

(b) 0 = a,
(c) Pb =
SC:
1 = 0.8b,
0.5e b 0 =
2 = 0.2b,
3 = 0.5c,
4 = 0.5c
6.21
h
iM
Po ut = 1 e0 /
= 0.099
MRC:
Po ut = 1 e
0 /
(0 /)2
1 + 0 / +
= 0.025
2
16. Autocorrelation and Cross-correlation for Gold codes, Kasawi codes from the small set
n=8
Kasawi codes ; small set
1
=
0.0039
2n 1
2n/2 +1
0.0667
( ) =
2n 1 =
2n/2 +12
= 0.0588
2n 1
number of sequences = 2n/2 = 16
large set
( ) =
1
2n 1 =
12n/2
=
2n 1
1(2n/2 +1)
=
2n 1
0.0039
0.0667, 0.0588
0.0627, 0.0706
number of sequences = 23n/2 = 4096

Gold codes; t(n) = 2(n+2)/2 + 1 = 33
( ) =
17.
1
2n 1 =
1t(n)
2n 1 =
t(n)2
2n 1 =
0.0039
0.1294
0.1216
1 1
1
1
1 1 1 1
H4 =
1 1 1 1
1 1 1 1
To see orthogonality of rows,
0
H4 H4T =
0
0
0
4
0
0
0
0
4
0
0
0
0
4
Due to the symmetric nature of H4 , shifts by TC /2 do not change orthogonality properties and crosscorrelation is 0 between any two users.
18.
2k
2
= 0.2512(6dB)
2 3N
k
= 1.9322
SIR = 2 (K1)
K = 40
N = 100
Pb = Q( 2rb ) = 0.0247
For synchronous users, the situation is worse (13.39)
P
19. r(t) = i i bi sci (t) + n(t)
RT
RT
P
rk = 0 b r(t)sck (t)dt = k bk kk + 0 b n(t)sck (t)dt + j6=k j bj kj
For 2 users, it is easy to see

r1
11 12 1 0
b1
n
=
+ 1
r2
21 22
0 2 b2
n2
We assume that noise statistics remain unchanged after despreading.
20. ML decision minimizes r = RAb + n

(r RAb)2
= rrT 2bT AT RT r + bT AT RT RAb
first term is same for all b , A is symmetric, R is symmetric
maximize 2bT ARr + bT AR2 Ab or minimize 2bT Ar bT ARAb
21. (a) If he transmits only one spreading sequence, there will be no other interference than noise. So
r
!
p
2P ()
BER = Q
=Q
2P ()
1
R bits/sec is achieved in this case.
If he transmits both with equal energy
BER = Q
P ()
1
2N P () + 1
2R bits/sec is achieved in this case

(b) Assume that
SN R0 is such that
Pb? = Q( SN R0 ) since we want to keep Pb Pb ?
S() = SN R0 for 0 1
P ()
= SN R0 for 1
1+ 1 P ()
2N
P () = 0 for 0
therefore
P () =
According to power constraint
0SN R
1 SN R0
1 SN R0
2N
0
0 1
1
P ()p()d = 1
0
Define K1 = SN R0
K2 =
power constraint will be
SN R0
SN R
1 2N 0
Z
K1
1
p()d + K2
and
0
P () =
(c)
K1
K2
1
p()d = 1
0
0 1
1
Z
J = Rp{0 1 } + 2Rp{1 } + K1
J
=0
0
J
=0
1
K1
0 =
R
(K2 K1 )
1 =
R
p(
d + K2
p(
d 1
Chapter 14
1. B=10 MHz
Btotal = [(10 + 2)100]M Hz = 1200M Hz
2. Btotal = 25M Hz
125 TDMA channels
8 user time slots per channel
R=270.833Kbps
Pre
26 training
bits
58 data bits
Trail
58 data bits
stop bits
start bits
(a)
8.25
data bits
as guard
time
(b)
2(58)
3+2(58)+26+3+8.25
= 0.7424
Information rate = 201.066 Kbps
(c) Frame duration = 8 timeslotduration

time slot duration = 2(3)+2(58)+26+8.25
= 576.92s
270.833103
Frame duration = 4.61 ms
Latency = 78 (frame duration) = 4.038 ms
3+8.25
(d) duration of guard band + stop bits = 270.83310
3 = 41.54s
max delay spread Tm < 41.54s for guard band to be useful.
3. B=10MHz
G=100
cross-correlation = 1/G
Interference limited system (N=0)
(a) SIR =
Pr
1
(k1)Pr
G
G
K1
(b) Pb = Q( 2b ) = 103
SIR = 3.0902
G
K = 1 + SIR
= b33.36c = 33
(c) B=10MHz
Information signal Bandwidth =
10M Hz
100
= 100KHz Total number of users = 100
G
= SIR
= 100 = 0.3269
(d) SIR
No, this is not a reasonable voice activity factor. It is too low.
4. (a) Since the system is interference limited
1
SIR = (m1)
= b
p
Pe = Q( (b )) (assuming binary signalling)
1
Pe = Q( m1
)
(b) Pe (more than one user occupy same freq band) = 1 Prob(only one user occupies it) Prob(no
user occupies
it)

K
K
(0.01)1 (0.99)K1
(0.99)K
= 1
1
0
(c)
Pe,avg
1
K
= 0.Prob (only one user) +
Q(
(0.01)m (0.99)Km
) =
m
m
1
m=2
K
X
1
K
=
Q(
)
(0.01)m (0.99)Km
m
m
1
m=2
K
X
5. pure Aloha : T= Le2L

T
2L (2) + e2L = 0
L = Le
2L
e
(2L + 1) = 0
L = 0.5 Tmax = 0.1839
2T
= (2L + 1)e2L (2) + e2L (2) = (4L 4)e2L
L2
2
L < 1, LT2 < 0 maxima
slotted Aloha : T = LeL
T
L (1) + eL = 0
L = Le
L = 1, Tmax = 0.3679
6. R=100 Mbps (pure Aloha)
N=103
= 103 packets per second
5
L = = 0.01,
T = Le2L = 0.0098
=N
R = 1 10 ,
effective data rate = RT=980.2 Kbps
L=2.835, also gives the same throughput
But psuccess earlier was 0.98 and now is 0.0034 (lots of packets are lost)
7. g1
g2
g3
r1?
=1
=3
=5
= r2? = r3? = 10,
n=1
(a) We expand
(I F )P
where
u=
? n T
N
1? n1 2? n2
N
,
,...,
g11 g22
gN N
(
0
k=j
Fkj =
k? gkj
k 6= j
gkk
P = (P1 P2 . . . PN )T
first row of (I-F)P

1? g12 P2 1? g13 P3
1? n1
g11
g11
g11
g11 P1 1? n1 + 1? g12 P2 + 1? g13 P3
g11 P1
1?
n1 + g12 P2 + g13 P3
= P1
which is precisely the SNR constraint for user 1.

We can similarly show this for users 2 and 3.
(b)
(
Fkj =
r1 g2
g1
r3 g1
g3
r2 g2
g3
r g
F = g22 1
r1 g3
g1
r2 g3
g2
rk? gj
gk
, if
, if
k=j
k 6= j
0
30 50
= 10/3 0 50/3 0.01
2
6
0
eigmax (F ) = 0.2 < 1

P ?exists
10 10 T
u = 10
1 , 3 , 5
12.5
P = (I F )1 u = 4.1667
2.5
8. P=10mW
B=100KHz
N0 = 103 W/Hz
P2
P1
, R2 = B log2 1 +
R1 = B log2 1 +
N0 B
N0 B + P1
CBC =
{P1 P2 :P1 +P2 =10mW }
MATLAB
P = 10e-3;
ss = P/1000;
B = 100e3;
N0 = 1e-9;
P1 = 0:ss:P;
for i = 1:length(P1)
P2(i) = P-P1(i);
R1(i) = B*log2(1+(P1(i)/N0*B));
R2(i) = B*log2(1+(P2(i)/(N0*B+P1(i))));
end
9. We show this result for 2 users
define gN10PB1 = x = nP11B
wherenk =
g2 P2
N0 B
= x = nP22B
say g1 > g2 , n1 < n2
n1
n2
N0
gk
<1
x 10
R (bps)
0.5
1.5
2.5
3.5
4
6
x 10
R1 (bps)
Figure 2: Problem 8
R1 = B log2 1 +
R2 = B log2 1 +
P1
n1 B
= B log2 (1 + x)
y
P2
=
B
log
1
+
2
n2 B+P1
1+x
y
R1 + R2 = Blog2 (1 + x) + B log2 1 + 1+x
1+x
< B log2 1 + x + y = B log2 1 + nP1 B
( 1)
R1 + R2 = B log2 1 + x + y 1+x
Hence we see that the sum rate for any P1 , P2 assignment is always less than or equal to the rate
achievable if all power is assigned to better user.
10. Based on the previous part; we know that the sum-rate maximizing strategy is to assign all power to
a single user at a time instant. The problem then reduces to water-filling for a single user, but the
fading statistic is given at each instant as
nmin (i) = min{n1 (i, n2 (i), . . . nk (i))}

The cut-off is determined by distribution of nmin (i)
1
P ()
0
=
0
P
=
P
nmin B
for k users
0
< 0
(distribution of is dictated by the distribution of nmin )

Z
1
1
11.
n=
1mW
5mW
nmin =
1mW
5mW
p()d = 1
w.p. 1/2
w.p. 1/2
w.p. 3/4
w.p. 1/4
Now we just treat it as a single user channel
10 w.p. 3/4
=
2 w.p. 1/4
If 0 < 2
1
1
0 1
p(1 ) +
1
1
0 2
p(2 ) = 1
1
p(1 ) p(2 )
=
+
+1
0
1
2
0 = 0.833 < 2 , so it is ok
(
1.2 11 = 1
P ()
=
1.2 12 = 2
P

1
0
Sum capacity CSR = B log2

12. = NP0 B = 10
1 1 e/
2 1 e/
max = max(1 , 2 )
P (1 ) = 11mW, P (2 ) = 7mW

p(1 ) + B log2 20 p(2 ) = 3.005B
1 e/ e/
[From (7.9) in Reader]
P () =
1
0
0
0
1
1
p()d = 1
0
0
1
1 2
1 e/ e/ d = 1
0
0
Z
Z
CSR = B
= 10 0 = 0.89 Using Matlab
1.1236 1 > 0.89

P () =
0
< 0.89
log2 (/0 ) p()d = 366.27Kbps (Using Matlab)
MATLAB
gammab = 10;
ss = .01;
gamma0 = [ss:ss:1e4];
i = 1;
gamma = [gamma0(i):ss:gamma0(length(gamma0))];
gammavec=(1/gamma0(i)-1./gamma)*(2/gammab).*(1-...
exp(-gamma/gammab)).*exp(-gamma/gammab);
sumcheck = sum(gammavec)*ss;
while sumcheck > 1

i = i + 1;
gamma = [gamma0(i):ss:gamma0(length(gamma0))];
gammavec = (1/gamma0(i)-1./gamma)*(2/gammab).*...
(1-exp(-gamma/gammab)).*exp(-gamma/gammab);
sumcheck = sum(gammavec)*ss;
end
gamma0ch = gamma0(i);
gamma = [gamma0ch:ss:gamma0(length(gamma0))];
gammavec=log2(gamma/gamma0ch).*(2/gammab).*(1-...
exp(-gamma/gammab)).*exp(-gamma/gammab);
Csr = B*sum(gammavec)*ss;
13. Consider the ith user
Ri = Bi log2 1 +
with time-division
Ri = Bti log2 1 +
Pi
N0 Bi
Pi
ti N0 Bi
ti =
14. B=100KHz
n1 =
n2 =
(a)
0 Bi B
0 ti 1
Bi
B
105 W/Hz
w.p. 3/4
2 105 W/Hz w.p. 1/4
105 W/Hz
w.p. 1/2
5
2 10 W/Hz w.p. 1/2
(1, 1)105 W/Hz
(1, 2)105 W/Hz

(n1 , n2 ) =
(2, 1)105 W/Hz
(2, 2)105 W/Hz
w.p.
w.p.
w.p.
w.p.
P = 10; n1 = 1e-5; n2 = 2e-5; B = 100e3;

tau = [.01:.01:1-.01]; for k = 1:length(tau)
P1 = [0.1:.1:P/tau(k)-.1];
P2 = (P-tau(k)*P1)/(1-tau(k));
%% For User 1
gamma1 = P1(i)/(n1*B);
gamma01try = 1/(1+(.75/gamma1)+(.25/gamma2));
gamma02try = .75/(1+(.75/gamma1));
if gamma01try < gamma2
gamma0ch = gamma01try;
3/8
3/8
1/8
1/8
C1(i) = tau(k)*B*(log2(gamma1/gamma0ch)*.75+log2(gamma2/gamma0ch)*.25);
elseif (gamma02try > gamma2) & (gamma02try < gamma1)
C1(i) = tau(k)*B*.75*log2(gamma1/gamma0ch);
else
C1(i) = 0;
end
end
%% For User 2
gamma01try = 1/(1+(.5/gamma1)+(.5/gamma2));
gamma02try = .5/(1+(.5/gamma1));
if gamma01try < gamma2
C2(i) = (1-tau(k))*B*(log2(gamma1/gamma0ch)*.5+log2(gamma2/gamma0ch)*.5);
elseif (gamma02try > gamma2) & (gamma02try < gamma1)
C2(i) = (1-tau(k))*B*.5*log2(gamma1/gamma0ch);
else
C2(i) = 0;
end
end
plot(C1,C2); hold on;
clear C1 C2
end
%% Now we use superposition coding and successive interference cancellation
P1 = 5; P2 = 5; C1 = .375*B*log2(1+(P1/(n1*B))) +
.375*B*log2(1+(P1/(n1*B))) + .125*B*log2(1+(P1/(n2*B+P2))) +
.125*B*log2(1+(P1/(n2*B+P2))); C2 = .125*B*log2(1+(P2/(n1*B))) +
.125*B*log2(1+(P2/(n2*B))) + .375*B*log2(1+(P2/(n2*B+P1))) +
.375*B*log2(1+(P2/(n1*B+P1))); plot(C1,C2,ro);
(b) Consider an assignment P1 , P2 & s.t
P1 + (1 )P2 = P
For user 1:
1 =
P1
,
n1 B
2 =
P1
n2 B
If 1 < 2
1
1
1
1
p(1 ) +
p(2 ) = 1
0 1
0 2
1
2
C1 = B log2
p(1 ) + B log2
p(2 )
0
0
3.5
x 10
2.5
C2
1.5
0.5
0.5
1.5
2.5
3.5
5
x 10
C1

for 2 < 0 < 1
1
1
0 1
p(1 ) = 1
C1 = B log2
1
0
p(1 )
for 0 > 1
We have similar equations for user 2
See Matlab for a sketch of capacity region
To achieve a point outside the TD-VP region we use code-division with interference cancellation
and superposition coding.
take the triplet
P1 = 5
N1 = 1 105
P2 = 5
N2 = 2 105
w.p. 3/8
P1
C1 = B log2 1 +
N1 B
P2
C2 = B log2 1 +
N1 B + P1
w.p. 3/8
P1
C1 = B log2 1 +
N1 B
P2
C2 = B log2 1 +
N2 B + P1
w.p. 1/8
C1 = B log2 1 +
P1
N2 B + P2
P2
C2 = B log2 1 +
N1 B
w.p. 1/8
C1 = B log2 1 +
P1
N2 B + P2
P2
C2 = B log2 1 +
N2 B
i = 1; snr = 10.^([5 10 20]/10); R = 100e3*log2(1+snr); ic = 10;

Tcur = zeros(1,3); Rcur = zeros(1,3); while i < 1e4
R1(i) = R(floor(unifrnd(1,4)));
Rcur(i,:) = [R1(i) R2(i) R3(i)];
if i == 1
[a,b] = max(Rcur(i,:));
Tcur(i+1,b) = Rcur(i,b);
else
Rcur_chk = Rcur(i,:)./(Tcur(i,:)+1e-10);
[a,b] = max(Rcur_chk);
Tcur(i+1,:) = (1-1/ic)*Tcur(i,:);
Tcur(i+1,b) = (1-1/ic)*Tcur(i,b) + (1/ic)*Rcur(i,b);
end
Rgot(i,b) = Rcur(i,b);
i = i+1;
end Rmean = [sum(Rgot(:,1))/i sum(Rgot(:,2))/i sum(Rgot(:,3))/i]
Rmeansys = sum(Rmean)
(c) B1 = B2 = B/2
P1 = P2 = P/2
For user 1
1 =
P1
= 10,
n1 B1
2 =
P1
=5
n2 B1
If 0 < 2
0 = 0.8889 < 2 = 5
C1 = 1.6209 105
For user 2
1 = 10 2 = 5
If 0 < 2
0 = 0.8696 < 2 = 5
C1 = 1.5118 105
(d) Consider the worst channel condition
(n1 , n2 ) = (2, 2) 105
P1
C1 = B log2 1 +
n1 B
P2
C2 = B log2 1 +
n2 B + P1
P1
P P1
C1 = C2 = log2 1 +
= log2 1 +
nB
nB + P1
P P1
P1
=
nB
nB + P1
P12 + 2P1 (nB) P nB = 0

P1 = 2.889
we get C1 = C2 = 1.2925 105
15.
(
CM AC =
(R1 R2 . . . Rk ) :
X
kS
)
P
kS gk Pk
S {1, . . . k}
Rk B log2 1 +
N0 b
Scale gk by ,
Pk by 1/
Since capacity region depends on gk Pk , it remains unchanged.
16. B = 100KHz
P1 = 3mW
P2 = 1mW
N0 = 0.001W/Hz
See Fig 4
y
(0,C )
2
(C *, C )
1 2
Code division with SIC
(C 1 , C *)
2
Time division
Code division without SIC

(C,1 ,0)
(a) C1 = B log2 1 + NP01B = 4.95B
C2 = B log2 1 + NP02B = 3.46B

R1 = 3B In TD rates lie on the straight line joining C1 & C2 , so
R1 R2
+
=1
C1
C2
3B
R2
+
=1
4.95B 3.46B
R2 = 136.30Kbps
With superposition
coding
and successive interference cancellation we have,

P1
?
C1 = B log2 1 + N0 B+P2 = 1.898B

C2? = B log2 1 +
P2
N0 B+P1
C2 C2?
?
C1? C1 (x C1 )
= 0.4B
(y C2 ) =
x = R1 = 3B
y = R2 = 235.51Kbps
1
(b) R1 = B1 log2 1 + NP0 B
1
P2
R2 = B2 log2 1 + N0 B2 For the point that touches the capacity region, we have:
B1
P1
3
=
=
B2
P2
1
B1 = 3B2
B2 = 25KHz
B1 = 75KHz
R1 = 401.816Kbps
R2 = 133.939Kbps
(c) For time-division
y
x
+
=1
C1 C2
R1 = R2 = R
R
R
+
=1
C1 C2
R1 = R2 = 203.65Kbps
With code division and successive interference cancellation
G=10
Bs = Bsignal = B/10 = 10KHz
P1
= 8.234Bs
C1 = Bs log2 1 +
N0 Bs
P2
C2 = Bs log2 1 +
= 6.658Bs
N0 Bs
P1
?
C1 = Bs log2 1 +
= 4.821Bs
N0 Bs + P2 /G
P2
?
C2 = Bs log2 1 +
= 2.079Bs
N0 Bs + P1 /G
C2 C2?
(y C2 ) = ?
(x C1? )
C1 C1
x=y=R
(R C2 ) = m(R C1? )
mC1? +C2
R = 1m
= 5.605Bs
R = 56.05Kbps
with code division and without interference cancellation
R1 = R2 = min(C1? , C2? ) = 20.79Kbps
17. The sum rate of a MAC is given by
CM AC,SR = B log2
Pk
k=1 gk pk
1+
N0 B
where Pk is the maximum power at user k. Clearly to achieve CM AC,SR , each user must transmit at
Pk
18.
"
CM AC,SR =
max eG B log2
PK
1+
P FM AC
k=1 gk Pk (g)
N0 B
!#
We know that for sum-rate we can treat the two users as a single user with a common power
g1 P1 + g1 P2
N0 B
g1 p(g1 )
g2 p(g2 )
p()
Find 0 s.t.
1
1
p()d = 1
0
0

Z
0
p()d
CM AC,SR =
B log2
0
1
1
P
0 0
=
0
o.w.
P
Z
where individual power adaptations are given as
19. P 1 = P 2 = 10mW
P1 =
P
P1
P
P2 =
P
P2
P
n1 = n2 =
P1
1 =
=
n1
P2
=
n2
1mW
5mW
w.p. 1/2
w.p. 1/2
10 w.p. 1/2
2 w.p. 1/2
10 w.p. 1/2
2 w.p. 1/2
20 w.p. 1/4
12 w.p. 1/2
= 1 + 2 =
4 w.p. 1/4
2 =
CM AC,SR = B log2 (1 + )
say 0 < 4
3
X
1
1
p(i ) = 1
0 i
i=1
0 = 0.8955
CM AC,SR,max = B
3
X
log2 (i /0 ) p(i ) = 3.5322B
i=1
20. 3 user
B=100KHz
5dB w.p. 1/3

10dB w.p. 1/3
=
20dB w.p. 1/3
where r is the SNR distribution for any user if it were the only user present.
(a) If the best user is always picked, the system is equivalent to a single user with fading statistics
governed by the max at each instant, max is distributed as
5dB w.p. 1/27

10dB w.p. 7/27
max =
20dB w.p. 19/27

If 0 < 3.1623 (5dB)
0 = 0.9572
(which is ok)
CBC,SR = 5.6612B = 566.12Kbps

(b) MATLAB CODE
i = 1;
snr = 10.^([5 10 20]/10);
R = 100e3*log2(1+snr);
ic = 10;
Tcur = zeros(1,3);
Rcur = zeros(1,3);
while i < 1e4
Rcur(i,:) = [R1(i) R2(i) R3(i)];
if i == 1
[a,b] = max(Rcur(i,:));
Tcur(i+1,b) = Rcur(i,b);
else
Rcur_chk = Rcur(i,:)./(Tcur(i,:)+1e-10);
[a,b] = max(Rcur_chk);
Tcur(i+1,:) = (1-1/ic)*Tcur(i,:);
Tcur(i+1,b) = (1-1/ic)*Tcur(i,b) + (1/ic)*Rcur(i,b);
end
Rgot(i,b) = Rcur(i,b);
i = i+1;
end
Rmean = [sum(Rgot(:,1))/i sum(Rgot(:,2))/i sum(Rgot(:,3))/i]

Rmeansys = sum(Rmean)
i=1000
ic = 1,
CBC,SR = 509.28Kbps
Rmean = 1.9385 1.7655 1.3887 105 bps

ic = 5,
CBC,SR = 536.13Kbps
Rmean = 1.7848 1.7818 1.7947 105 bps

ic = 10,
CBC,SR = 558.98Kbps
Rmean = 1.8728 1.8621 1.8549 105 bps

Notice that as ic Rmean becomes equal for all users.
Chapter 15
1. City has 10 macro-cells

each cell has 100 users
total number of users = 1000
Cells are of size 1 sqkm
maximumdistance traveled to traverse = 2km

time = 302 = 169.7s
In the new setup
number of cells = 105 microcells
total number
of users = 1000 1002 users
210
time = 30103 =1.69s
number of users increases by 10000 and handoff time reduces by 1/100
2. See Fig 1
u
(1,1)
60
(1,0)
(0,1)
(0,0)
Figure 1: Problem 2
D2 = (j20)2 + (i20)2 2(j20)(i20) cos(2/3)

p
p
D = 2a i2 + j 2 + ij = 3R i2 + j 2 + ij
3. diamond shaped cells, R= 100m
Dmin = 600m
D = 2KR
D
600
K = 2R
= 2100
=3
2
N =K =9
(a) number of cells per cluster = N = 9
(b) number of channels per cell = total number/N = 450/9 = 50
4. (a) R=1km
D=6km
2
3D /2
cluster
N = AA
=
= 31 (D/R)2 = 13 62 = 12
cell
3 3R2 /2
number of cells per cluster = N = 12
SHADED CELLS USE

SAME FREQUENCY
D
Figure 2: Problem 3
(b) number of channels in each cell = 1200/12 = 100
p
(c) i2 + j 2 + ij = 2 3 i = 2, j = 2
5. R=10m
D=60m
I = 2
0 = 4
M = 4 for diamond shaped cells
SIRa =
RI
R2
=
= 32400
MD 0
4D4
R4
= 324
4D4
R2
=9
SIRc =
4D2
SIRa > SIRb > SIRc
SIRb =
6. = 2
BP SK
Pb = 106 Pb = Q( 2b ) b = SIR0 = 4.7534

B = 50M Hz
each user100KHz
= Bs

1 D
SIR = M
M=6
for hexagonal cells
R
a1 = 0.167
a2 = 3
2/
0
N > a12 SIR
N 9.4879 N = 10
a1
Cu = 50
7. G = 100
=1
= 1.5
With no sectorization
SIR =
3G (Nc
1)(1 + )
= 4.7534
Nc = b26.2450c = 26
With sectorization, interference is reduced by a factor of 3
Nc = b76.7349c = 76
8. SINR =
G
PNc 1
Xi +N
i=1
= p(Xi = 1)
N G(0.247Nc , 0.078Nc )
Po ut = p(SIR < SIR0 )
(a) Pout = p PNc 1G

i=1
Xi +N
< SIR0
=p
P
Nc 1
i=1
Xi + N >
G
SIR0
PNc 1
(b) X = i=1
Xi then X Bin(, Nc 1)
P c 1
p(x + N > G/SIR0 ) = N
N =0 p(n + N > G/SIR0 |x = n)p(x = n)
Nc 1
p(x = n) =
n (1 )Nc 1n
n
p(x + N > G/SIR0 ) =
NX
c 1
p(N > G/SIR0 n|x = n)p(x = n)
N =0
NX
c 1
G
n 0.247Nc
N 0.247Nc
> SIR0
|x = n
=
p
0.078N
0.078N
c
c
N =0
G
!
NX
c 1
0.247N
c
p(x = n)
=
Q SIR0
0.078Nc
N =0
(c) Nc = 35
= 0.5
SIR0 = 5
G = 150
p = 0.0973
MATLAB
for i = 1:length(n)
pn(i) = (factorial(Nc-1)./(factorial(n(i)).*factorial(Nc-1-n(i))))...
*alpha.^n(i)*(1-alpha).^(Nc-1-n(i));
end
sump = 0; for i = 1:length(n)
f = ((G/sir0)-n(i)-.247*Nc)/(sqrt(.078*Nc));
sump = sump + .5*erfc((f)/sqrt(2))*pn(i);
end
(d) If x can be approximated as Gaussian then
x G((Nc 1), (Nc 1)(1 ))
x + N G(0.247Nc + (Nc 1), 0.078Nc + (Nc 1)(1 ))
G
!
SIR0 (0.247Nc + (Nc 1))
p(x + N > G/SIR0 ) = Q p
0.078Nc + (Nc 1)(1 )
(e) p= 0.0969 (very accurate approximation!)
9. define
k =
nk +
gk Pk
P
k6=j
gkj pj
k, j {1, . . . K}
where,
gk is channel power gain from user k to his base station nk is thermal noise power at user ks base
station
is interference reduction factor ( 1/G)
gkj is channel power gain from j th interfering transmitter to user ks base station
pk is user ks Tx power
pj is user js Tx power
define a matrix F such that
(
0
k=j
Fkj =
k? gkj
k 6= j
gk
k, j {1, . . . K}
u=
? nK
1? n1 2? n2
,
,..., K
g1
g2
gK
If Perron Ferbinius eigenvalue of F is less than 1 , then a power control policy exists. The optimal
power control policy is given to be P ? = (I F )1 u
10. Matlab
D = 2:.01:10;
R = 1;
gamma = 2;
Pdes = R^(-gamma);
for i = 1:length(D)
Pint = 6*(.2*(D(i)-R)^(-gamma)+.2*(D(i)-R/2)^(-gamma)+.2*(D(i))^(-gamma)...
+.2*(D(i)+R/2)^(-gamma)+.2*(D(i)+R)^(-gamma));
Pintbest = 6*((D(i)+R)^(-gamma));
Pintworst = 6*((D(i)-R)^(-gamma));
ASE(i) = log(1+Pdes/Pint)/(pi*(.5*D(i))^2);
ASEbest(i) = log(1+Pdes/Pintbest)/(pi*(.5*D(i))^2);
ASEworst(i) = log(1+Pdes/Pintworst)/(pi*(.5*D(i))^2);
end
0.35
5 Case ASE
Best Case ASE
Worst Case ASE
0.3
0.25
ASE(D)
0.2
0.15
0.1
0.05
10
11. Pt = 5W
B = 100KHz
N0 = 1016 W/Hz
3
Pr = Pt K dd0
d0 = 1, K = 100
(a) D=2R
2 users share the band available
Each user gets 50KHz
R=1Km
BASESTATION
USER

(b) P b = 41 103 = 41 b = 250
b
b
If D(n) = 2nR, number of users that share band = 2(2(n-1)+1)
each user gets 100KHz
2(2n1) = Bu (n)
interference is only from first tier
Pt K
SIR(n) =
N0
2 Bu (n)
using Matlab ,
Matlab
Pt = 5;
R = 1000;
sigma_2 = 1e-16;
n = 1;
n = 4,
+ 2 Pt K
SIR = 261.9253,
d0
R
d0
R2 +D(n)2
D = 8R
3 ! > 25
D = 2*n*R;
Bu = (100/(2*(2*n-1)))*1e3;
K = 100;
d0 = 1;
Pdes = Pt*K*(d0/R)^3;
Pint = 2*(Pt*K*(d0/sqrt(R^2+D^2))^3);
Npower = sigma_2*Bu;
sir = Pdes/(Npower+Pint);
while sir < 250
n = n+1;
D = 2*n*R;
Bu = (100/(2*(2*n-1)))*1e3;
K = 100;
d0 = 1;
Pdes = Pt*K*(d0/R)^3;
Pint = 2*(Pt*K*(d0/sqrt(R^2+D^2))^3);
Npower = sigma_2*Bu;
sir = Pdes/(Npower+Pint);
end
1 +R2 )/B
(c) ASE = (R
2km2km
R1 = R2 = Bu (1) log(1 + SIR(1))
Bu (1) = 50KHz, SIR(1) = 5.5899
ASE = 0.6801bps/Hz/km2
12. B=100KHz
N0 = 109 W/Hz
K = 10
P = 10mW per user
(a) 0 1
is channel gain between cells.
See Matlab
If is large, interference can be decoded and subtracted easily so capacity grows with as high
SNRs (beyond an value) .
For low SNR values ( less than a value) c decreases with increase in as interference is increased
which cannot be easily decoded due to low SNR.
MATLAB CODE:
B = 100e3;
sigma_2 = 1e-9;
P = 10e-3;
K = 10;
ss = .001;
alpha = 0:.01:1;
theta = 0:ss:1;
for i = 1:length(alpha)
capvec = log2(1+(K*P*(1+2*alpha(i)*cos(2*pi*theta)).^2)/(sigma_2*B));
C(i) = (1/K)*sum(capvec)*ss;
end
B = 100 KHz
1.02
0.98
0.96
C()/B
0.94
0.92
0.9
0.88
0.86
0.84
0.82
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9

(b) C(K) as K because as the number of mobile per cell increases system resources get shared more
and so per user capacity C(K) has to fall.
MATLAB CODE:
BB = 100e3;
sigma_2 = 1e-9;
P = 10e-3;
K = 1:.1:30;
ss = .001;
alpha = .5;
theta = 0:ss:1;
for i = 1:length(K)
capvec = log2(1+(K(i)*P*(1+2*alpha*cos(2*pi*theta)).^2)/(sigma_2*B));
C(i) = (1/K(i))*sum(capvec)*ss;
end
B = 100 KHz
6
C(K)/B
10
15
20
25
30

(c) as transmit power P , capacity C but gets saturated after a while as the system becomes
interference limited.
MATLAB CODE:
B = 100e3;
sigma_2 = 1e-9;
P = [0:.1:100]*1e-3;
K = 10;
ss = .001;
alpha = .5;
theta = 0:ss:1;
for i = 1:length(P)
capvec = log2(1+(K*P(i)*(1+2*alpha*cos(2*pi*theta)).^2)/(sigma_2*B));
C(i) = (1/K)*sum(capvec)*ss;
end
B = 100 KHz
1.4
1.2
C(P)/B
0.8
0.6
0.4
0.2
10
20
30
40
50
60
70
80
P (in mW)
Figure 7: Problem 12c
90
100
Chapter 16
1. d=1Km , P = Pt d
(a) Pt = P d
Pt ( = 2) = 10KW
Pt ( = 2) = 1 1010 W
(b) d= 500m
Pt ( = 2) = 2.5KW
Total power is simply twice.
Ptotal = 5KW
Pt ( = 4) = 6.25 108 W
Total power Ptotal = 1.25 109 KW
(c)
Pt ( = 2) = 10 10
Pt ( = 4) = 10 10
1
103
N +1
1
103
N +1
Pt ( = 2) = (N + 1)
Pt ( = 4) = (N + 1)
1 1010
(N + 1)4
1 104
1 104
=
(N + 1)2
(N + 1)
1 1010
1 1010
=
(N + 1)4
(N + 1)3
(a) We expand
(I F )P
where
u=
1 104
(N + 1)2
2. 1 = 2 = 7dB
3 = 10dB
=1
ni =
1i
1 0.06 0.04
0.9 0.126
G = 0.09
0.064 0.024
0.8
? n T
N
1? n1 2? n2
N
,
,...,
g11 g22
gN N
(
0
k=j
Fkj =
k? gkj
k 6= j
gkk
P = (P1 P2 . . . PN )T
first row of (I-F)P

= P1
1? g12 P2 1? g13 P3
1? n1
g11
g11
g11
g11 P1 1? n1 + 1? g12 P2 + 1? g13 P3
g11 P1
1?
n1 + g12 P2 + g13 P3
which is precisely the SNR constraint for user 1.

We can similarly show this for users 2 and 3.
0 0.3007 0.2005
0 0.7017 From Matlab:
(b) F = 0.5012
0.8
0.3
0
= abs(eig(F )) = (0.8667 0.4791 0.4791)
As max(||) = 0.8667 < 1 a feasible power vector for the system exists.
(c) u = [5.0119 5.5687 12.5]
P ? = (I F )1 u = [39.7536 71.6634 65.8019]
3. P(0) = [50 50 50]
MATLAB CODE
P = [50 50 50];
k = 1;
while k<50
for i = 1:3
sum_int = 0;
for j = 1:3
if i~=j
sum_int = sum_int + rho*(G(i,j)*P(j));
end
end
gamma(i) = (G(i,i)*P(i))/(n(i)+sum_int);
end
P_plot(:,k) = P(:);
Gamma_plot(:,k) = gamma(:);
k = k+1;
for i = 1:3
P(i) = (gamma_des(i)/gamma(i))*P(i);
end
end
50
48
46
44
42
40
38
36
34
32
30
10
15
20
25
30
35
40
45
50
Figure 1: Problem 3:Power, User 1

75
70
65
60
55
50
10
15
20
25
30
35
40
45
50

68
66
64
62
60
58
56
54
52
50
10
15
20
25
30
35
40
45
50
8.5
7.5
6.5
5.5
4.5
3.5
10
15
20
25
30
35
40
45
50
Figure 4: Problem 3:SNR, User 1

5.4
5.2
4.8
4.6
4.4
4.2
3.8
10
15
20
25
30
35
40
45
50

12
11.5
11
10.5
10
9.5
8.5
7.5
10
15
20
25
30
35
40
45
50
4. (a) = 2 , for N= 2 , we want P (10m) = 10mW which will ensure that P (d > 10m) < 10mW
Pmax (10)2 = 10 103
Pmax = 1W
(b) = 2 , N= 4
P (20m) = 10mW
Pmax (20)2 = 10 103
Pmax = 4W
(c) = 4 , N= 4
P (20m) = 10mW
Pmax (20)4 = 10 103
Pmax = 1600W
5. MATLAB CODE:
clear;
Ra = 20:-1:1;
for j = 1:length(Ra)
for k = 1:100
no_of_Cluster = 1;
node(1,1,no_of_Cluster) = unifrnd(0,100);
R = Ra(j);
no_of_nodes = size(node,1);
while (no_of_nodes == 1) | (no_of_Cluster > 1)
no_of_Cluster = no_of_Cluster + 1;
no_of_nodes = no_of_nodes + 1;
new_node = reduce_clusters(node,R);
node = new_node;
no_of_Cluster = size(node,3);
end
no_of_nodes_this_run(k) = no_of_nodes;
clear node;
end
des_ans(j) = mean(no_of_nodes_this_run);
end
d/D
6. p(d) = e D d > 0
Prob that one of the copies arrives after D = prob(dD)
Z
1 x/D
=
e
dx = e1
D
D
Prob that all N copies arrive after D assuming independence = eN
N=1, Prob = 0.3679
N=5, Prob = 0.0067
Since the paths were identical(had similar delay profile), throughput goes down by a facto of 5 as we
are sending same information on all paths.
As we try to decrease delay, the throughput also goes down.
Delay is proportional to throughput. This is the trade-off
ij
7. Dij = cij
The function is linear in fij , so it is both convex and concave. Specifically, it is convex.
Dij
fij
= 2
cij
cij
fij
2 Dij
= 2 3 > 0 for fij , cij > 0
2
cij
cij
Hence it is convex in cij too.
8. C=10Mbps
4
10
D = F/C
D = F/(CF)
3
10
10
Delay
10
10
10
10
10
10
10
6
F (flow)
x 10
Figure 7: Problem 8
9. =
fij /cij fij

fij /cij
(a) x =
=
fij
cij
1
1x
cij
cij fij
0x<1
(b) 1 <
(c) > 10
1
1x
> 10 x > 9/10 fij > 0.9cij
(d) Network for which delay is calculated based on (16.6) will be more congested as the metric will
tell us that the delay less (say below a threshold fixed apriori) and so we will keep on placing more
traffic.
10. (a) Since there is full cooperation, we can adapt R based on the values of C.
Using Matlab
Ropt = [29.322.336.3]Kbps
Dmin = 0.993
(b) Dmin = 1.3028
(c) Notice that even with a lower data rate, we get 31 % increase in average distortion without
cross-layer design
MATLAB
clear D0 = .38;
R0 = 18.3e3;
Theta = 2537;
K = 1;
L = 3040;
T = 350e-3;
C = [45 24 60]*1e3;
p = [.5 .25 .25];
count = 1;
ss = .1;
Rpos1 = [(R0/1e3)+ss:ss:(C(1)/1e3)-ss]*1e3;
Rpos2 = [(R0/1e3)+ss:ss:(C(2)/1e3)-ss]*1e3;
Rpos3 =[(R0/1e3)+ss:ss:(C(3)/1e3)-ss]*1e3;
for i1 = 1:length(Rpos1)
R(1) = Rpos1(i1);
R(2) = Rpos2(i2);
R(3) = Rpos3(i3);
Rv(count,:) = R;
AvgDist(count) = 0;
for i = 1:length(C)
Dist = D0+(Theta/(R(i)-R0))+K*exp(-(C(i)-R(i))*(T/L));
AvgDist(count) = AvgDist(count) + Dist*p(i);
end
count = count+1;
end
end
end
[AvgDistmin,d] = min(AvgDist) Ropt = Rv(d,:)
11.
2C/B 1
Eb
(CB ) =
N0
C/B
As C , numerator exponentially whereas denominator only linearly.
B
Eb
2C/B 1
(B) = B
N0
C
ENB0 (B)
B
2C/B ln2 2C/B

1
+
BC
C
C
2C/B > 1forC, B > 0
Eb
(B) as B for C fixed
N0
Eb
N0 (CB )
as C for fixed

Goldsmith Solution

Uploaded by

Copyright:

Available Formats

Goldsmith Solution

Uploaded by

Document Information

Original Description:

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

Goldsmith Solution

Uploaded by

Copyright:

Available Formats

Chapter 1

= 5 1011 (very high)

4. Geo: 35,786 Km above earth RT T = 23578610

0.8 0.2 $0.2

0.8 0.2 $0.4 $0.6

(d) Macrocell: $50 M

Attenuation is very high for high frequencies

d ht , hr , we need to keep only first order terms

4. Signal nulls occur when = (2n + 1)

Pref lection = 54.54dB

dmin = 2.41km with the same argument dmin = 4km

= 10 log10 K 10r log1 0 dd0

(b) PL(2Km) = 10 log10 K 10r log10 d = 161.76dB

P rob(X < 10) = P rob

(d) = Precvd (due to path loss alone)

P (Precd (d) > 55) = 0.9

20. MATLAB CODE

(b) = 4dB, outage prob < 1%

To perform integration by parts, we let du = rdr and v = Q a + b ln Rr . Then u = 12 r2 and

where k = a + b ln Rr . Then we get

Since Q(x) = 1 Q(x).

1 (t) = 2fc 1 (t) D1

So, when c = 0 , Tm = 44 ns. When c = 1 , Tm = 19 ns.

5. Use CDF strategy.

6. For Rayleigh fading channel

2 2 = 80dBm, P0 = 95dBm, P r(Pr < P0 ) = 0.0311

0.01 = 1 eP0 /Pr

9. CDF of Ricean distribution is

P [Pr,1 T , Pr,2 < T |Y = y] fy (y) dy

Pr,1 |Y = y ~ N W + by,a2 2 , and [Pr,1 |Y = y] [Pr,2 |Y = y]

, = 8, 4 = 5. With independent fading we get

With correlated fading we get Pout = 0.1316.

Minumum required sampling rate

% Ts contains the time instances at which z_dB is specified

% Use even number of samples in calculation

Frequency samples used in calculation

% Generate complex Gaussian samples with line spectra in frequency domain

p(z > Z) = 1 p(z Z) = 1 (1 e ) = e

Nj level crossing rate at level Aj

The antenna setup is shown in Fig. 15

dc = 768 m. Since d dc , power fall-off is proportional to d2 .

Figure 4: Antenna Setting

Figure 5: Time Axis for Ray Arrival

(c) Avg Delay Spread =

(d) u > Coherence BW Freq. Selective Fading

= 105 u > = 105 kHz

Figure 6: Scattering Function

Figure 1: Capacity vs Pout

Assume all channel states are used

Now assume only the best 3 channel states are used

Construct the Lagrangian function

Evaluating this with p() =

(b) Capacity can be computed as

To compute 0 we use the power constraint as:

First assume that 0 < 0.0625, then we have

This gives us, C = 23.4 Mbps.

To compute 0 we use the power constraint as:

First assume that 0 < 0.0625, then we have