This action might not be possible to undo. Are you sure you want to continue?
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. P
b
= 10
−12
1
2γ
= 10
−12
γ =
10
12
2
= 5 ×10
11
(very high)
4. Geo: 35,786 Km above earth ⇒RTT =
2×35786×10
3
c
= 0.2386s
Meo: 8,000 20,000 Km above earth ⇒RTT =
2×8000×10
3
c
= 0.0533s
Leo: 500 2,000 Km above earth ⇒RTT =
2×500×10
3
c
= 0.0033s
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)
⇒ 0.5
2
2
1
×$1 + 0.5
2
×$1 +
6
1
0.8 ×0.2
5
×$0.2 +
6
2
0.8
2
×0.2
4
×$0.4+
1 −
6
1
0.8 ×0.2
5
×$0.2 −
6
2
0.8
2
×0.2
4
×$0.4
×$0.6
⇒$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. Handoﬀ becomes a big problem.
2. Intercell 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
100km
2
Area/cell
= 2500cells ⇒
Area
cell
= .04km
2
(c) From Rappaport or iteration of formula, we get that 100
channels
cell
⇒ 89
channels
cell
@P
b
= .02
Each subscriber generates
1
30
of an Erlang of traﬃc.
Thus, each cell can support 30 ×89 = 2670 subscribers
Macrocell: 2670 ×100 ⇒267, 000 subscribers
Microcell: 6,675,000 subscribers
(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 W
imax
∼ 40Mbps
∴ number of CDPD lines ∼ 2 ×10
3
Chapter 2
1.
P
r
= P
t
_√
G
l
λ
4πd
_
2
λ = c/f
c
= 0.06
10
−3
= P
t
_
λ
4π10
_
2
⇒P
t
= 4.39KW
10
−3
= P
t
_
λ
4π100
_
2
⇒P
t
= 438.65KW
Attenuation is very high for high frequencies
2. d= 100m
h
t
= 10m
h
r
= 2m
delay spread = τ =
x+x
−l
c
= 1.33×
3. ∆φ =
2π(x
+x−l)
λ
x
+x −l =
_
(h
t
+h
r
)
2
+d
2
−
_
(h
t
−h
r
)
2
+d
2
= d
_
_
¸
_
h
t
+h
r
d
_
2
+ 1 −
¸
_
h
t
−h
r
d
_
2
+ 1
_
_
d h
t
, h
r
, we need to keep only ﬁrst order terms
∼ d
_
_
_
_
_
1
2
¸
_
h
t
+h
r
d
_
2
+ 1
_
_
−
_
_
1
2
¸
_
h
t
−h
r
d
_
2
+ 1
_
_
_
_
_
=
2(h
t
+h
r
)
d
∆φ ∼
2π
λ
2(h
t
+h
r
)
d
4. Signal nulls occur when ∆φ = (2n + 1)π
2π(x
+x −l)
λ
= (2n + 1)π
2π
λ
_
_
(h
t
+h
r
)
2
+d
2
−
_
(h
t
−h
r
)
2
+d
2
_
= π(2n + 1)
_
(h
t
+h
r
)
2
+d
2
−
_
(h
t
−h
r
)
2
+d
2
=
λ
2
(2n + 1)
Let m = (2n + 1)
_
(h
t
+h
r
)
2
+d
2
= m
λ
2
+
_
(h
t
−h
r
)
2
+d
2
square both sides
(h
t
+h
r
)
2
+d
2
= m
2
λ
2
4
+ (h
t
−h
r
)
2
+d
2
+mλ
_
(h
t
−h
r
)
2
+d
2
x = (h
t
+h
r
)
2
, y = (h
t
−h
r
)
2
, x −y = 4h
t
h
r
x = m
2
λ
2
4
+y +mλ
_
y +d
2
⇒d =
¸
_
1
mλ
_
x −m
2
λ
2
4
−y
__
2
−y
d =
¸
_
4h
t
h
r
(2n + 1)λ
−
(2n + 1)λ
4
_
2
−(h
t
−h
r
)
2
, n ∈ Z
5. h
t
= 20m
h
r
= 3m
f
c
= 2GHz λ =
c
f
c
= 0.15
d
c
=
4h
t
h
r
λ
= 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 R
3
The length of the normal to the building from the top of Tx antenna = h
t
The length of the normal to the building from the top of Rx antenna = h
r
In this situation the 2 ray model is same as that analyzed in the book.
7. h(t) = α
1
δ(t −τ) +α
2
δ(t −(τ + 0.22µs))
G
r
= G
l
= 1
h
t
= h
r
= 8m
f
c
= 900MHz, λ = c/f
c
= 1/3
R = −1
delay spread =
x +x
−l
c
= 0.022 ×10
−6
s
⇒
2
_
8
2
+
_
d
2
_
2
−d
c
= 0.022 ×10
−6
s
⇒d = 16.1m
∴ τ =
d
c
= 53.67ns
α
1
=
_
λ
4π
√
G
l
l
_
2
= 2.71 ×10
−6
α
2
=
_
λ
4π
√
RG
r
x +x
_
2
= 1.37 ×10
−6
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 G
r
(gain of reﬂected path) is decreased, the asymptotic behavior of P
r
tends toward d
−2
from
d
−4
, 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 G
r
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+(hthr)^2).^.5;
r=(d.^2+(ht+hr)^2).^.5;
phd=2*pi/lambda*(r1);
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
0 20 40 60
−150
−100
−50
0
50
1
0
*
l
o
g
1
0
(
P
r
)
10 * log10(d)
G
r
= 1
0 20 40 60
−150
−100
−50
0
50
1
0
*
l
o
g
1
0
(
P
r
)
10 * log10(d)
G
r
= .316
0 20 40 60
−150
−100
−50
0
50
1
0
*
l
o
g
1
0
(
P
r
)
10 * log10(d)
G
r
= .1
0 20 40 60
−150
−100
−50
0
50
1
0
*
l
o
g
1
0
(
P
r
)
10 * log10(d)
G
r
= .01
0 20 40 60
0
100
200
300
400
P
h
a
s
e
(
d
e
g
)
10 * log10(d)
Figure 1: Problem 8
9. As indicated in the text, the power fall oﬀ with distance for the 10ray model is d
−2
for relatively large
distances
10. The delay spread is dictated by the ray reaching last d =
_
(500/6)
2
+ 10
2
= 83.93m
Total distance = 6d = 503.59m
τ
0
= 503.59/c = 1.68µs
L.O.S ray d = 500m
τ
0
= 500/c = 1.67µs
∴ delay spread = 0.01µs
11. f
c
= 900MHz
λ = 1/3m
G = 1 radar cross section 20dBm
2
= 10 log
1
0σ ⇒σ = 100
d=1 , s = s
=
_
(0.5d)
2
+ (0.5d)
2
= d
√
0.5 =
√
0.5
Path loss due to scattering
P
r
P
t
=
_
λ
√
Gσ
(4π)
3/2
ss
_
2
= 0.0224 = −16.498dB
Path loss due to reﬂection (using 2 ray model)
P
r
P
t
=
_
R
√
G
s +s
_
2 _
λ
4π
_
2
= 3.52 ×10
−4
= −34.54dB
d = 10 P
scattering
= −56.5dB P
reflection
= −54.54dB
d = 100 P
scattering
= −96.5dB P
reflection
= −74.54dB
d = 1000 P
scattering
= −136.5dB P
reflection
= −94.54dB
Notice that scattered rays over long distances result in tremendous path loss
12.
P
r
= P
t
K
_
d
0
d
_
γ
→simpliﬁed
P
r
= P
t
_√
G
l
4π
_
2
_
λ
d
_
2
→free space
∴ when K =
_
√
G
l
4π
_
2
and d
0
= λ
The two models are equal.
13. P
noise
= −160dBm
f
c
= 1GHz, d
0
= 1m, K = (λ/4πd
0
)
2
= 5.7 ×10
−4
, λ = 0.3, γ = 4
We want SNR
recd
= 20dB = 100
∵ Noise power is 10
−19
P = P
t
K
_
d
0
d
_
γ
10
−17
= 10K
_
0.3
d
_
4
d ≤ 260.7m
14. d = distance between cells with reused freq
p = transmit power of all the mobiles
_
S
I
_
uplink
≥ 20dB
(a) Min. S/I will result when main user is at A and Interferers are at B
d
A
= distance between A and base station #1 =
√
2km d
B
= distance between B and base station
#1 =
√
2km
_
S
I
_
min
=
P
_
Gλ
4πd
A
_
2
2P
_
Gλ
4πd
B
_
2
=
d
2
B
2d
2
A
=
(d
min
−1)
2
4
= 100
⇒ d
min
− 1 = 20km ⇒ d
min
= 21km since integer number of cells should be accommodated in
distance d ⇒d
min
= 22km
(b)
P
γ
P
u
= k
_
d
0
d
_
γ
⇒
_
S
I
_
min
=
Pk
_
d
0
d
A
_
γ
2Pk
_
d
0
d
B
_
γ
=
1
2
_
d
B
d
A
_
γ
=
1
2
_
d
min
−1
√
2
_
γ
=
1
2
_
d
min
−1
√
2
_
3
= 100
⇒d
min
= 9.27 ⇒ with the same argument ⇒d
min
= 10km
(c)
_
S
I
_
min
=
k
_
d
0
d
A
_
γ
A
2k
_
d
0
d
B
_
γ
B
=
(d
min
−1)
4
0.04
= 100
⇒d
min
= 2.41km ⇒ with the same argument d
min
= 4km
15. f
c
= 900MHz, h
t
= 20m, h
r
= 5m, d = 100m
Large urban city PL
largecity
= 353.52dB
small urban city PL
smallcity
= 325.99dB
suburb PL
suburb
= 207.8769dB
rural area PL
ruralarea/countryside
= 70.9278dB
As seen , path loss is higher in the presence of multiple reﬂectors, diﬀractors and scatterers
1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3
−60
−50
−40
−30
−20
−10
0
Plot for Gr = 0 dB
y = −15x+8
y = −35x+56
Figure 2: Problem 16
16. Piecewise linear model for 2path model. See Fig 2
17. P
r
= P
t
−P
L
(d) −
3
i
FAF
i
−
2
j
PAF
j
FAF =(5,10,6), PAF =(3.4,3.4)
P
L
(d)K
_
d
0
d
_
γ
0
= 10
−8
= −8dB
−110 = P
t
−80 −5 −10 −6 −3.4 −3.4
⇒P
t
= −2.2dBm
18. (a)
P
r
P
t
dB = 10 log
10
K −10r log
1
0
d
d
0
using least squares we get
10 log
10
K = −29.42dB
γ = 4
(b) PL(2Km) = 10 log
10
K −10r log
10
d = −161.76dB
(c) Receiver power can be assumed to be Gaussian with variance σ
2
ψdB
X ∼ N(0, σ
2
ψdB
)
Prob(X < −10) = Prob
_
X
σ
ψdB
<
−10
σ
ψdB
_
= 6.512 ×10
−4
19. Assume free space path loss parameters
f
c
= 900MHz →λ = 1/3m
σ
ψdB
= 6
SNR
recd
= 15dB
P
t
= 1W
g = 3dB
P
noise
= −40dBm ⇒P
recvd
= −55dB
Suppose we choose a cell of radius d
µ(d) = P
recvd
(due to path loss alone)
= P
t
_√
G
l
λ
4πd
_
2
=
1.4 ×1−
−3
d
2
µ
dB
= 10 log
1
0(µ(d))
P(P
recd
(d) > −55) = 0.9
P
_
P
recd
(d) −µ
dB
σ
ψdB
>
−55 −µ
dB
6
_
= 0.9
⇒
−55 −µ
dB
6
= −1.282
⇒µ
dB
= −47.308
⇒µ(d) = 1.86 ×10
−5
⇒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((ij)/Xc)*y(j);
end
end
21. Outage Prob. = Prob. [received power
dB
Tp
dB
]
Tp = 10dB
(a)
outageprob. = 1 −Q
_
Tp −µ
ψ
σ
ψ
_
= 1 −Q
_
−5
8
_
= Q
_
5
8
_
= 26%
(b) σ
ψ
= 4dB, outage prob < 1% ⇒
Q
_
Tp −µ
ψ
σ
ψ
_
> 99% ⇒
Tp −µ
ψ
σ
ψ
< −2.33 ⇒
µ
ψ
≥ 19.32dB
(c)
σ
ψ
= 12dB,
Tp −µ
ψ
σ
ψ
< −6.99 ⇒µ
ψ
≥ 37.8dB
0 20 40 60 80 100 120 140 160 180 200
−60
−50
−40
−30
−20
−10
0
10
20
White Noise Process
Filtered Shadowing Processing
Figure 3: Problem 20
(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.
C =
2
R
2
R
_
r=0
rQ
_
a +b ln
r
R
_
dr
To perform integration by parts, we let du = rdr and v = Q
_
a +b ln
r
R
_
. Then u =
1
2
r
2
and
dv =
∂
∂r
Q
_
a +b ln
r
R
_
=
∂
∂x
Q(x)
x=a+b ln(r/R)
∂
∂r
_
a +b ln
r
R
_
=
−1
√
2π
exp(−k
2
/2)
b
r
dr. (1)
where k = a +b ln
r
R
. Then we get
C =
2
R
2
_
1
2
r
2
Q
_
a +b ln
r
R
_
_
R
r=0
+
2
R
2
R
_
r=0
1
2
r
2
1
√
2π
exp(−k
2
/2)
b
r
dr (2)
= Q(a) +
1
R
2
R
_
r=0
r
2
1
√
2π
exp(−k
2
/2)
b
r
dr (3)
= Q(a) +
1
R
2
R
_
r=0
1
√
2π
R
2
exp
_
2(k −a)
b
_
exp(−k
2
/2)
b
r
dr (4)
= Q(a) +
a
_
k=−∞
1
√
2π
exp
_
−k
2
2
+
2k
b
−
2a
b
_
dk (5)
= Q(a) + exp
_
−2a
b
+
2
b
2
_
a
_
k=−∞
1
√
2π
exp
_
−
1
2
(k −
2
b
)
2
_
dk (6)
= Q(a) + exp
_
2 −2ab
b
2
_
_
1 −Q
_
a −
2
b
1
__
(7)
= Q(a) + exp
_
2 −2ab
b
2
_
Q
_
2 −ab
b
_
(8)
(9)
Since Q(−x) = 1 −Q(x).
23. γ = 3
d
0
= 1
k = 0dB
σ = 4dB
R = 100m
P
t
= 80mWP
min
= −100dBm = −130dB
P
γ
(R) = P
t
K
_
d
0
d
_
γ
= 80 ×10
−9
= −70.97dB
a =
P
min
−P
γ
(R)
σ
= 14.7575
b =
10γ log
10
e
σ
ψdB
= 3.2572
c = Q(a) +e
2−2ab
b
2
Q
_
2 −2ab
b
_
1
24. γ = 6
σ = 8
P
γ
(R) = 20 +P
min
a = −20/8 = −2.5
b =
10 ×6 ×log
10
e
8
= 20.3871
c = 0.9938
25.
γ/σ
ψ
dB
2 4 6
4 0.7728 0.8587 0.8977
8 0.6786 0.7728 0.8255
12 0.6302 0.7170 0.7728
Since P
r
(r) ≥ P
min
for all r ≤ R, the probability of nonoutage is proportional to Q
_
−1
σ
_
, and thus
decreases as a function of σ. Therefore, C decreases as a function of σ. Since the average power at the
boundary of the cell is ﬁxed, C increases with γ, because it forces higher transmit power, hence more
received power at r < R. Due to these forces, we have minimum coverage when γ = 2 and σ = 12. By
a similar argument, we have maximum coverage when γ = 6 and σ = 4. The same can also be seen
from this ﬁgure:
2
3
4
5
6
4
6
8
10
12
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
γ
σ
ψ
dB
C
o
v
e
r
a
g
e
Figure 4: Problem 25
The value of coverage for middle point of typical values i.e. γ = 4 and σ = 8 can be seen from the
table or the ﬁgure to be 0.7728.
Chapter 3
1. d = vt
r +r
= d +
2h
2
d
Equivalent lowpass channel impulse response is given by
c(τ, t) = α
0
(t)e
−jφ
0
(t)
δ(τ −τ
0
(t)) +α
1
(t)e
−jφ
1
(t)
δ(τ −τ
1
(t))
α
0
(t) =
λ
√
G
l
4πd
with d = vt
φ
0
(t) = 2πf
c
τ
0
(t) −φ
D
0
τ
0
(t) = d/c
φ
D
0
=
_
t
2πf
D
0
(t)dt
f
D
0
(t) =
v
λ
cos θ
0
(t)
θ
0
(t) = 0 ∀t
α
1
(t) =
λR
√
G
l
4π(r+r
)
=
λR
√
G
l
4π(d+
2h
2
d
)
with d = vt
φ
1
(t) = 2πf
c
τ
1
(t) −φ
D
1
τ
1
(t) = (r +r
)/c = (d +
2h
2
d
)/c
φ
D
1
=
_
t
2πf
D
1
(t)dt
f
D
1
(t) =
v
λ
cos θ
1
(t)
θ
1
(t) = π −arctan
h
d/2
∀t
2. For the 2 ray model:
τ
0
=
l
c
τ
1
=
x +x
c
∴ delay spread(T
m
) =
x +x
−l
c
=
_
(h
t
+h
r
)
2
+d
2
−
_
(h
t
−h
r
)
2
+d
2
c
when d (h
t
+h
r
)
T
m
=
1
c
2h
t
h
r
d
h
t
= 10m, h
r
= 4m, d = 100m
∴ T
m
= 2.67 ×10
−9
s
3. Delay for LOS component = τ
0
= 23 ns
Delay for First Multipath component = τ
1
= 48 ns
Delay for Second Multipath component = τ
2
= 67 ns
τ
c
= Delay for the multipath component to which the demodulator synchronizes.
T
m
= max
m
τ
m
−τ
c
So, when τ
c
= τ
0
, T
m
= 44 ns. When τ
c
= τ
1
, T
m
= 19 ns.
4. f
c
= 10
9
Hz
τ
n,min
=
10
3×10
8
s
∴ min f
c
τ
n
=
10
10
3×10
8
= 33 1
5. Use CDF strategy.
F
z
(z)= P[x
2
+y
2
≤ z
2
] =
_ _
x
2
+y
2
≤z
2
1
2πσ
2
e
−(x
2
+y)
2
2σ
2
dxdy =
2π
_
0
z
_
0
1
2πσ
2
e
−r
2
2σ
2
rdrdθ = 1 −e
−z
2
2σ
2
(z ≥ 0)
df
z
(z)
dz
=
z
σ
2
e
−
z
2
2σ
2
→Rayleigh
For Power:
F
z
2(z)=P[Z ≤
√
z] = 1 −e
−z
2
2σ
2
f
z
(z)=
1
2σ
2
e
−z
2σ
2
→Exponential
6. For Rayleigh fading channel
Pr(P
r
< P
0
) = 1 −e
−P
0
/2σ
2
2σ
2
= −80dBm, P
0
= −95dBm, Pr(P
r
< P
0
) = 0.0311
2σ
2
= −80dBm, P
0
= −90dBm, Pr(P
r
< P
0
) = 0.0952
7. For Rayleigh fading channel
P
outage
= 1 −e
−P
0
/2σ
2
0.01 = 1 −e
−P
0
/P
r
∴ P
r
= −60dBm
8. 2σ
2
= 80dBm = 10
−11
Target Power P
0
= 80 dBm = 10
−11
Avg. power in LOS component = s
2
= 80dBm = 10
−11
Pr[z
2
≤ 10
−11
] = Pr[z ≤
10
−5
√
10
]
Let z
0
=
10
−5
√
10
=
_
z
0
0
z
σ
2
e
−
−(z
2
+s
2
)
2σ
2
I
0
_
zs
σ
2
_
dz, z ≥ 0
= 0.3457
To evaluate this, we use Matlab and I
0
(x) = besseli(0,x). Sample Code is given:
clear P0 = 1e11; s2 = 1e11; sigma2 = (1e11)/2; z0 =
sqrt(1e11); 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
F
Ricean
Z
(z) =
_
z
0
p
Ricean
Z
(z)
where
p
Ricean
Z
(z) =
2z(K + 1)
Pr
exp
_
−K −
(K + 1)z
2
Pr
_
I
0
_
2z
_
K(K + 1)
Pr
_
, z ≥ 0
For the Nakagamim approximation to Ricean distribution, we set the Nakagami m parameter to be
(K + 1)
2
/(2K + 1). CDF of Nakagamim distribution is
F
Nakagamim
Z
(z) =
_
z
0
p
Nakagamim
Z
(z)
where
p
Nakagamim
Z
(z) =
2m
m
z
2m−1
Γ(m)Pr
m
exp
_
−mz
2
Pr
_
, z ≥ 0, m ≥ 0.5
We need to plot the two CDF curves for K = 1,5,10 and Pr =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*m1))/(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 Nakagamim approximation becomes better as K increases. Also, for a
ﬁxed value of K and x, prob(γ<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 Nakagamidistribution is
always above the Ricean distribution.
10. (a) W = average received power
Z
i
= Shadowing over link i
P
ri
= Received power in dBW, which is Gaussian with mean W, variance σ
2
(b)
P
outage
= P [P
r,1
< T ∩ P
r,2
< T] = P [P
r,1
< T] P[P
r,2
< T] since Z
1
, Z
2
independent
=
_
Q
_
W −T
σ
__
2
=
_
Q
_
σ
__
2
(c)
P
out =
∞
_
−∞
P [P
r,1
≤ T , P
r,2
< TY = y] f
y
(y) dy
P
r,1
Y = y ~ N
_
W +by,a
2
σ
2
_
, and [P
r,1
Y = y] ⊥ [P
r,2
Y = y]
P
outage=
∞
_
−∞
_
Q
_
W +by −T
aσ
__
2
1
√
2πσ
e
−
y
2
2σ
2
dy
0 0.5 1 1.5 2 2.5 3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Ricean
Nakagami −m
0 0.5 1 1.5 2 2.5 3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
K = 5
Ricean
Nakagami−m
0 0.5 1 1.5 2 2.5 3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
K = 10
Ricean
Nakagami−m
0 0.5 1 1.5 2 2.5 3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
K = 1
Ricean
Nakagami − m
0 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
K = 5
Ricean
Nakagami−m
0 0.5 1 1.5 2 2.5 3
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
K = 10
Ricean
Nakagami−m
3.5 4 4.5 5 5.5 6
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
x 10
−31
Tail Behavior
K = 10
Ricean
Nakagami−m
Figure 1: The CDF and PDF for K = 1, 5, 10 and the Tail Behavior
let
y
σ
= u
=
∞
_
−∞
1
√
2π
_
Q
_
W −T +buσ
aσ
__
2
e
−u
2
2
du =
∞
_
−∞
1
√
2π
_
Q
_
+byσ
aσ
__
2
e
−y
2
2
dy
(d) Let a = b =
1
√
2
, σ = 8, = 5. With independent fading we get
P
out
=
_
Q
_
5
8
__
2
= 0.0708.
With correlated fading we get P
out
= 0.1316.
Conclusion : Independent shadowing is prefarable for diversity.
11. There are many acceptable techniques for this problem. Sample code for both the stochastic tech
nique(preferred) and the Jake’s technique are included.
Jakes: Summing of appropriate sine waves
%Jake’s Method
close all; clear all;
%choose N=30
N=30; M=0.5*(N/21); 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
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
−10
−5
0
5
fd = 1 Hz
E
n
v
e
l
o
p
e
(
d
B
)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
−20
−10
0
10
fd = 10 Hz
E
n
v
e
l
o
p
e
(
d
B
)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
−20
−10
0
10
fd = 100 Hz
E
n
v
e
l
o
p
e
(
d
B
)
Figure 2: Problem 11
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; % [Hz] Minumum required sampling rate
end;
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; % Use even number of samples in calculation
end;
f = linspace(f_s,f_s,N); % [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));
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
−30
−20
−10
0
1 Hz
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
−30
−20
−10
0
10
10Hz
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
−60
−40
−20
0
20
l00 Hz
Figure 3: Problem 11
12. P
r
= 30dBm
f
D
= 10Hz
P
0
= 0dBm, t
z
=
e
ρ
2
−1
ρf
D
√
2π
= 0.0013s
P
0
= 15dBm, t
z
= 0.0072s
P
0
= 25dBm, t
z
= 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
L
Z
(above) = L
Z
(below) =
√
2πf
D
ρe
−ρ
2
And,
t
Z
(above) =
p(z > Z)
L
Z
(above)
p(z > Z) = 1 −p(z ≤ Z) = 1 −(1 −e
−ρ
2
) = e
−ρ
2
t
Z
(above) =
1
√
2πf
D
ρ
The values of t
Z
(above) for f
D
= 10,50,80 Hz are 0.0224s, 0.0045s, 0.0028s respectively. Notice that as
f
D
increases, t
Z
(above) reduces.
14. Note: The problem has been solved for T
s
= 10µs
P
r
= 10dB
f
D
= 80Hz
R
1
: −∞≤ γ ≤ −10dB, π
1
= 9.95 ×10
−3
R
2
: −10dB ≤ γ ≤ 0dB, π
2
= 0.085
R
3
: 0dB ≤ γ ≤ 5dB, π
3
= 0.176
R
4
: 5dB ≤ γ ≤ 10dB, π
4
= 0.361
R
5
: 10dB ≤ γ ≤ 15dB, π
5
= 0.325
R
6
: 15dB ≤ γ ≤ 20dB, π
6
= 0.042
R
7
: 20dB ≤ γ ≤ 30dB, π
7
= 4.54 ×10
−5
R
8
: 30dB ≤ γ ≤ ∞, π
8
= 3.72 ×10
−44
N
j
→ level crossing rate at level A
j
N
1
= 0, ρ =
_
0
10
N
2
= 19.85, ρ =
_
0.1
10
N
3
= 57.38, ρ =
_
1
10
N
4
= 82.19, ρ =
_
10
0.5
10
N
5
= 73.77, ρ =
_
10
10
N
6
= 15.09, ρ =
_
10
1.5
10
N
7
= 0.03, ρ =
_
10
2
10
N
8
= 0, ρ =
_
10
3
10
MATLAB CODE:
N = [0 19.85 57.38 82.19 73.77 15.09 .03 0];
Pi = [9.95e3 .085 .176 .361 .325 .042 4.54e5 3.72e44];
T = 10e3;
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 0.0199 0.9801
% 0.0023 0.0068 0.9909
% 0.0033 0.0047 0.9921
% 0.0023 0.0020 0.9957
% 0.0023 0.0005 0.9973
% 0.0036 0.0000 0.9964
% 0.0066 0 0.9934
% 0 0 1.0000
15. (a)
S(τ, ρ) =
_
_
_
α
1
δ(τ) ρ = 70Hz
α
2
δ(τ −0.022µsec) ρ = 49.5Hz
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
2
¸
_
d
2
_
2
+ 64
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
2
¸
_
d
2
_
2
+ 8
2
−d = 0.022 ×10
−6
3 ×10
8
⇒4
_
d
2
4
+ 8
2
_
= 6.6
2
+d
2
+ 2d(6.6)
⇒d = 16.1m
f
D
= v cos(θ)/λ. v = f
D
λ/ cos(θ). For the LOS ray, θ = 0 and for the multipath component,
θ = 45
o
. We can use either of these rays and the corresponding f
D
value to get v = 23.33m/s.
(b)
d
c
=
4h
t
h
r
λ
d
c
= 768 m. Since d d
c
, power falloﬀ is proportional to d
−2
.
(c) T
m
= 0.022µs, B
−1
= 0.33µs. Since T
m
B
−1
, we have ﬂat fading.
16. (a) Outdoor, since delay spread ≈ 10 µsec.
Consider that 10 µsec ⇒ d = ct = 3km diﬀerence between length of ﬁrst and last path
(b) Scattering function
S(τ, ρ) = F
∆t
[A
c
(τ, ∆t)]
=
1
W
rect
_
1
W
ρ
_
for 0 ≤ τ ≤ 10µsec
The Scattering function is plotted in Fig. 16
8m 8m
d meters
Figure 4: Antenna Setting
0.022 us
0
t
t0 = (d/3e8)
t1 = 2 sqrt[(d/2)^2+8^2]/3e8
t1 t0
Figure 5: Time Axis for Ray Arrival
(c) Avg Delay Spread =
∞
0
τA
c
(τ)dτ
∞
0
A
c
(τ)dτ
= 5µsec
RMS Delay Spread =
¸
¸
¸
¸
_
∞
0
(τ−µ
Tm
)
2
A
c
(τ)dτ
∞
0
A
c
(τ)dτ
= 2.89µsec
Doppler Spread =
W
2
= 50 Hz
(d) β
u
> Coherence BW ⇒ Freq. Selective Fading ≈
1
T
m
= 10
5
⇒ β
u
> = 10
5
kHz
Can also use µ
Tm
or σ
Tm
instead of T
m
(e) Rayleigh fading, since receiver power is evenly distributed relative to delay; no dominant LOS
path
(f) t
R
=
e
ρ
2
−1
ρF
D
√
2π
with ρ = 1, f
D
=
W
2
→ t
r
= .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 errorfree
transmission. So
1
t
R
= 72.94 symbols/sec
17. (a) T
m
≈ .1msec = 100µsec
B
d
≈ .1Hz
Answers based on µ
T
m
or σ
T
m
are ﬁne too. Notice, that based on the choice of either T
m
, µ
T
m
or
σ
T
m
, the remaining answers will be diﬀerent too.
(b) B
c
≈
1
T
m
= 10kHz
∆f > 10kHz for u
1
⊥ u
2
(c) (∆t)
c
= 10s
(d) 3kHz < B
c
⇒ Flat
30kHz > B
c
⇒ Freq. Selective
rho
tau
W/2 W/2
10 us
S(tau,rho)
Figure 6: Scattering Function
Chapter 4
1. C = Blog
2
_
1 +
S
N
0
B
_
C =
log
2
1+
S
N
0
B
1
B
As B →∞ by L’Hospital’s rule
C =
S
N
0
1
ln 2
2. B = 50 MHz
P = 10 mW
N
0
= 2 ×10
−9
W/Hz
N = N
0
B
C = 6.87 Mbps.
P
new
= 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. P
noise
= 0.1mW
B = 20MHz
(a) C
user1→base station
= 0.933B = 18.66Mbps
(b) C
user2→base station
= 3.46B = 69.2Mbps
4. (a) Ergodic Capacity (with Rcvr CSI only)= B[
6
i=1
log
2
(1 +γ
i
)p(γ
i
)] = 2.8831×B = 57.66 Mbps.
(b) p
out
= Pr(γ < γ
min
)
C
o
= (1p
out
)Blog
2
(1 +γ
min
)
For
γ
min
> 20dB, p
out
= 1, C
o
= 0
15dB < γ
min
< 20dB, p
out
= .9, C
o
= 0.1Blog
2
(1 +γ
min
), max C
o
at γ
min
≈ 20dB.
10dB < γ
min
< 15dB, p
out
= .75, C
o
= 0.25Blog
2
(1 +γ
min
), max C
o
at γ
min
≈ 15dB.
5dB < γ
min
< 10dB, p
out
= .5, C
o
= 0.5Blog
2
(1 +γ
min
), max C
o
at γ
min
≈ 10dB.
0dB < γ
min
< 5dB, p
out
= .35, C
o
= 0.65Blog
2
(1 +γ
min
), max C
o
at γ
min
≈ 5dB.
−5dB < γ
min
< 0dB, p
out
= .1, C
o
= 0.9Blog
2
(1 +γ
min
), max C
o
at γ
min
≈ 0dB.
γ
min
< −5dB, p
out
= 0, C
o
= Blog
2
(1 +γ
min
), max C
o
at γ
min
≈ 5dB.
Plot is shown in Fig. 1. Maximum at γ
min
= 10dB, p
out
=0.5 and C
o
= 34.59 Mbps.
5. (a) We suppose that all channel states are used
1
γ
0
= 1 +
4
i=1
1
γ
i
p
i
⇒ γ
0
= 0.8109
1
γ
0
−
1
γ
4
> 0 ∴ true
S(γ
i
)
S
=
1
γ
0
−
1
γ
i
0 0.2 0.4 0.6 0.8 1
0
0.5
1
1.5
2
2.5
3
3.5
x 10
7
P
out
C
a
p
a
c
i
t
y
(
b
p
s
)
Figure 1: Capacity vs P
out
S(γ)
S
=
_
¸
¸
_
¸
¸
_
1.2322 γ = γ
1
1.2232 γ = γ
2
1.1332 γ = γ
3
0.2332 γ = γ
4
C
B
=
4
i=1
log
2
_
γ
i
γ
0
_
p(γ
i
) = 5.2853bps/Hz
(b) σ =
1
E[1/γ]
= 4.2882
S(γ
i
)
S
=
σ
γ
i
S(γ)
S
=
_
¸
¸
_
¸
¸
_
0.0043 γ = γ
1
0.0029 γ = γ
2
0.4288 γ = γ
3
4.2882 γ = γ
4
C
B
= log
2
(1 +σ) = 2.4028bps/Hz
(c) To have p
out
= 0.1 or 0.01 we will have to use all the subchannels as leaving any of these will
result in a p
out
of at least 0.2 ∴ truncated channel power control policy and associated spectral
eﬃciency are the same as the zerooutage case in part b .
To have p
out
that maximizes C with truncated channel inversion, we get
max
C
B
= 4.1462bps/Hz p
out
= 0.5
6. (a)
SNR
recvd
=
P
γ
(d)
P
noise
=
_
¸
¸
_
¸
¸
_
10dB w.p. 0.4
5dB w.p. 0.3
0dB w.p. 0.2
−10dB w.p. 0.1
Assume all channel states are used
1
γ
0
= 1 +
4
i=1
1
γ
i
p
i
⇒ γ
0
= 0.4283 > 0.1 ∴ not possible
Now assume only the best 3 channel states are used
0.9
γ
0
= 1 +
3
i=1
1
γ
i
p
i
⇒ γ
0
= 0.6742 < 1 ∴ ok!
S(γ)
S
=
_
¸
¸
_
¸
¸
_
1.3832 γ = γ
1
= 10
1.1670 γ = γ
2
= 3.1623
0.4832 γ = γ
3
= 1
0 γ = γ
4
= 0.1
C/B = 2.3389bps/Hz
(b) σ = 0.7491
C/B = log
2
(1 +σ) = 0.8066bps/Hz
(c)
_
C
B
_
max
= 2.1510bps/Hz obtained by using the best 2 channel states.
With p
out
= 0.1 + 0.2 = 0.3
7. (a) Maximize capacity given by
C = max
S(γ):
S(γ)p(γ)dγ=S
_
γ
Blog
_
1 +
S(γ)γ
S
_
p(γ)dγ.
Construct the Lagrangian function
L =
_
γ
Blog
_
1 +
S(γ)γ
S
_
p(γ)dγ −λ
_
S(γ)
S
p(γ)dγ
Taking derivative with respect to S(γ), (refer to discussion section notes) and setting it to zero,
we obtain,
S(γ)
S
=
_
1
γ
0
−
1
γ
γ ≥ γ
0
0 γ < γ
0
Now, the threshold value must satisfy
_
∞
γ
0
_
1
γ
0
−
1
γ
_
p(γ)dγ = 1
Evaluating this with p(γ) =
1
10
e
−γ/10
, we have
1 =
1
10γ
0
_
∞
γ
0
e
−γ/10
dγ −
1
10
_
∞
γ
0
e
−γ/10
γ
dγ (1)
=
1
γ
0
e
−γ
0
/10
−
1
10
_
∞
γ
0
10
e
−γ
γ
dγ (2)
=
1
γ
0
e
−γ
0
/10
−
1
10
EXPINT(γ
0
/10) (3)
where EXPINT is as deﬁned in matlab. This gives γ
0
= 0.7676. The power adaptation becomes
S(γ)
S
=
_
1
0.7676
−
1
γ
γ ≥ 0.7676
0 γ < 0.7676
(b) Capacity can be computed as
C/B =
1
10
_
∞
0.7676
log (γ/0.7676) e
−γ/10
dγ = 2.0649 nats/sec/Hz.
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 γ
C/B =
1
10
_
∞
0
log (1 +γ) e
−γ/10
dγ = 2.0150 nats/sec/Hz.
(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
1
10
_
∞
γ0
e
−γ/10
dγ = 0.95
which gives γ
0
= 0.5129. This gives us the capacity with truncated channel inversion as
C/B = log
_
1 +
1
1
10
_
∞
γ
0
1
γ
e
−γ/10
dγ
_
∗ 0.95 (4)
= log
_
1 +
1
1
10
EXPINT(γ
0
/10)
_
∗ 0.95 (5)
= 1.5463 nats/s/Hz. (6)
(f) Channel Mean=5 dB = 0.3162. So for perfect channel knowledge at transmitter and receiver we
compute γ
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 Jensen’s 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,
C = Blog
_
1 +
S
N
0
B +I
_
= 10.7Kbps.
(b) Suppose we transmit at power S
1
when jammer is oﬀ and S
2
when jammer is oﬀ,
C = Bmax
_
log
_
1 +
S
1
N
o
B
_
0.75 + log
_
1 +
S
2
N
o
B +I
_
0.25
_
subject to
0.75S
1
+ 0.25S
2
= S.
This gives S
1
= 12.25mW, S
2
= 3.25mW and C = 53.21Kbps.
(c) The jammer should transmit −x(t) to completely cancel oﬀ the signal.
S = 10mW
N
0
= .001 µW/Hz
B = 10 MHz
Now we compute the SNR’s as:
γ
j
=
H
j

2
S
N
0
B
This gives: γ
1
=
1
2
10
−3
0.001×10
−6
10×10
6
= 1, γ
2
= .25, γ
3
= 4, γ
4
= 0.0625
To compute γ
0
we use the power constraint as:
j
_
1
γ
0
−
1
γ
j
_
+
= 1
First assume that γ
0
< 0.0625, then we have
4
γ
0
= 1 +
_
1
1
+
1
.25
+
1
4
+
1
.0625
_
⇒γ
0
= .1798 > 0.0625
So, our assumption was wrong. Now we assume that 0.0625 < γ
0
< .25, then
3
γ
0
= 1 +
_
1
1
+
1
.25
+
1
4
_
⇒γ
0
= .48 > 0.25
So, our assumption was wrong again. Next we assume that 0.25 < γ
0
< 1, then
2
γ
0
= 1 +
_
1
1
+
1
4
_
⇒γ
0
= .8889 < 1
This time our assumption was right. So we get that only two subbands each of bandwidth 10 MHz
are used for transmission and the remaining two with lesser SNR’s are left unused.
Now, we can ﬁnd capacity as:
C =
j:γ
j
≥γ
0
Blog
2
_
γ
j
γ
0
_
This gives us, C = 23.4 Mbps.
9. Suppose transmit power is P
t
, interference power is P
int
and noise power is P
noise
.
In the ﬁrst strategy C/B = log
2
_
1 +
P
t
P
int
+P
noise
_
In the second strategy C/B = log
2
_
1 +
P
t
−P
int
P
noise
_
Assuming that the transmitter transmits x[k] added to its message always, power remaining for actual
messages is P
t
−P
int
The ﬁrst or second strategy may be better depending on
P
t
P
int
+P
noise
≷
P
t
−P
int
P
noise
⇒P
t
−P
int
−P
noise
≷ 0
P
t
is generally greater than P
int
+P
noise
, and so strategy 2 is usually better.
1
0.5
2
0.25
f (in MHz)
fc
fc+10
fc+20
fc10
fc20
H(f)
Figure 2: Problem 11
10. We show this for the case of a discrete fading distribution
C = Σlog
_
1 +
(1 +j)
2
P
j
N
0
B
_
L =
i
log
_
1 +
(1 +j)
2
P
j
N
0
B
_
−d
j
_
_
j
P
j
−P
_
_
∂L
∂P
j
= 0
⇒
(1 +j)
2
P
j
N
0
B
=
1
λ
(1 +j)
2
N
0
B
−1
letγ
j
=
(1 +j)
2
P
N
0
B
⇒
P
j
P
=
1
λP
−
1
γ
j
denote
1
γ
0
=
1
λP
∴
P
j
P
=
1
γ
0
−
1
γ
j
subject to the constraint
ΣP
j
P
= 1
11. S = 10mW
N
0
= .001 µW/Hz
B = 10 MHz
Now we compute the SNR’s as:
γ
j
=
H
j

2
S
N
0
B
This gives: γ
1
=
1
2
10
−3
0.001×10
−6
10×10
6
= 1, γ
2
= .25, γ
3
= 4, γ
4
= 0.0625
To compute γ
0
we use the power constraint as:
j
_
1
γ
0
−
1
γ
j
_
+
= 1
First assume that γ
0
< 0.0625, then we have
4
γ
0
= 1 +
_
1
1
+
1
.25
+
1
4
+
1
.0625
_
⇒γ
0
= .1798 > 0.0625
So, our assumption was wrong. Now we assume that 0.0625 < γ
0
< .25, then
3
γ
0
= 1 +
_
1
1
+
1
.25
+
1
4
_
⇒γ
0
= .48 > 0.25
So, our assumption was wrong again. Next we assume that 0.25 < γ
0
< 1, then
2
γ
0
= 1 +
_
1
1
+
1
4
_
⇒γ
0
= .8889 < 1
This time our assumption was right. So we get that only two subbands each of bandwidth 10 MHz
are used for transmission and the remaining two with lesser SNR’s are left unused.
Now, we can ﬁnd capacity as:
C =
j:γ
j
≥γ
0
Blog
2
_
γ
j
γ
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
max
s.t.
i
P(f
i
)≤P
i
log
2
_
1 +
H(f
i
)
2
P(f
i
)
N
0
_
denote γ(f
i
) =
H(f
i
)
2
P(f
i
)
N
0
L =
i
log
2
_
1 +γ(f
i
)
P(f
i
)
P
_
+λ(
P(f
i
))
denote x
i
=
P(f
i
)
P
, the problem is similar to problem 10
⇒x
i
=
1
γ
0
−
1
γ(f
i
)
where γ
0
is found from the constraints
i
_
1
γ
0
−
1
γ(f
i
)
_
= 1 and
1
γ
0
−
1
γ(f
i
)
≥ 0∀i
13. (a) C=13.98Mbps
MATLAB
Gammabar = [1 .5 .125];
ss = .001;
P = 30e3;
N0 = .001e6;
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;
for i = 1:length(Gammabar)
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((sumP1)));
gamma0ch = gamma0v(c);
C = 0;
for i = 1:length(Gammabar)
a = gamma.*(gamma>gamma0ch);
[b,c] = max(a>0);
gammac = a(find(a));
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 = 30e3;
N0 = .001e6;
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);
gammac = a(find(a));
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((sumP1)));
gamma0ch = gamma0v(c);
a = gamma.*(gamma>gamma0ch);
[b,c] = max(a>0);
gammac = a(find(a));
pgammac = pgamma(c:length(gamma));
C(k) = Bc*ss*sum(log2(gammac/gamma0ch).*pgammac);
end Ctot = sum(C);
Chapter 6
1. (a) For sinc pulse, B =
1
2T
s
⇒T
s
=
1
2B
= 5 ×10
−5
s
(b) SNR =
P
b
N
0
B
= 10
Since 4QAM is multilevel signalling
SNR =
P
b
N
0
B
=
E
s
N
0
BT
s
=
2E
s
N
0
B
_
∵ BT
s
=
1
2
_
∴ SNR per symbol =
E
s
N
0
= 5
SNR per bit =
E
b
N
0
= 2.5 (a symbol has 2 bits in 4QAM)
(c) SNR per symbol remains the same as before =
E
s
N
0
= 5
SNR per bit is halved as now there are 4 bits in a symbol
E
b
N
0
= 1.25
2. p
0
= 0.3, p
1
= 0.7
(a)
P
e
= Pr(0 detected, 1 sent — 1 sent)p(1 sent) +Pr(1 detected, 0 sent — 0 sent)p(0 sent)
= 0.7Q
_
d
min
√
2N
0
_
+ 0.3Q
_
d
min
√
2N
0
_
= Q
_
d
min
√
2N
0
_
d
min
= 2A
= Q
_
_
¸
2A
2
N
0
_
_
(b)
p( ˆ m = 0m = 1)p(m = 1) = p( ˆ m = 1m = 0)p(m = 0)
0.7Q
_
_
A+a
_
N
0
2
_
_
= 0.3Q
_
_
A−a
_
N
0
2
_
_
, a > 0
Solving gives us ’a’ for a given A and N
0
(c)
p( ˆ m = 0m = 1)p(m = 1) = p( ˆ m = 1m = 0)p(m = 0)
0.7Q
_
_
A
_
N
0
2
_
_
= 0.3Q
_
_
B
_
N
0
2
_
_
, a > 0
Clearly A > B, for a given A we can ﬁnd B
(d) Take
E
b
N
0
=
A
2
N
0
= 10
In part a) P
e
= 3.87 ×10
−6
In part b) a=0.0203 P
e
= 3.53 ×10
−6
In part c) B=0.9587 P
e
= 5.42 ×10
−6
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));
t2=.3*Q(a/sqrt(N0/2));
diff = abs(t1t2);
[c,d] = min(diff);
a(d)
c
3. s(t) = ±g(t) cos 2πf
c
t
r = ˆ r cos ∆φ
where ˆ r is the signal after the sampler if there was no phase oﬀset. Once again, the threshold that
minimizes P
e
is 0 as (cos ∆φ) acts as a scaling factor for both +1 and 1 levels. P
e
however increases
as numerator is reduced due to multiplication by cos ∆φ
P
e
= Q
_
d
min
cos ∆φ
√
2N
0
_
4.
A
2
c
_
T
b
0
cos
2
2πf
c
tdt = A
2
c
_
T
b
0
1 + cos 4πf
c
t
2
= A
2
c
_
¸
¸
¸
_
T
b
2
+
sin(4πf
c
T
b
)
8πf
c
. ¸¸ .
→0 as f
c
1
_
¸
¸
¸
_
=
A
2
c
T
b
2
= 1
x(t) = 1 +n(t)
Let prob 1 sent =p
1
and prob 0 sent =p
0
P
e
=
1
6
[1.p
1
+ 0.p
0
] +
2
6
[0.p
1
+ 0.p
0
] +
2
6
[0.p
1
+ 0.p
0
] +
1
6
[0.p
1
+ 1.p
0
]
=
1
6
[p
1
+p
0
] =
1
6
(∵ p
1
+p
0
= 1 always )
5. We will use the approximation P
e
∼ (average number of nearest neighbors).Q
_
d
min
√
2N
0
_
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 of neighbors = 4.5
P
e
= 4.5Q
_
2a
√
2N
0
_
(b) 16QAM, P
e
= 4
_
1 −
1
4
_
Q
_
2a
√
2N
0
_
= 3Q
_
2a
√
2N
0
_
(c) P
e
∼
2×3+3×2
5
Q
_
2a
√
2N
0
_
= 2.4Q
_
2a
√
2N
0
_
(d) P
e
∼
1×4+4×3+4×2
9
Q
_
3a
√
2N
0
_
= 2.67Q
_
3a
√
2N
0
_
6.
P
s,exact
= 1 −
_
1 −
2(
√
M −1)
√
M
Q
_
_
3γ
s
M −1
__
2
Figure 1: Problem 5
P
s,approx
=
4(
√
M −1)
√
M
Q
_
_
3γ
s
M −1
_
approximation is better for high SNRs as then the multiplication factor is not important and P
e
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=1exp(2*log((1((2*(sqrt(M)1))/(sqrt(M)))*Q(sqrt((3*gamma)/(M1))))));
Ps_approx = ((4*(sqrt(M)1))/sqrt(M))*Q(sqrt((3*gamma)/(M1)));
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 overestimates the error rate because it overcounts the probability that the transmitted signal
is mistaken for something other than its nearest neighbors. At high SNR, this is very unlikely and this
overcounting becomes negligible.
8. (a)
I
x
(a) =
_
∞
0
e
−at
2
x
2
+t
2
dt
since the integral converges we can interchange integral and derivative for a¿0
∂I
x
(a)
∂a
=
_
∞
0
−te
−at
2
x
2
+t
2
dt
x
2
I
x
(a) −
∂I
x
(a)
∂a
=
_
∞
0
(x
2
+t
2
)e
−at
2
x
2
+t
2
dt =
_
∞
0
e
−at
2
dt =
1
2
_
π
a
0 5 10 15 20 25 30
10
−300
10
−200
10
−100
10
0
P
s
Problem 2 − Symbol Error Probability for QPSK
Approximation
Exact Formula
0 1 2 3 4 5 6 7 8 9 10
10
−3
10
−2
10
−1
10
0
SNR(dB)
P
s
Problem 2 − Symbol Error Probability for QPSK
Approximation
Exact Formula
Figure 2: Problem 7
(b) Let I
x
(a) = y, we get
y
−x
2
y = −
1
2
_
π
a
comparing with
y
+P(a)y = Q(a)
P(a) = −x
2
, Q(a) = −
1
2
_
π
a
I.F. = e
P(u)u
= e
−x
2
a
∴ e
−x
2
a
y =
_
−
1
2
_
π
a
e
−x
2
u
du
solving we get
y =
π
2x
e
ax
2
erfc(x
√
a)
(c)
erfc(x
√
a) = I
x
(a)
2x
π
e
−ax
2
=
2x
π
e
−ax
2
_
∞
0
e
−at
2
x
2
+t
2
dt
a = 1
erfc(x) =
2x
π
e
−ax
2
_
∞
0
e
−at
2
x
2
+t
2
dt
=
2
π
_
π/2
0
e
−x
2
/sin
2
θ
dθ
Q(x) =
1
2
erfc(x/
√
2) =
1
π
_
π/2
0
e
−x
2
/2sin
2
θ
dθ
9. P = 100W
N
0
= 4W, SNR = 25
P
e
= Q(
√
2γ) = Q(
√
50) = 7.687 ×10
−13
data requires P
e
∼ 10
−6
voice requires P
e
∼ 10
−3
so it can be used for both.
with fading P
e
=
1
4γ
b
= 0.01
So the system can’t be used for data at all. It can be used for very low quality voice.
10. T
s
= 15µsec
at 1mph T
c
=
1
B
d
=
1
v/λ
= 0.74s T
s
∴ outage probability is a good measure.
at 10 mph T
c
= 0.074s T
s
∴ outage probability is a good measure.
at 100 mph T
c
= 0.0074s = 7400µs > 15µs outage or outage combined with average prob of error can
be a good measure.
11.
M
γ
(s) =
_
∞
0
e
sγ
p(γ)dγ
=
_
∞
0
e
sγ
1
γ
e
−γ/γ
dγ
=
1
1 −γs
12. (a) When there is path loss alone, d =
√
100
2
+ 500
2
= 100
√
6 ×10
3
P
e
=
1
2
e
−γ
b
⇒γ
b
= 13.1224
P
γ
N
0
B
= 13.1224 ⇒P
γ
= 1.3122 ×10
−14
P
γ
P
t
=
_
√
Gλ
4πd
_
2
⇒4.8488W
(b)
x = 1.3122 ×10
−14
= −138.82dB
P
γ,dB
∼ N(µP
γ
, 8), σ
dB
= 8
P(P
γ,dB
≥ x) = 0.9
P
_
P
γ,dB
−µP
γ
8
≥
x −µP
γ
8
_
= 0.9
⇒Q
_
x −µP
γ
8
_
= 0.9
⇒
x −µP
γ
8
= −1.2816
⇒µP
γ
= −128.5672dB = 1.39 ×10
−13
13. (a) Law of Cosines:
c
2
= a
2
+b
2
−2ab cos C with a,b =
√
E
s
, c = d
min
, C = Θ = 22.5
c = d
min
=
_
2E
s
(1 −cos 22.5) = .39
√
E
s
Can also use formula from reader
(b) P
s
= α
m
Q
_√
β
m
γ
s
_
= 2Q
_
_
d
min
2
2N
o
_
= 2Q(
√
.076γ
s
)
α
m
= 2, β
m
= .076
(c) P
e
=
∞
_
0
P
s
(γ
s
)f(γ
s
)dγ
s
=
∞
_
0
α
m
Q(
√
β
m
γ
s
)f(γ
s
)dγ
s
Using alternative Q form
=
α
m
π
π
2
_
0
_
1 +
gγ
s
(sinφ)
2
_
−1
dφ with g =
β
m
2
=
α
m
2
_
1 −
_
gγ
s
1+gγ
s
_
= 1 −
_
.038γ
s
1+.038γ
s
=
1
.076γ
s
, where we have used an integral table to evaluate
the integral
(d) P
d
=
P
s
4
(e) BPSK: P
b
=
1
4γ
b
= 10
−3
, ⇒ γ
b
= 250, 16PSK: From above get γ
s
= 3289.5
Penalty =
3289.5
250
= 11.2dB
Also will accept γ
b
(16PSK) = 822 ⇒= 5.2dB
14.
P
b
=
_
∞
0
P
b
(γ)p(γ)dγ
P
b
(γ) =
1
2
e
−γ
P
b
=
1
2
_
∞
0
e
−γb
p
γ
(γ)dγ =
1
2
M
But from 6.65
M
γ
(s) =
_
1 −
sγ
m
_
−m
∴ P
b
=
1
2
_
1 +
γ
m
_
−m
For M = 4, γ = 10
P
b
= 3.33 ×10
−3
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 mNakagami 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 BER
1 2.33×10
−2
2 5.53×10
−3
4 1.03×10
−3
16. For DPSK in Rayleigh fading, P
b
=
1
2γ
b
⇒γ
b
= 500
N
o
B = 3 ×10
−12
mW ⇒P
target
= γ
b
N
0
B = 1.5 ×10
−9
mW = 88.24 dBm
Now, consider shadowing:
P
out
= P[P
r
< P
target
] = P[Ψ < P
target
−P
r
] = Φ
_
P
target
−P
r
σ
_
⇒Φ
−1
(.01) = 2.327 =
P
target
−P
r
σ
P
r
= −74.28 dBm = 3.73 ×10
−8
mW = P
t
_
λ
4πd
_
2
⇒ d = 1372.4 m
17. (a)
γ
1
=
_
0 w.p. 1/3
30 w.p. 2/3
γ
2
=
_
5 w.p. 1/2
10 w.p. 1/2
In MRC, γ
Σ
= γ
1
+γ
2
. So,
γ
Σ
=
_
¸
¸
_
¸
¸
_
5 w.p. 1/6
10 w.p. 1/6
35 w.p. 1/3
40 w.p. 1/3
(b) Optimal Strategy is waterﬁlling with power adaptation:
S(γ)
S
=
_
1
γ
0
−
1
γ
, γ ≥ γ
0
0 γ < γ
0
Notice that we will denote γ
Σ
by γ only hereon to lighten notation. We ﬁrst assume γ
0
< 5,
4
i=1
_
1
γ
0
−
1
γ
i
_
p
i
= 1
⇒
1
γ
0
= 1 +
4
i=1
p
i
γ
i
⇒γ
0
= 0.9365 < 5
So we found the correct value of γ
0
.
C = B
4
i=1
log
2
_
γ
i
γ
0
_
p
i
C = 451.91 Kbps
(c) Without, receiver knowledge, the capacity of the channel is given by:
C = B
4
i=1
log
2
(1 +γ
i
)p
i
C = 451.66 Kbps
Notice that we have denote γ
Σ
by γ to lighten notation.
18. (a)
s(k) = s(k −1)
z(k −1) = g
k−1
s(k −1) +n(k −1)
z(k) = g
k
s(k) +n(k)
From equation 5.63 , the input to the phase comparator is
z(k)z
(k −1) = g
k
g
(
k −1)
s(k)s
(k −1) +g
k
s(k −1)n
k−1
+
g
(
k −1)
s
(k −1)n
k
+n
k
n
k−1
but s(k) = s(k −1)
s(k)s
(k −1) = s
k

2
= 1 (normalized)
(b)
¯ n
k
= s
k−1
n
k
¯ n
k−1
= s
k−1
n
k−1
¯ z = g
k
g
k−1
+g
k
¯ n
k−1
+g
k−1
¯ n
k
φ
x
(s) =
p
1
p
2
(s −p
1
)(s −p
2
)
=
A
s −p
1
+
B
s −p
2
A = (s −p
1
)φ
x
(s)
s=p
1
=
p
1
p
2
p
1
−p
2
B = (s −p
2
)φ
x
(s)
s=p
2
=
p
1
p
2
p
2
−p
1
(c) Relevant part of the pdf
φ
x
(s) =
p
1
p
2
(p
2
−p
1
)(s −p
2
)
∴ p
x
(x) =
p
1
p
2
(p
2
−p
1
)
L
−1
_
1
(s −p
2
)
_
=
p
1
p
2
(p
2
−p
1
)
e
p
2
x
, x < 0
(d)
P
b
= prob(x < 0) =
p
1
p
2
(p
2
−p
1
)
_
0
−∞
e
p
2
x
dx = −
p
1
p
2
−p
1
(e)
p
2
−p
1
=
1
2N
0
[γ
b
(1 −ρ
c
) + 1]
+
1
2N
0
[γ
b
(1 +ρ
c
) + 1]
=
γ
b
+ 1
N
0
[γ
b
(1 −ρ
c
) + 1][γ
b
(1 +ρ
c
) + 1]
∴ P
b
=
γ
b
(1 −ρ
c
) + 1
2(γ
b
+ 1)
(f) ρ
c
= 1
∴ P
b
=
1
2(γ
b
+ 1)
19. γ
b
0 to 60dB
ρ
c
= J
0
(2πB
D
T) with B
D
T = 0.01, 0.001, 0.0001
where J
0
is 0 order Bessel function of 1
st
kind.
P
b
=
1
2
_
1 +γ
b
(1 −ρ
c
)
1 +γ
b
_
when B
D
T = 0.01, ﬂoor can be seen about γ
b
= 40dB
when B
D
T = 0.001, ﬂoor can be seen about γ
b
= 60dB
when B
D
T = 0.0001, ﬂoor can be seen between γ
b
= 0 to 60dB
20. Data rate = 40 Kbps
Since DQPSK has 2 bits per symbol. ∴ T
s
=
2
40×10
3
= 5 ×10
−5
sec
DQPSK
Gaussian Doppler power spectrum, ρ
c
= e
−(πB
D
T)
2
B
D
= 80Hz
Rician fading K = 2
ρ
c
= 0.9998
P
floor
=
1
2
_
1 −
¸
(ρ
c
/
√
2)
2
1 −(ρ
c
/
√
2)
2
_
exp
_
−
(2 −
√
2)K/2
1 −ρ
c
/
√
2
_
= 2.138 ×10
−5
21. ISI:
Formula based approach:
P
floor
=
_
σT
m
T
s
_
2
Since its Rayleigh fading, we can assume that σ
T
m
≈ µ
T
m
= 100ns
P
floor
≤ 10
−4
which gives us
_
σT
m
T
s
_
2
≤ 10
−4
T
s
≥
σ
T
m
_
P
floor
= 10µsec
So, T
s
≥ 10µs. T
b
≥ 5µs. R
b
≤ 200 Kbps.
Thumb  Rule approach:
µ
t
= 100 nsec will determine ISI. As long as T
s
µ
T
, ISI will be negligible. Let T
s
≥ 10 µ
T
. Then
R ≤
2bits
symbol
1
T
s
symbols
sec
= 2Mbps
Doppler:
B
D
= 80 Hz
P
floor
= 10
−4
≥
1
2
_
_
1 −
¸
¸
¸
_
_
ρ
c
/
√
2
_
2
1 −
_
ρ
c
/
√
2
_
2
_
_
⇒ρ
c
≥ 0.9999
But ρ
c
for uniform scattering is J
0
(2πB
D
T
s
), so
ρ
c
= J
0
(2πB
D
T
s
) = 1 −(πf
D
T
s
)
2
≥ 0.9999
⇒T
s
≤ 39.79µs
T
b
≤ 19.89µs. R
b
≥ 50.26 Kbps.
Combining the two, we have 50.26 Kbps ≤ R
b
≤ 200 Kbps (or 2 Mbps).
22. From ﬁgure 6.5
with P
b
= 10
−3
, d = θ
T
m
/T
s
, θ
T
m
= 3µs
BPSK
d = 5 ×10
−2
T
s
= 60µsec
R = 1/T
s
= 16.7Kbps
QPSK
d = 4 ×10
−2
T
s
= 75µsec
R = 2/T
s
= 26.7Kbps
MSK
d = 3 ×10
−2
T
s
= 100µsec
R = 2/T
s
= 20Kbps
Chapter 7
1. P
s
= 10
−3
QPSK, P
s
= 2Q(
√
γ
s
) ≤ 10
−3
, γ
s
≥ γ
0
= 10.8276.
P
out
(γ
0
) =
M
i=1
_
1 −e
−
γ
0
γ
i
_
γ
1
= 10, γ
2
= 31.6228, γ
3
= 100.
M = 1
P
out
=
_
1 −e
−
γ
0
γ
1
_
= 0.6613
M = 2
P
out
=
_
1 −e
−
γ
0
γ
1
_ _
1 −e
−
γ
0
γ
2
_
= 0.1917
M = 3
P
out
=
_
1 −e
−
γ
0
γ
1
_ _
1 −e
−
γ
0
γ
2
_ _
1 −e
−
γ
0
γ
3
_
= 0.0197
2. p
γ
Σ
(γ) =
M
γ
_
1 −e
−γ/γ
¸
M−1
e
−γ/γ
γ = 10dB = 10
as we increase M, the mass in the pdf keeps on shifting to higher values of γ and so we have higher
values of γ 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)*(1exp(gamma/gamma_bar)).^...
(M(i)1).*(exp(gamma/gamma_bar));
end
3.
P
b
=
_
∞
0
1
2
e
−γ
p
γ
Σ
(γ)dγ
=
_
∞
0
1
2
e
−γ
M
γ
_
1 −e
−γ/γ
_
M−1
e
−γ/γ
dγ
=
M
2γ
_
∞
0
e
−(1+1/γ)γ
_
1 −e
−γ/γ
_
M−1
dγ
=
M
2γ
M−1
n=0
_
M −1
n
_
(−1)
n
e
−(1+1/γ)γ
dγ
=
M
2
M−1
n=0
_
M −1
n
_
(−1)
n
1
1 + n + γ
= desired expression
0 10 20 30 40 50 60
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
γ
p
γ
Σ
(
γ
)
M = 1
M = 2
M = 4
M = 8
M = 10
Figure 1: Problem 2
4.
p
γ
Σ
(γ) =
_
Pr{γ
2
< γ
τ
, γ
1
< γ} γ < γ
τ
Pr{γ
τ
≤ γ
1
≤ γ} + Pr{γ
2
< γ
τ
, γ
1
< γ} γ > γ
τ
If the distribution is iid this reduces to
p
γ
Σ
(γ) =
_
P
γ
1
(γ)P
γ
2
(γ
τ
) γ < γ
τ
Pr{γ
τ
≤ γ
1
≤ γ} + P
γ
1
(γ)P
γ
2
(γ
τ
) γ > γ
τ
5.
P
b
=
_
∞
0
1
2
e
−γ
p
γ
Σ
(γ)dγ
p
γ
Σ
(γ) =
_
_
1 −e
−γ
T
/γ
_
1
γ
e
−γ
r
/γ
γ < γ
T
_
2 −e
−γ
T
/γ
_
1
γ
e
−γ
r
/γ
γ > γ
T
P
b
=
1
2γ
_
1 −e
−γ
T
/γ
_
_
γ
T
0
e
−γ/γ
e
−γ
dγ +
1
2γ
_
2 −e
−γ
r
/γ
_
_
∞
γ
T
e
−γ/γ
e
−γ
dγ
=
1
2(γ + 1)
_
1 −e
−γ
T
/γ
+ e
−γ
T
e
−γ
T
/γ
_
6.
P
b
P
b
(10dB) P
b
(20dB)
no diversity
1
2(γ+1)
0.0455 0.0050
SC(M=2)
M
2
M−1
m=0
(−1)
m
M −1
m
1+m+γ
0.0076 9.7 ×10
−5
SSC
1
2(γ+1)
_
1 −e
−γ
T
/γ
+ e
−γ
T
e
−γ
T
/γ
_
0.0129 2.7 ×10
−4
As SNR increases SSC approaches SC
7. See
MATLAB CODE:
gammab_dB = [0:.1:20];
gammab = 10.^(gammab_dB/10);
M= 2;
0 2 4 6 8 10 12 14 16 18 20
10
−7
10
−6
10
−5
10
−4
10
−3
10
−2
10
−1
10
0
γ
avg
P
b
,a
v
g
(
D
P
S
K
)
M = 2
M = 3
M = 4
Figure 2: Problem 7
for j = 1:length(gammab)
Pbs(j) = 0
for m = 0:M1
f = factorial(M1)/(factorial(m)*factorial(M1m));
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;
for j = 1:length(gammab)
Pbs(j) = 0
for m = 0:M1
f = factorial(M1)/(factorial(m)*factorial(M1m));
Pbs(j) = Pbs(j) + (M/2)*((1)^m)*f*(1/(1+m+gammab(j)));
end
end
semilogy(gammab_dB,Pbs,’b.’);
hold on
M = 4;
for j = 1:length(gammab)
Pbs(j) = 0
for m = 0:M1
f = factorial(M1)/(factorial(m)*factorial(M1m));
Pbs(j) = Pbs(j) + (M/2)*((1)^m)*f*(1/(1+m+gammab(j)));
end
end
semilogy(gammab_dB,Pbs,’b:’);
hold on
8.
γ
Σ
=
1
N
0
_
M
i=1
a
i
γ
i
_
2
M
i=1
a
2
i
≤
1
N
0
a
2
i
γ
2
i
a
2
i
=
γ
2
i
N
0
Where the inequality above follows from CauchySchwartz condition. Equality holds if a
i
= cγ
i
where
c is a constant
9. (a) γ
i
= 10 dB = 10, 1 ≤ i ≤ N
N = 1, γ = 10, M = 4
P
b
= .2e
−1.5
γ
(M−1)
= .2e
−15/3
= 0.0013.
(b) In MRC, γ
Σ
= γ
1
+ γ
2
+ . . . + γ
N
.
So γ
Σ
= 10N
P
b
= .2e
−1.5
γ
Σ
(M−1)
= .2e
−5N
≤ 10
−6
⇒N ≥ 2.4412
So, take N = 3, P
b
= 6.12 ×10
−8
≤ 10
−6
.
10. Denote N(x) =
1
√
2π
e
−x
2
/2
, Q
(x) = −N(x)
P
b
=
_
∞
0
Q(
_
2γ)dP(γ)
Q(∞) = 0, P(0) = 0
d
dγ
Q(
_
2γ) = −N(
_
2γ)
√
2
2
√
γ
= −
1
√
2π
e
−γ
1
2
√
γ
P
b
=
_
∞
0
1
√
2π
e
−γ
1
2
√
γ
P(γ)dγ
P(γ) = 1 −e
−γ/γ
M
k=1
(γ/γ)
k−1
(k −1)!
1
_
∞
0
1
√
2π
e
−γ
1
2
√
γ
dγ =
1
2
2
_
∞
0
1
√
2π
e
−γ
1
2
√
γ
e
−γ/γ
M
k=1
(γ/γ)
k−1
(k −1)!
dγ =
M
k=1
1
(k −1)!
_
1
2
√
π
_
∞
0
e
−γ
1+
1
γ
γ
−1/2
_
γ
γ
_
k−1
dγ
_
Denote A =
_
1 +
1
γ
_
−1/2
=
M−1
m=0
1
m!
1
2
√
pi
_
∞
0
e
−γ/A
2
γ
−1/2
_
γ
γ
_
m
dγ
let γ/A
2
= u
=
M−1
m=0
1
m!
1
2
√
pi
_
∞
0
e
−u
u
−1/2
A
_
uA
2
γ
_
m
A
2
du
=
A
2
+
M−1
m=1
_
2m−1
m
_
A
2m
2
2m
A
γ
m
P
b
=
1 −A
2
−
M−1
m=1
_
2m−1
m
_
A
2m+1
2
2m
γ
m
11.
DenoteN(x) =
1
√
2π
e
−x
2
/2
Q
(x) = [1 −φ(x)]
= −N(x)
P
b
=
_
∞
0
Q(
_
2γ)dP(γ) =
_
∞
0
1
√
2π
e
−γ
1
√
2γ
P(γ)dγ
_
∞
0
1
√
2π
e
−γ
1
√
2γ
dγ =
1
√
π
Γ
_
1
2
_
=
1
2
(1)
_
∞
0
1
√
2π
e
−γ
1
√
2γ
e
−2γ/γ
dγ =
1
2
_
1 +
2
γ
(2)
_
∞
0
1
√
2π
e
−γ
1
√
2γ
_
πγ
γ
e
−γ/γ
_
1 −2Q
__
2γ
γ
__
dγ =
1
2
√
γ
1
B
√
Aγ
(3)
where A = 1 +
2
γ
, B = 1 +
1
γ
overall P
b
=
1
2
_
1 −
¸
1 −
1
(1 + γ)
2
_
12.
P
b
P
b
(10dB) P
b
(20dB)
no diversity
1
2
_
1 −
_
γ
b
1+γ
b
_
0.0233 0.0025
two branch SC
_
Q(
√
2γ)p
γ
Σ
dγ 0.0030 3.67 ×10
−5
two branch SSC
_
Q(
√
2γ)p
γ
Σ
dγ 0.0057 1.186 ×10
−4
two branch EGC
_
Q(
√
2γ)p
γ
Σ
dγ 0.0021 2.45 ×10
−5
two branch MRC
_
Q(
√
2γ)p
γ
Σ
dγ 0.0016 0.84 ×10
−5
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)*(1exp(gammat/gammab)).*exp(gamma1/gammab));
tointeg2 = Q(sqrt(2*gamma2)).*((1/gammab)*(2exp(gammat/gammab)).*exp(gamma2/gammab));
anssum(i) = sum(tointeg1)*.01+sum(tointeg2)*.01;
end
13. gammab_dB = [10];
gammab = 10.^(gammab_dB/10);
Gamma=sqrt(gammab./(gammab+1));
pb_mrc =(((1Gamma)/2).^2).*(((1+Gamma)/2).^0+2*((1+Gamma)/2).^1);
pb_egc = .5*(1sqrt(1(1./(1+gammab)).^2));
0 2 4 6 8 10 12 14 16 18 20
10
−5
10
−4
10
−3
10
−2
10
−1
γ
P
b
(
γ
)
MRC
EGC
dB penalty ~ .5 dB
Figure 3: Problem 13
14. 10
−3
= P
b
= Q(
√
2γ
b
) ⇒4.75, γ = 10
MRC P
out
= 1 −e
−γ
0
/γ
M
k=1
(γ
0
/γ)
k−1
(k−1)!
= 0.0827
ECG P
out
= 1 −e
−2γ
R
−
√
πγ
R
e
−γ
R
(1 −2Q(
√
2γ
R
)) = 0.1041 > P
out,MRC
15. P
b,MRC
= 0.0016 < 0.0021P
b,EGC
16. If each branch has γ = 10dB Rayleigh
γ
Σ
= overall recvd SNR =
γ
1
+γ
2
2
∼
γe
−γ/(γ/2)
(γ/2)
2
γ ≥ 0
BPSK
P
b
=
_
∞
0
Q(
_
2γ)p
γ
Σ
dγ = 0.0055
17. p(γ) where
_
∞
0
p(γ)e
−xγ
dγ =
0.01γ
√
x
we will use MGF approach
P
b
=
1
π
_
π/2
0
Π
2
i=1
M
γ
i
_
−
1
sin
2
φ
_
dφ
=
1
π
_
π/2
0
(0.01γ sinφ)
2
dφ
=
(0.01γ)
2
4
= 0.0025
18.
P
b
=
_
1 −π
2
_
3 2
m=0
_
l + m
m
__
1 + π
2
_
m
; π =
_
γ
1 + γ
Nakagami2 fading
M
γ
_
−
1
sin
2
φ
_
=
_
1 +
γ
2 sin
2
φ
_
−2
P
b
=
1
π
_
π/2
0
_
M
γ
_
−
1
sin
2
φ
__
3
dφ, γ = 10
1.5
= 5.12 ×10
−9
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 = ((1Gamma)/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.
P
b
=
1
π
_
π/2
0
_
1 +
γ
2 sin
2
φ
_
−2
_
1 +
γ
sin
2
φ
_
−1
dφ
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
5 10 15 20
10
−7
10
−6
10
−5
10
−4
10
−3
10
−2
γ
avg
(dB)
P
b
a
v
g
Figure 4: Problem 19
20.
P
b
=
2
3
Q
_
_
2γ
b
(3) sin
_
π
8
__
α = 2/3, g = 3 sin
2
_
π
8
_
M
γ
_
−
g
sin
2
φ
_
=
_
1 +
gγ
sin
2
φ
_
−1
P
b
=
α
π
_
π/2
0
_
1 +
gγ
sin
2
φ
_
−M
dφ
MATLAB CODE:
M = [1 2 4 8];
alpha = 2/3; g = 3*sin(pi/8)^2;
gammab_dB = [5:.1:20];
gammabvec = 10.^(gammab_dB/10);
for k = 1:length(M)
for i = 1:length(gammabvec)
gammab = gammabvec(i);
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
5 10 15 20
10
−15
10
−10
10
−5
10
0
γ
avg
(dB)
P
b
a
v
g
Figure 5: Problem 20
21.
Q(z) =
1
π
_
π/2
0
exp
_
−
z
2
sin
2
φ
_
dφ , z > 0
Q
2
(z) =
1
π
_
π/4
0
exp
_
−
z
2
2 sin
2
φ
_
dφ , z > 0
P
s
(γ
s
) =
4
π
_
1 −
1
√
M
__
π/2
0
exp
_
−
gγ
s
sin
2
φ
_
dφ −
4
π
_
1 −
1
√
M
2
__
π/4
0
exp
_
−
gγ
s
sin
2
φ
_
dφ
P
s
=
_
∞
0
P
s
(γ
Σ
)p
γ
Σ
(γ
Σ
)dγ
Σ
=
4
π
_
1 −
1
√
M
__
π/2
0
_
∞
0
exp
_
gγ
Σ
sin
2
φ
_
p
γ
Σ
(γ)dγ
Σ
dφ −
4
π
_
1 −
1
√
M
_
2
_
π/4
0
_
∞
0
exp
_
gγ
Σ
sin
2
φ
_
p
γ
Σ
(γ)dγ
Σ
dφ
But γ
Σ
= γ
1
+ γ
2
+ . . . + γ
M
= Σγ
i
=
4
π
_
1 −
1
√
M
__
π/2
0
Π
M
i=1
M
γ
i
_
−
g
sin
2
φ
_
dφ −
4
π
_
1 −
1
√
M
_
2
_
π/4
0
Π
M
i=1
M
γ
i
_
−
g
sin
2
φ
_
dφ
22. Rayleigh: M
γ
s
(s) = (1 −sγ
s
)
−1
Rician: M
γ
s
(s) =
1+k
1+k−sγ
s
exp
_
k
s
γ
s
1+k−sγ
s
_
MPSK
P
s
=
_
(M−1)π/M
0
Mγ
s
_
−
g
sin
2
φ
_
dφ →no diversity
Three branch diversity
P
s
=
1
π
_
(M−1)π/M
0
_
1 +
gγ
sin
2
φ
_
−1
_
(1 + k) sin
2
φ
(1 + k) sin
2
φ + gγ
s
exp
_
−
kγ
s
g
(1 + k) sin
2
φ + gγ
s
__
2
dφ
g = sin
2
_
π
16
_
= 0.1670
MQAM:
Formula derived in previous problem with g =
1.5
16−1
=
1.5
15
P
s
= 0.0553
MATLAB CODE:
gammab_dB = 10;
gammab = 10.^(gammab_dB/10);
K = 2;
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/(161);
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;
5 10 15 20
10
−9
10
−8
10
−7
10
−6
10
−5
10
−4
10
−3
10
−2
10
−1
10
0
P
s
a
v
g
Figure 6: Problem 22
23. MATLAB CODE:
M = [1 2 4 8];
alpha = 2/3;
g = 1.5/(161);
gammab_dB = [5:.1:20];
gammabvec = 10.^(gammab_dB/10);
for k = 1:length(M)
for i = 1:length(gammabvec)
gammab = gammabvec(i);
phi1 = [0.001:.001:pi/2];
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 10
1. (a)
(AA
H
)
T
= (A
H
)
T
.A
T
= (A
T
)
T
A
T
= AA
H
∴ (AA
H
)
H
= AA
H
For AA
H
, λ = λ, i.e. eigenvalues are real
AA
H
= QΛQ
H
(b) X
H
AA
H
X = (X
H
A)(X
H
A)
H
= X
H
A ≥ 0
∴ AA
H
is positive semideﬁnite.
(c) I
M
+AA
H
= I
M
+QΛQ
H
= Q(I + Λ)Q
H
A
H
positive semideﬁnite ⇒λ
i
≥ 0∀i
∴ 1 +λ
i
> 0∀i
∴ I
M
+AA
H
positive deﬁnite
(d)
det[I
M
+AA
H
] = det[I
M
+QΛQ
H
]
= det[Q(I
M
+ Λ
M
)Q
H
]
= det[I
M
+ Λ
M
]
= Π
Rank(A)
i=1
(1 +λ
i
)
det[I
N
+A
H
A] = det[I
N
+
¯
QΛ
¯
Q
H
]
= det[
¯
Q(I
N
+ Λ
N
)
¯
Q
H
]
= det[I
N
+ Λ
N
]
= Π
Rank(A)
i=1
(1 +λ
i
)
∵ AA
H
and A
H
A have the same eigenvalue
∴ det[I
M
+AA
H
] = det[I
N
+A
H
A]
2. H = UΣV
T
U =
_
_
−0.4793 0.8685 −0.1298
−0.5896 −0.4272 −0.6855
−0.6508 −0.2513 0.7164
_
_
Σ =
_
_
1.7034 0 0
0 0.7152 0
0 0 0.1302
_
_
V =
_
¸
¸
_
−0.3458 0.6849 0.4263
−0.5708 0.2191 0.0708
−0.7116 −0.6109 0.0145
−0.2198 0.3311 −0.9017
_
¸
¸
_
3. H = UΣV
T
Let
U =
_
_
1 0
0 1
0 0
_
_
V =
_
_
1 0
0 1
0 0
_
_
Σ =
_
1 0
0 2
_
∴ H =
_
_
1 0 0
0 2 0
0 0 0
_
_
4. Check the rank of each matrix
rank(H
I
) = 3
∴ multiplexing gain = 3
rank(H
2
) = 4
∴ multiplexing gain = 4
5.
C =
R
H
i=1
log
2
_
1 +
λ
i
ρ
M
t
_
Constraint
V
i
= ρ
λ
i
= constant
∴
∂C
∂λ
i
=
ρ
M
t
ln 2
1
(1 +
λ
i
ρ
M
t
)
−
ρ
M
t
ln 2
1
(1 +
λ
i
ρ
M
t
)
= 0
⇒λ
i
= λ
j
∴ when all R
H
singular values are equal, this capacity is maximized.
6. (a) Any method to show H ≈ UΛV is acceptable. For example:
D =
_
_
.13 .08 .11
.05 .09 .14
.23 .13 .10
_
_
where : d
ij
=
¸
¸
¸
H
ij
−H
H
ij
¸
¸
¸ ×100
(b) precoding ﬁlter M = V
−1
shaping ﬁlter F = U
−1
F =
_
_
−.5195 −.3460 −.7813
−.0251 −.9078 .4188
−.8540 .2373 .4629
_
_
M =
_
_
−.2407 −.8894 .3887
−.4727 −.2423 −.8473
−.8478 .3876 .3622
_
_
Thus Y = F(H)MX + FN = U
∗
UΛVV
∗
X +U
∗
N
= ΛX +U
∗
N
(c)
P
i
P
=
1
γ
o
−
1
γ
i
for
1
γ
i
>
1
γ
o
, 0 else
γ
i
=
λ
i
2
P
N
o
B
= 94.5 for i = 1, 6.86 for i = 2, .68 for i = 3
Assume γ
2
> γ
0
> γ
3
since γ
3
= .68 is clearly too small for data transmission
P
i
P
= 1 ⇒
2
γ
0
−
1
γ
1
−
1
γ
2
= 1 ⇒γ
0
= 1.73
P
1
P
= .5676
P
2
P
= .4324
C = B
_
log
2
_
1 +γ
1
P
1
P
_
+ log
2
_
1 +γ
2
P
2
P
_¸
= 775.9 kbps
(d) With equal weight beamforming, the beamforming vector is given by c =
1
√
(3)
[1 1 1]. The SNR
is then given by:
SNR =
c
H
H
H
Hc
N
0
B
= (.78)(100) = 78. (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 Blog
2
det[I
Mγ
+HR
X
H
H
]
R
X
: T
γ
(R
X
) = ρ If the channel is known to the transmitter, it will perform an SVD decomposition of
H as
H = UΣV
HR
X
H
H
= (UΣV )R
X
(UΣV )
H
By Hadamard’s inequality we have that for A ∈
n×n
det(A) ≤ Π
n
i=1
A
ii
with equality iﬀ A is diagonal.
We choose R
X
to be diagonal, say = Ω then
det(I
MR
+HR
X
H
H
) = det(I + ΩΣ
2
)
∴ C = max
i
ρ
i
≤ρ
Bσ
i
log
2
(1 +λ
i
ρ
i
)
where
√
λ
i
are the singular values.
8. The capacity of the channel is found by the decomposition of the channel into R
H
parallel channels,
where R
H
is the rank of the channel matric H.
C = max
ρ
i
:
i
ρ
i
≤ρ
i
Blog
2
(1 +λ
i
ρ
i
)
where
√
λ
i
are the R
H
nonzero singular values of the channel matrix H and ρ is the SNR constraint.
γ
i
= λ
i
ρ
Then the optimal power allocation is given as
P
i
P
=
_
1
γ
0
−
1
γ
i
γ
i
≥ γ
0
0 γ
i
< γ
0
(2)
for some cutoﬀ value γ
0
. The resulting capacity is given as
C =
i:γ
i
≥γ
0
Blog
2
(γ
i
/γ
0
)
For
H =
_
¸
¸
_
1 1 −1 1
1 1 −1 −1
1 1 1 1
1 1 1 −1
_
¸
¸
_
R
H
= 3, γ
1
= 80, γ
2
= 40, γ
3
= 40. We ﬁrst assume that γ
0
is less than the minimum γ
i
which is 40.
γ
0
=
3
1 +
3
i=1
1
γ
i
which gives γ
0
= 2.8236 < min
i
γ
i
, hence the assumption was correct.
C
B
= 12.4732 bits/sec/Hz
For
H =
_
¸
¸
_
1 1 1 −1
1 1 −1 1
1 −1 1 1
1 −1 −1 −1
_
¸
¸
_
R
H
= 4, γ
1
= 40, γ
2
= 40, γ
3
= 40, γ
4
= 40. We ﬁrst assume that γ
0
is less than the minimum γ
i
which is 40.
γ
0
=
4
1 +
4
i=1
1
γ
i
which gives γ
0
= 3.6780 < min
i
γ
i
, hence the assumption was correct.
C
B
= 13.7720 bits/sec/Hz
9. H =
_
¸
¸
¸
¸
_
h
11
. . . h
1M
t
. . . . .
. . . . .
. . . . .
h
M
r
1
. . . h
M
r
M
t
_
¸
¸
¸
¸
_
M
r
×M
t
Denote G = HH
T
lim
M
t
→∞
1
M
t
G
ii
= lim
M
t
→∞
1
M
t
[h
i1
. . . h
iM
t
]
_
¸
¸
¸
¸
_
h
i1
.
.
.
h
iM
t
_
¸
¸
¸
¸
_
= lim
M
t
→∞
1
M
t
M
t
j=1
h
ij
2
= E
j
h
ij
2
= σ
2
= 1 ∀i
lim
M
t
→∞,i=j
1
M
t
G
ij
= lim
M
t
→∞
1
M
t
[h
i1
. . . h
iM
t
]
_
¸
¸
¸
¸
_
h
j1
.
.
.
h
jM
t
_
¸
¸
¸
¸
_
= lim
M
t
→∞
1
M
t
M
t
k=1
h
ik
h
jk
= E
k
h
ik
h
jk
= E
k
(h
ik
)E
k
(h
jk
)
= 0 ∀i, j, i = j
∴ lim
M→∞
1
M
HH
T
= I
M
∴ lim
M→∞
Blog
2
det
_
I
M
+
ρ
M
HH
T
_
= Blog
2
det [I
M
+ρI
M
]
= Blog
2
[1 +ρ] det I
M
= MBlog
2
[1 +ρ]
10. We ﬁnd the capacity by randomly generating 10
3
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));
gammatemp1 = gammatemp;
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
0 5 10 15 20 25
0
5
10
15
20
25
ρ (dB)
C
e
r
g
o
d
i
c
M
t
= M
r
= 3
M
t
= 2 M
r
=3
M
t
= M
r
= 2
M
t
= 2 M
r
=1
M
t
= M
r
= 1
Figure 1: Problem 10
11. We ﬁnd the capacity by randomly generating 10
4
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 k = 1:length(rho)
for i = 1:1000
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;
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;
pout = sum(C<Cout(k))/length(C);
end
if Cout(k)<0;
Cout(k) = 0;
end
end
0 5 10 15 20 25 30
0
5
10
15
20
25
ρ (dB)
C
o
u
t
a
g
e
M
t
= M
r
= 3
M
t
= 2 M
r
=3
M
t
= M
r
= 2
M
t
= 2 M
r
=1
M
t
= M
r
= 1
Figure 2: Problem 11
12.
P (u
n < X) = P
_
M
r
i=1
u
i
n
i
< X
_
=
M
r
i=1
u
i
P (n
i
< X)
= P (n
i
< X)
∴ the statistics of u
n are the same as the statistics of each of these elements
13.
Σ
x
= u
Hvx
= u
Hv
2
x
2
= v
H
H
H
u
H
u
Hvx
2
= v
H
H
H
Hvx
2
= v
H
Q
H
Qvx
2
≤ λ
max
x
2
with equality when u, v are the principal left and right singular vectors of the channel matrix H
∴ SNR
max
= λ
max
x
2
N
= λ
max
ρ
14.
H =
_
_
0.1 0.5 0.9
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 corre
sponding to this value are
u
opt
=
_
_
−0.7101
−0.4641
−0.5294
_
_
and
v
opt
=
_
_
−0.1818
−0.4190
−0.8896
_
_
It is easy to check that u
T
opt
u
opt
= 1 and v
T
opt
v
opt
= 1 and that
u
T
opt
∗ H ∗ v
opt
= 1.4480
Since, during beamforming from eq. 10.17 in reader,
y = (u
T
Hv)x +u
T
n
and for a given transmit SNR of ρ, the received SNR is given as
SNR
rcvd
= ρ(u
T
opt
Hv
opt
)
2
since, u
opt
has norm 1, noise power is not increased. For, ρ = 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
v
2
=
1
√
3
_
_
1
1
1
_
_
Deﬁne
h = Hv
2
So, eq. 10.17 in the reader can be written as
y = (u
T
h)x +u
T
n
To maximize SNR we need to ﬁnd a u
2
of norm 1 such that (u
h) is maximized.
Using Matlab, we get that the maximum singular value of h is 1.2477 and the singular vector corre
sponding to this value is
u
2
=
_
_
0.6941
0.5090
0.5090
_
_
It is easy to check that u
T
2
u
2
= 1 and that
u
T
2
∗ h = 1.2477
Alternatively, from MRC concept we know that:
u
2
=
h
H
h
=
_
_
0.6941
0.5090
0.5090
_
_
where h is the L
2
norm of h.
For a given transmit SNR of ρ, the received SNR is given as
SNR
rcvd
= ρ(u
T
2
Hv
2
)
2
since, u
2
has norm 1, noise power is not increased. For, ρ = 1, SNR is simply (1.2477)
2
= 1.5567.
15. (a) ρ = 10 dB = 10
P
e
= ρ
−d
So to have P
e
≤ 10
−3
, we should have d ≥ 3, or at least d = 3. Solving the equation that relates
diversity gain d to multiplexing gain r at high SNR’s we get
d = (M
r
−r)(M
t
−r)
⇒3 = (8 −r)(4 −r)
Solving for r we get
r = 3.35 or 8.64
We have that r ≤ min{M
r
, M
t
}, 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 SNR’s too and we are
given that we can use base 2 for logarithms, we can ﬁnd the data rate as
R = r log
2
(ρ) = 9.96 bits/s/Hz
(b) With, r = 3, we can ﬁnd d as
d = (M
r
−r)(M
t
−r) = (8 −3)(4 −3) = 5
For this value of d,
P
e
= ρ
−d
= 10
−5
16. According to SVD of h
√
λ = 1.242
∴ C/B = log
2
(1 +λρ) = log
2
(1 + 1.242
2
.10) = 4.038bps/Hz
17.
H =
_
.3 .5
.7 .2
_
=
_
−.5946 .8041
−.8041 .5946
_ _
.8713 0
0 .3328
_ _
−.8507 .5757
−.5757 −.8507
_
P= 10mW
N
0
= 10
−9
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 chan
nel into 2 parallel channels. We can then do waterﬁlling over the two parallel channels available
to get capacity.
Finding the γ
i
’s
γ
1
=
λ
2
1
P
N
0
B
= 75.92
γ
2
=
λ
2
2
P
N
0
B
= 11.08
Finding γ
0
Now, we have to ﬁnd the cutoﬀ value γ
0
. First assume that γ
0
is less than both γ
1
and γ
2
. Then
_
1
γ
0
−
1
γ
1
_
+
_
1
γ
0
−
1
γ
2
_
= 1
⇒
2
γ
0
= 1 +
1
γ
1
+
1
γ
2
⇒γ
0
=
1
1 +
1
γ
1
+
1
γ
2
= 1.81
which is less than both γ
1
and γ
2
values so our assumption was correct.
Finding capacity
Now we can use the capacity expression as
C =
2
i=1
Blog
2
_
γ
i
γ
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(γ) = 1 +γK
S(γ)
S
Finding K
K =
−1.5
ln(5P
b
)
= .283
γ
K
= γ
0
/K.
Finding γ
0
or γ
K
We now ﬁnd the cutoﬀ γ
0
. First assume that γ
0
< {γ
1
, γ
2
}. Notice that γ
1
and γ
2
have already
been calculated in part (a) as γ
1
= 75.92 and γ
2
= 11.08.
_
1
γ
0
−
1
γ
1
K
_
+
_
1
γ
0
−
1
γ
2
K
_
= 1
⇒
2
γ
0
= 1 +
1
γ
1
K
+
1
γ
2
K
⇒γ
0
=
1
1 +
1
K
_
1
γ
1
+
1
γ
2
_ = 1.4649
which is less than both γ
1
and γ
2
values, so our assumption was correct.
γ
K
= γ
0
/K = 5.1742
Finding Rate R
Therefore the total rate, R is given as
R = B
_
log
2
_
γ
1
γ
K
_
+ log
2
_
γ
2
γ
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 HH
H
.
Once this is done the SNR at the combiner output is simply λ
max
ρ, where λ
max
is the maximum
eigen value of the Wishart Matrix HH
H
and ρ is
P
N
0
B
Finding γ
s
As given in the question, λ
max
is 0.7592 and ρ was calculated to be 100. So we get that γ
s
= 75.92 .
Finding P
b
When using BPSK, γ
s
= γ
b
. Now we can use the expression for P
b
for BPSK
P
b
= Q
_
_
2γ
b
_
= Q
_√
2 ×75.92
_
=
_
0 Using the approx. given in the Ques.
3.4 ×10
−35
Using Matlab
Credit is given for either value.
Finding Rate R
Since we are using BPSK and are given that B = 1/T
b
, 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 P
b
im
proves as P
b
is now 3.4 ×10
−35
∼ 0 whereas earlier it was 10
−3
.
(d) Therefore we see that we can tradeoﬀ rate for robustness of the system. If we are willing to de
crease the rate at which we transmit, we can get more diversity advantage i.e. one strong channel
which gives a much less value of P
b
.
18. (a) clear;
clc;
Mt = 4;
Mr = Mt;
rho_dB = [0:20];
rho = 10.^(rho_dB/10);
for k = 1:length(rho)
for i = 1:1000
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));
gammatemp1 = gammatemp;
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 k = 1:length(rho)
for i = 1:1000
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;
C(i) = sum(log2(1+gamma/Mt));
end
Cout(k) = mean(C);
end
19. using Matlab we get C
out
= 7.8320
MATLAB CODE
clear;
clc;
Mt = 4;
Mr = Mt;
rho_dB = 10;
rho = 10.^(rho_dB/10);
for k = 1:length(rho)
for i = 1:1000
0 2 4 6 8 10 12 14 16 18 20
0
5
10
15
20
25
ρ (dB)
C
C
ergodic
M
t
= M
r
= 1
C
out
M
t
= M
r
= 1
C
ergodic
M
t
= M
r
= 4
C
out
M
t
= M
r
= 4
Figure 3: Problem 18
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;
C(i) = sum(log2(1+gamma/Mt));
end
Cout(k) = mean(C);
pout = sum(C<Cout(k))/length(C);
while pout > .1
Cout(k) = Cout(k).01;
pout = sum(C<Cout(k))/length(C);
end
if Cout(k)<0;
Cout(k) = 0;
end
end
20. As µ 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 = 1:1000
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*sigma_used;
C(i) = sum(log2(1+gamma/Mt));
end
[f,x] = ecdf(C);
6 7 8 9 10 11 12 13 14 15 16 17 18
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Empirical CDF’s of Capacity w/o Tx CSI
x
F
X
(
x
)
M = 4
M = 6
M = 8
Figure 4: Problem 20
Chapter 11
1. See Fig 1
fc = 100 MHz
fc+B fcB
2B = 100 KHz
Figure 1: Band of interest.
B = 50 KHz, f
c
= 100 MHz
H
eq
(f) =
1
H(f)
= f
Noise PSD = N
0
W/Hz. Using this we get
Noise Power =
_
f
c
+B
f
c
−B
N
0
H
eq
(f)
2
df (1)
= N
0
_
f
c
+B
f
c
−B
f
2
df (2)
= N
0
f
3
3
_
(f
c
+B)
(f
c
−B)
(3)
=
N
0
3
(f
c
+ B)
3
−(f
c
−B)
3
(4)
= 10
21
N
0
W (5)
Without the equalizer, the noise power will be 2BN
0
= 10
5
N
0
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 = A
2
T
b
/T
b
= A
2
For the 2nd channel:
ISI power over a bit time =
A
2
T
b
∞
n=1
_
(n+1)T
b
nT
b
e
−t/T
m
dt = 2e
−1/2
A
2
A
S(t)
T
m
/2
1
h
T
m
1
(t)
h
2
(t)
t
t
t
Figure 2: Problem 2a
(b) No ISI: pulse interval = 11/2µs = 5.5µs
∴ Data rate = 1/5.5µs = 181.8Kbps
If baseband signal =100KHz: pulse width = 10µs
Data rate = 2/10µs + 10µs = 100Kbps
3. (a)
h(t) =
_
e
−
t
τ
t ≥ 0
0 o.w.
(6)
τ = 6 µ sec
H
eq
(f) =
1
H(f)
H(f) =
_
∞
0
e
−
t
τ
e
−j2πft
dt (7)
=
1
1
τ
+ j2πf
(8)
Hence,
H
eq
(f) =
1
τ
+ j2πf
(b)
SNR
eq
SNR
ISI
=
B
−B
S
x
(f)H(f)
2
H
eq
(f)
2
df
B
−B
N
0
H
eq
(f)
2
df
B
−B
S
x
(f)H(f)
2
df
2BN
0
1
h
10us
1
(t)
t
1
X
1us
(t)
t
h
12us
1
(t)*X(t)
t
1
Figure 3: Problem 2b
Assume S
x
(f) = S, −B ≤ f ≤ B ⇒
2BS
N
0
2B
τ
2
+
8π
2
3
B
3
S
B
−B
H(f)
2
df
2BN
0
=
2B
1
τ
2
+
4π
2
3
B
2
1.617 ×10
−6
= 0.9364 = −0.28 dB
(c)
h[n] = 1 + e
−T
s
τ
δ [n −1] + e
−2T
s
τ
δ [n −2] + ...
H(z) = 1 +e
−T
s
τ
z
−1
+ e
−2T
s
τ
z
−2
+ e
−3T
s
τ
z
−3
+ ...
=
∞
n=0
_
e
−
T
s
τ
z
−1
_
n
=
z
z −e
−
T
s
τ
=
1
1 −e
−
T
s
τ
z
−1
⇒ H
eq
(z) =
1
H(z)+N
0
. Now, we need to use some approximation to come up with the ﬁlter tap
coeﬃcient values. If we assume N
0
0 (the zeroforcing assumption), we get H
eq
(z) = 1−e
−
T
s
τ
z
−1
.
Thus, a two tap ﬁlter is suﬃcient. For T
s
=
1
30
ms, we have a
0
=1, a
1
= −0.0039 as the tap
coeﬃcient values. Any other reasonable way is also accepted.
4. ω
i
= c
i
where {c
i
} is the inverse Z transform of 1/F(z)
Show that this choice of tap weights minimizes
¸
¸
¸
¸
1
F(z)
−(ω
−N
z
N
+ . . . + ω
N
z
−N
)
¸
¸
¸
¸
2
. . . (1)
at z = e
jω
If F(z) is of length 2 and monic, say F(z) = 1 −a
1
z then
1
F(z)
= 1 + a
1
z
−1
+ a
2
1
z
−2
+ . . . where c
1
= a
1
, c
2
= a
2
1
, a
1
< 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 ω
i
= c
i
minimizes (1). The result can be similarly proved for length of F(z) greate than 2 or nonmonic.
5. (a) H
eq
(f) =
1
H(f)
for ZF equalizer
H
eq
(f) =
_
¸
¸
¸
¸
_
¸
¸
¸
¸
_
1 f
c
−20MHz ≤ f < f
c
−10MHz
2 f
c
−10MHz ≤ f < f
c
0.5 f
c
≤ f < f
c
+ 10MHz
4 f
c
+ 10MHz ≤ f < f
c
+ 20MHz
0 o.w.
(9)
(b) S=10mW Signal power
N = N
0
[1
2
×10MHz + 2
2
×10MHz + 0.5
2
×10MHz + 4
2
×10MHz] = 0.2125mW
∴ SNR = 47.0588 = 16.73dB
(c) T
s
= 0.0125µsec
P
b
≤ 0.2e
−1.5γ/M−1
or M ≤ 1 +
1.5SNR
−ln(5P
b
)
for P
b
= 10
−3
M ≤ 14.3228 (M ≥ 4 thus using the formula is reasonable)
R =
log
2
M
T
s
= 307.2193Mbps
(d) We use M=4 non overlapping subchannels, each with B=10MHz bandwidth
1: f
c
−20MHz ≤ f < f
c
−10MHz α
1
= 1
2: f
c
−10MHz ≤ f < f
c
α
2
= 0.5
3: f
c
≤ f < f
c
+ 10MHz α
3
= 2
4: f
c
+ 10MHz ≤ f < f
c
+ 20MHz α
4
= 0.25
Power optimization: γ
i
=
Pα
2
i
N
0
B
for i = 1, 2, 3, 4
γ
1
= 1000 γ
2
= 250 γ
3
= 4000 γ
4
= 62.5
for P
b
= 10
−3
K = 0.2831
P
i
P
=
_
1
γ
0
−
1
Kγ
i
γ
i
≥ γ
0
/K
0 γ
i
≤ γ
0
/K
(10)
We can see that all subchannels will be used and
P
1
= 2.6523 P
2
= 2.5464 P
3
= 2.6788 P
4
= 2.1225
and
γ
0
= 3.7207
thus R = 2B
4
i=1
log(Kγ
i
/γ
0
) = 419.9711Mbps
6. (a)
F{f(t)} =
_
T f < 1/T
0 o.w.
F
Z
(f) =
1
T
S
∞
n=−∞
F
_
f +
n
T
s
_
= 1
∴ folded spectrum of f(t) is ﬂat.
(b)
y
k
= y(kT + t
0
)
=
∞
i=−∞
X
i
f(kT + t
0
−iT)
=
N
i=−N
X
i
f ((k −i)T + t
0
)
= X
k
sinc(t
0
) +
N+k
i=−N+k,i=k
X
i
f ((k −i)T + t
0
)
. ¸¸ .
ISI
(c)
ISI =
N+k
i=−N+k,i=k
X
i
sin (π(k −i) + t
0
/Tπ)
π(k −i) + t
0
/Tπ
= sin(πt
0
/T)
N
i=−N,i=0
1
πt
0
/T −πi
= sin(πt
0
/T)
N
i=1
_
−1
πt
0
/T −πi
+
1
πt
0
/T −πi
_
=
2
π
sin(πt
0
/T)
N
n=1
n
n
2
−t
2
0
/T
2
Thus, ISI →∞ as N →∞
7. g
m
(t) = g
(−t) = g(t) = sinc(t/T
S
), t < T
s
Noise whitening ﬁlter :
1
G
m
(1/z
)
8. J
min
= 1 −
∞
j=−∞
c
j
f
−j
B(z) = C(z)F(z)
=
F(z)F
(z
−1
)
F(z)F
(z
−1
) + N
0
=
X(z)
X(z) + N
0
∴ b
0
=
1
2πj
_
B(z)
z
dz
=
1
2πj
_
X(z)
z[X(z) + N
0
]
dz
=
T
2π
_
π/T
−π/T
X(e
jωT
)
X(e
jωT
) + N
0
dω
∴ J
min
= 1 −
T
2π
_
π/T
−π/T
X(e
jωT
)
X(e
jωT
) + N
0
dω
=
T
2π
_
π/T
−π/T
N
0
X(e
jωT
) + N
0
dω
=
T
2π
_
π/T
−π/T
N
0
T
−1
∞
n=−∞
H(ω + 2πn/T)
2
+ N
0
dω
= T
s
_
−0.5T
s
−0.5T
s
N
0
F
Σ
(f) + N
0
df
9.
V
W
J =
_
∂J
∂w
0
, . . . ,
∂J
∂w
N
_
J = w
T
M
v
w
−2{V
d
w
} + 1
∴
∂J
∂w
= 2M
v
w
T
−2V
d
∂J
∂w
= 0 ⇒ 2M
v
w
T
= 2V
d
⇒ w
opt
=
_
M
T
v
_
−1
V
H
d
10.
J
min
= T
s
_
0.5T
s
−0.5T
s
N
0
F
Σ
(f) + N
0
df
∵
N
0
F
Σ
(f) + N
0
≥ 0 ∴ J
min
≥ 0
N
0
F
Σ
(f) + N
0
≤
N
0
N
0
= 1
∴ J
min
≤ T
s
_
0.5T
s
−0.5T
s
1df = 1
∴ 0 ≤ J
min
≤ 1
11.
F
Σ
(f) =
1
T
s
∞
n=−∞
F
_
f +
n
T
s
_
=
1
T
s
∞
n=−∞
1 + 0.5e
−j2π
f+
n
T
s
+ 0.3e
−j4π
f+
n
T
s
MMSE equalizer :
J
min
= T
s
_
0.5/T
s
−0.5/T
s
N
0
F
Σ
(f) + N
0
df
DF equalizer :
J
min
= exp
_
T
s
_
0.5/T
s
−0.5/T
s
ln
_
N
0
F
Σ
(f) + N
0
_
_
df
12. (a) G(f) is a sinc(), so theoretically inﬁnite. But 2/T is also acceptable (Null to Null bandwidth)
(b) τ T is more likely since T = 10
−9
sec
As long as τ > T
b
, get ISI and so, frequency selective fading
(c) Require T
b
= T
m
+ T ⇒R =
1
T
m
+T
= 49997.5bps
(d) H
eq
(z) =
1
F(z)
for ZF equalizer
⇒H
eq
(z) =
1
d
0
+d
1
z
−1
+d
2
z
−2
Long division yields the ﬁrst 2 taps as
w
0
= 1/α
0
w
1
= −α
1
/α
2
0
13. (a)
H
zf
(f) =
1
H(f)
=
_
¸
¸
¸
¸
¸
¸
_
¸
¸
¸
¸
¸
¸
_
1 0 ≤ f ≤ 10KHz
2 10KHz ≤ f ≤ 20KHz
3 20KHz ≤ f ≤ 30KHz
4 30KHz ≤ f ≤ 40KHz
5 40KHz ≤ f ≤ 50KHz
(b) The noise spectrum at the output of the ﬁlter is given by N(f) = N
0
H
eq
(f)
2
, and the noise
power is given by the integral of N(f) from 50 kHz to 50 kHz:
N =
_
50kHz
f=−50kHz
N(f)df = 2N
0
_
50kHz
f=0kHz
H
eq
(f)
2
df
= 2N
0
(1 + 4 + 9 + 16 + 25)(10kHz)
= 1.1mW
(c) The noise spectrum at the output of the ﬁlter is given by N(f) =
N
0
(H(f)+α)
2
, and the noise power
is given by the integral of N(f) from 50 kHz to 50 kHz. For α = .5 we get
N = 2N
0
(.44 + 1 + 1.44 + 1.78 + 2.04))(10kHz) = 0.134 mW
For α = 1 we get
N = 2N
0
(.25 + .44 + .56 + .64 + .69))(10kHz) = 0.0516 mW
(d) As α increases, the frequency response H
eq
(f) decreases for all f. Thus, the noise power decreases,
but the signal power decreases as well. The factor α 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 α → ∞, the noise power goes to 0 because H
eq
(f) → 0 for all f. However, the signal power
also goes to zero.
14. The equalizer must be retrained because the channel decorrelates. 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
LMSDFE: 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 LMSDFE/RLS, T
b
(LMSDFE) <T
b
(RLS).
But time to convergence is faster for RLS.
Case 1: f
D
= 100 Hz ⇒(∆t
c
) ≡ 10 msec, must retrain every 5 msec.
LMSDFE: R =
10
7
9

1000 bits
5 msecs
= 911 Kbps
RLS: R =
10
7
58

50 bits
5 msec
= 162 Kbps
Case2: f
D
= 1000 Hz ⇒ retrain every 0.5 msec
R
LMSDFE
= 0 bps
R
RLS
= 72.4 Kbps
16. In the adaptive method, we start with some initial value of tap coeﬃcients W
0
and then use the
steepest descent method
W
k+1
= W
K
−∆G
K
. . . (1)
where ∆ is some small positive number and G
K
is the gradient of MSE = E
ˆ
d
k
−
ˆ
ˆ
d
k

2
is RW
k
−p
(Notice that 11.37 was a solution of gradient =0 , ∴ RW = p)
∴ G
K
= RW
k
−p = −E[ε
k
Y
k
]
where Y
k
= [y
k+L
. . . y
K−L
]
T
and ε
k
=
ˆ
ˆ
d
k
−
ˆ
d
k
Approximately (1) can be rewritten as
W
k+1
= W
k
+∆ε
k
Y
k
Chapter 15
1. City has 10 macrocells
each cell has 100 users
∴ total number of users = 1000
Cells are of size 1 sqkm
maximum distance traveled to traverse =
√
2km
∴ time =
√
2
30
= 169.7s
In the new setup
number of cells = 10
5
microcells
total number of users = 1000 ×100
2
users
time =
√
2×10
30×10
3
=1.69s
∴ number of users increases by 10000 and handoﬀ time reduces by 1/100
2. See Fig 1
(0,0)
(0,1)
v
u
(1,1)
(1,0)
60
Figure 1: Problem 2
D
2
= (j20)
2
+ (i20)
2
−2(j20)(i20) cos(2π/3)
⇒ D = 2a
_
i
2
+j
2
+ij =
√
3R
_
i
2
+j
2
+ij
3. diamond shaped cells, R= 100m
D
min
= 600m
D = 2KR
K =
D
2R
=
600
2×100
= 3
N = K
2
= 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
N =
A
cluster
A
cell
=
√
3D
2
/2
3
√
3R
2
/2
=
1
3
(D/R)
2
=
1
3
6
2
= 12
number of cells per cluster = N = 12
D
SHADED CELLS USE
SAME FREQUENCY
Figure 2: Problem 3
(b) number of channels in each cell = 1200/12 = 100
(c)
_
i
2
+j
2
+ij = 2
√
3 ⇒ i = 2, j = 2
5. R=10m
D=60m
γ
I
= 2
γ
0
= 4
M = 4 for diamond shaped cells
SIR
a
=
R
−γ
I
MD
−γ
0
=
R
−2
4D
−4
= 32400
SIR
b
=
R
−4
4D
−4
= 324
SIR
c
=
R
−2
4D
−2
= 9
SIR
a
> SIR
b
> SIR
c
6. γ = 2
BPSK
P
b
= 10
−6
→ P
b
= Q(
√
2γ
b
) ⇒ γ
b
= SIR
0
= 4.7534
B = 50MHz
each user100KHz = B
s
SIR =
1
M
_
D
R
_
γ
M=6 for hexagonal cells
a
1
= 0.167
a
2
= 3
N >
1
a
2
_
SIR
0
a
1
_
2/γ
⇒ N ≥ 9.4879 ∴ N = 10
C
u
= 50
7. G = 100
ξ = 1
λ = 1.5
With no sectorization
SIR =
1
ξ
3G
(N
c
−1)(1 +λ)
= 4.7534
N
c
= 26.2450 = 26
With sectorization, interference is reduced by a factor of 3
N
c
= 76.7349 = 76
8. SINR =
G
N
c
−1
i=1
X
i
+N
α = p(X
i
= 1)
N ∼ G(0.247N
c
, 0.078N
c
)
P
o
ut = p(SIR < SIR
0
)
(a) P
out
= p
_
G
N
c
−1
i=1
X
i
+N
< SIR
0
_
= p
_
N
c
−1
i=1
X
i
+N >
G
SIR
0
_
(b) X =
N
c
−1
i=1
X
i
then X ∼ Bin(α, N
c
−1)
p(x +N > G/SIR
0
) =
N
c
−1
N=0
p(n +N > G/SIR
0
x = n)p(x = n)
p(x = n) =
_
N
c
−1
n
_
α
n
(1 −α)
N
c
−1−n
p(x +N > G/SIR
0
) =
N
c
−1
N=0
p(N > G/SIR
0
−nx = n)p(x = n)
=
N
c
−1
N=0
p
_
N −0.247N
c
√
0.078N
c
>
G
SIR
0
−n −0.247N
c
√
0.078N
c
x = n
_
=
N
c
−1
N=0
Q
_
G
SIR
0
−n −0.247N
c
√
0.078N
c
_
p(x = n)
(c) N
c
= 35
α = 0.5
SIR
0
= 5
G = 150
p = 0.0973
MATLAB
for i = 1:length(n)
pn(i) = (factorial(Nc1)./(factorial(n(i)).*factorial(Nc1n(i))))...
*alpha.^n(i)*(1alpha).^(Nc1n(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((N
c
−1)α, (N
c
−1)α(1 −α))
x +N ∼ G(0.247N
c
+ (N
c
−1)α, 0.078N
c
+ (N
c
−1)α(1 −α))
p(x +N > G/SIR
0
) = Q
_
G
SIR
0
−(0.247N
c
+ (N
c
−1)α)
_
0.078N
c
+ (N
c
−1)α(1 −α)
_
(e) p= 0.0969 (very accurate approximation!)
9. deﬁne
γ
k
=
g
k
P
k
n
k
+ρ
k=j
g
kj
p
j
k, j ∈ {1, . . . K}
where,
g
k
is channel power gain from user k to his base station n
k
is thermal noise power at user k’s base
station
ρ is interference reduction factor (ρ ∼ 1/G)
g
kj
is channel power gain from j
th
interfering transmitter to user k’s base station
p
k
is user k’s Tx power
p
j
is user j’s Tx power
deﬁne a matrix F such that
F
kj
=
_
0 k = j
γ
k
g
kj
ρ
g
k
k = j
k, j ∈ {1, . . . K}
u =
_
γ
1
n
1
g
1
,
γ
2
n
2
g
2
, . . . ,
γ
K
n
K
g
K
_
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
2 3 4 5 6 7 8 9 10
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
D
A
S
E
(
D
)
5 Case ASE
Best Case ASE
Worst Case ASE
Figure 3: Problem 10
11. P
t
= 5W
B = 100KHz
N
0
= 10
−16
W/Hz
P
r
= P
t
K
_
d
0
d
_
3
d
0
= 1, K = 100
(a) D=2R
2 users share the band available
Each user gets 50KHz
BASESTATION
USER
R=1Km
Figure 4: Problem 11a
(b) P
b
=
1
4γ
b
⇒ 10
−3
=
1
4γ
b
⇒ γ
b
= 250
If D(n) = 2nR, number of users that share band = 2(2(n1)+1)
∴ each user gets
100KHz
2(2n−1)
= B
u
(n)
interference is only from ﬁrst tier
SIR(n) =
P
t
K
_
d
0
R
_
3
N
0
2
B
u
(n) + 2
_
P
t
K
_
d
0
√
R
2
+D(n)
2
_
3
_ > 25
using Matlab , n = 4, SIR = 261.9253, D = 8R
Matlab
Pt = 5;
R = 1000;
sigma_2 = 1e16;
n = 1;
D = 2*n*R;
Bu = (100/(2*(2*n1)))*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*n1)))*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
(c) ASE =
(R
1
+R
2
)/B
2km×2km
R
1
= R
2
= B
u
(1) log(1 +SIR(1))
B
u
(1) = 50KHz, SIR(1) = 5.5899
ASE = 0.6801bps/Hz/km
2
12. B=100KHz
N
0
= 10
−9
W/Hz
K = 10
P = 10mW per user
(a) 0 ≤ α ≤ 1 α is channel gain between cells.
See Matlab
If α is large, interference can be decoded and subtracted easily so capacity grows with α as high
SNR’s (beyond an α value) .
For low SNR values (α less than a value) c decreases with increase in α as interference is increased
which cannot be easily decoded due to low SNR.
MATLAB CODE:
B = 100e3;
sigma_2 = 1e9;
P = 10e3;
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
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.82
0.84
0.86
0.88
0.9
0.92
0.94
0.96
0.98
1
1.02
α
C
(
α
)
/
B
B = 100 KHz
Figure 5: Problem 12a
(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 = 1e9;
P = 10e3;
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
0 5 10 15 20 25 30
0
1
2
3
4
5
6
K
C
(
K
)
/
B
B = 100 KHz
Figure 6: Problem 12b
(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 = 1e9;
P = [0:.1:100]*1e3;
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
0 10 20 30 40 50 60 70 80 90 100
0
0.2
0.4
0.6
0.8
1
1.2
1.4
P (in mW)
C
(
P
)
/
B
B = 100 KHz
Figure 7: Problem 12c
7. 8. 1. Handoﬀ becomes a big problem. 2. Intercell 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 Each subscriber generates 30 of an Erlang of traﬃc. Thus, each cell can support 30 × 89 = 2670 subscribers Macrocell: 2670 × 100 ⇒ 267, 000 subscribers Microcell: 6,675,000 subscribers (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
channels cell
⇒ 89
channels cell
@Pb = .02
Chapter 2
1. Pr = Pt 10−3 = Pt 10−3 = Pt √ Gl λ 4πd λ 4π10 λ 4π100
2
λ = c/fc = 0.06
2
⇒ Pt = 4.39KW
2
⇒ Pt = 438.65KW
Attenuation is very high for high frequencies 2. d= 100m ht = 10m hr = 2m delay spread = τ = 3. ∆φ =
2π(x +x−l) λ
x+x −l c
= 1.33×
x +x−l =
(ht + hr )2 + d2 − (ht − hr )2 + d2 ht + hr 2 ht − hr 2 = d +1− + 1 d d ht + hr d
2
d
ht , hr , we need to keep only ﬁrst order terms
1 ∼ d 2 =
ht − hr d
2
1 + 1 − 2
+ 1
∆φ ∼
2(ht + hr ) d 2π 2(ht + hr ) λ d
4. Signal nulls occur when ∆φ = (2n + 1)π 2π(x + x − l) λ (ht + hr )2 + d2 − (ht + hr )2 + d2 − (ht − hr )2 + d2 = (2n + 1)π = π(2n + 1) λ (2n + 1) 2 λ + 2 (ht − hr )2 + d2
2π λ
(ht − hr )2 + d2 =
Let m = (2n + 1) (ht + hr )2 + d2 = m square both sides (ht + hr )2 + d2 = m2 x = (ht + hr )2 , y = (ht − hr )2 , x − y = 4ht hr x = m2 ⇒d = d = λ2 + y + mλ y + d2 4 1 mλ x − m2 λ2 −y 4
2
λ2 + (ht − hr )2 + d2 + mλ (ht − hr )2 + d2 4
−y
2
4ht hr (2n + 1)λ − (2n + 1)λ 4
− (ht − hr )2
,n ∈ Z
5. ht = 20m hr = 3m fc = 2GHz λ = fcc = 0.15 t dc = 4hλhr = 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) = α1 δ(t − τ ) + α2 δ(t − (τ + 0.22µs)) Gr = Gl = 1 ht = hr = 8m fc = 900M Hz, λ = c/fc = 1/3 R = −1 delay spread = ⇒ x+x −l c
d 2 2
= 0.022 × 10−6 s = 0.022 × 10−6 s
2 82 +
−d
c ⇒ d = 16.1m d ∴ τ = = 53.67ns c
and the dominant factor is the phase diﬀerence between the paths. Also note that the the received power actually increases with distance up to some point. R=1.^.counter). clear all. the reﬂected path is approximately two times the LOS path. making the phase diﬀerence very small. Pr=(lambda/4/pi)^2*(abs(Vec)).1. dc=4*ht*hr/lambda. which makes sense since the eﬀect of reﬂected path is reduced and it is more like having only a LOS path.5. Thus. From the plots it can be seen that as Gr (gain of reﬂected path) is decreased. subplot(2.√ λ Gl α1 = 4π l √ λ RGr α2 = 4π x + x 2 = 2. Gr=GR(counter). the lengths of both paths are initially dominated by the diﬀerence between the antenna heights (which is 35 meters).^. the ﬁrst local maxima is achieved. Vec=Gl. d=[1:1:100000]..37 × 10−6 8.20*log10(dnew)). plot(10*log10(dnew).^2.2. Additionally. A program to plot the ﬁgures is shown below. r=(d.. When the phase diﬀerence becomes 180 degrees. plot(10*log10(d).71 × 10−6 2 = 1. Also the variation of power before and around dc is reduced because the strength of the reﬂected path decreases as Gr decreases. Since R = 1. the powers of both paths are roughly constant for small values of d. dnew=[dc:1:100000].^2+(hthr)^2). The power versus distance curves and a plot of the phase diﬀerence between the two paths is shown on the following page. c=3e8./l+R*Gr.*exp(phd*sqrt(1)). lambda=c/f. Gl=1.316. close all.10*log10(Pr)10*log10(Pr(1))). ht=50..01].40*log10(dnew)). for counter = 1:1:4.^2+(ht+hr)^2). hr=15. d = 1). counter=1. phd=2*pi/lambda*(r1). hold on. GR=[1. the asymptotic behavior of Pr tends toward d−2 from d−4 . this causes the two paths to nearly cancel each other out.5. l=(d. f=900e6. This is because for very small distances (i. plot(10*log10(dnew). end hold off .e. figure(1)./r.
S ray d = 500m τ0 = 500/c = 1.01µs 11.316 20 40 G = . s = s = (0.54dB Pref lection = −94.68µs L.Gr = 1 50 10 * log10(Pr) 0 −50 −100 −150 50 10 * log10(Pr) 0 −50 −100 −150 400 Phase (deg) 300 200 100 0 0 20 40 10 * log10(d) 60 0 20 40 10 * log10(d) 60 10 * log10(Pr) 0 20 40 Gr = .5dB d = 100 Pscattering = −96.1 10 * log10(d) 60 10 * log10(Pr) 50 0 −50 −100 −150 50 0 −50 −100 −150 0 0 Gr = .5 Path loss due to scattering √ λ Gσ Pr = Pt (4π)3/2 ss Path loss due to reﬂection (using 2 ray model) √ 2 R G λ Pr = Pt s+s 4π d = 10 Pscattering = −56.5d)2 + (0.5d)2 = d 0.54dB Pref lection = −74.59/c = 1.54dB distances result in tremendous path loss d0 d 2 γ → simpliﬁed λ d 2 → free space .0224 = −16. Pr = Pt K Pr = Pt √ Gl 4π 2 (500/6)2 + 102 = 83.52 × 10−4 = −34.01 10 * rlog10(d) 60 20 40 10 * log10(d) 60 Figure 1: Problem 8 9.498dB 2 = 3. The delay spread is dictated by the ray reaching last d = Total distance = 6d = 503. fc = 900M Hz λ = 1/3m G = 1 radar cross section 20dBm2 = 10 log1 0σ ⇒ σ = 100 √ √ d=1 . As indicated in the text. the power fall oﬀ with distance for the 10ray model is d−2 for relatively large distances 10.5 = 0.59m τ0 = 503.5dB d = 1000 Pscattering = −136.5dB Notice that scattered rays over long 12.O.67µs ∴ delay spread = 0.93m = 0.54dB Pref lection = −54.
7 × 10−4 .G ∴ when K = 4π l and d0 = λ The two models are equal. d = distance between cells with reused freq p = transmit power of all the mobiles S I ≥ 20dB uplink (a) Min.04 ⇒ dmin = 2. hr = 5m. ht = 20m. path loss is higher in the presence of multiple reﬂectors. fc = 900M Hz. K = (λ/4πd0 )2 = 5.7m 14. 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 S I = min P 2P Gλ 4πdA Gλ 4πdB 2 2 = d2 (dmin − 1)2 B = 100 = 4 2d2 A ⇒ dmin − 1 = 20km ⇒ dmin = 21km since integer number of cells should be accommodated in distance d ⇒ dmin = 22km (b) Pγ d0 =k Pu d 1 dB 2 dA γ γ ⇒ S I = min Pk 2P k d0 dA d0 dB γ γ = ⇒ dmin (c) 1 dmin − 1 γ 1 dmin − 1 √ √ = = 2 2 2 2 = 9.3 d 4 d ≤ 260. √ 2 13.52dB P Lsmallcity = 325.3.8769dB P Lruralarea/countryside = 70.99dB P Lsuburb = 207. Pnoise = −160dBm fc = 1GHz.27 ⇒ with the same argument ⇒ dmin = 10km S I k = min γ d0 dA A γ d0 dB B 3 = 100 = 2k (dmin − 1)4 = 100 0. d = 100m Large urban city small urban city suburb rural area P Llargecity = 353. λ = 0. diﬀractors and scatterers . γ = 4 We want SN Rrecd = 20dB = 100 Noise power is 10−19 d0 γ P = Pt K d 10−17 = 10K 0.41km ⇒ with the same argument dmin = 4km 15. d0 = 1m.9278dB As seen .
2 1.512 × 10−4 .2dBm 18. Pr = Pt − PL (d) − 3 F AFi − 2 P AFj i j FAF =(5.6).2 2.4. Assume free space path loss parameters fc = 900M Hz → λ = 1/3m σψdB = 6 SN Rrecd = 15dB Pt = 1W X σψdB < −10 σψdB = 6.Plot for Gr = 0 dB 0 −10 y = −15x+8 −20 −30 −40 y = −35x+56 −50 −60 1 1.76dB 2 (c) Receiver power can be assumed to be Gaussian with variance σψdB 2 X ∼ N (0. Piecewise linear model for 2path model.4 − 3.3. PAF =(3.4) PL (d)K d0 d γ 0 = 10−8 = −8dB −110 = Pt − 80 − 5 − 10 − 6 − 3. (a) d = 10 log10 K − 10r log1 0 d0 using least squares we get 10 log10 K = −29.4 2.42dB γ=4 Pr Pt dB (b) PL(2Km) = 10 log10 K − 10r log10 d = −161.4 ⇒ Pt = −2.10.6 2.8 2 2. See Fig 2 17.4 1.8 3 Figure 2: Problem 16 16.6 1. σψdB ) P rob(X < −10) = P rob 19.
01.9 P Precd (d) − µdB −55 − µdB > = 0.9 σψdB 6 −55 − µdB ⇒ = −1. T p − µψ < −6.68m 20. [received powerdB Tp = 10dB (a) outageprob. MATLAB CODE Xc = 20.308 ⇒ µ(d) = 1.282 6 ⇒ µdB = −47.32dB (c) σψ = 12dB. = Prob.8dB σψ . y = wgn(1. Outage Prob.200*(1/ss)). for j = 1:i x(i) = x(i)+exp((ij)/Xc)*y(j). end end 21.86 × 10−5 ⇒ d = 8.99 ⇒ µψ ≥ 37. outage prob < 1% ⇒ Q T p − µψ σψ > 99% ⇒ T p − µψ < −2.33 ⇒ σψ T p − µψ σψ =1−Q −5 8 =Q 5 8 = 26% T pdB ] µψ ≥ 19. for i = 1:length(y) x(i) = y(i).4 × 1−−3 = Pt = 4πd d2 = 10 log1 0(µ(d)) µdB P (Precd (d) > −55) = 0.g = 3dB Pnoise = −40dBm ⇒ Precvd = −55dB Suppose we choose a cell of radius d µ(d) = Precvd (due to path loss alone) √ 2 Gl λ 1. = 1 − Q (b) σψ = 4dB. ss = .
∂r R ∂x ∂r R r 2π (1) . 22. 2 C= 2 R R rQ a + b ln r=0 r dr R r To perform integration by parts. if they are uncorrelated. we let du = rdr and v = Q a + b ln R . 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. we can use macroscopic diversity.20 White Noise Process Filtered Shadowing Processing 10 0 −10 −20 −30 −40 −50 −60 0 20 40 60 80 100 120 140 160 180 200 Figure 3: Problem 20 (d) For mitigating the eﬀect of shadowing. The idea in macroscopic diversity is to send the message from diﬀerent base stations to achieve uncorrelated shadowing. Then u = 1 r2 and 2 dv = ∂ r ∂ ∂ r −1 b Q a + b ln = Q(x)x=a+b ln(r/R) a + b ln = √ exp(−k 2 /2) dr.
97dB Pmin − Pγ (R) = 14.3871 b= 8 c = 0. γ = 6 σ=8 Q 2 − 2ab b 1 Pγ (R) = 20 + Pmin a = −20/8 = −2. Then we get C = 2 1 2 r r Q a + b ln 2 2 R R 1 R2 1 R2 R R 2 + 2 R r=0 R r=0 1 2 1 b r √ exp(−k 2 /2) dr 2 r 2π (2) = Q(a) + r=0 R 1 b r2 √ exp(−k 2 /2) dr r 2π 1 √ R2 exp 2π 2(k − a) b b exp(−k 2 /2) dr r dk (3) = Q(a) + (4) r=0 a = Q(a) + k=−∞ 1 √ exp 2π −k 2 2k 2a + − 2 b b a (5) = Q(a) + exp −2a 2 + 2 b b 2 − 2ab b2 2 − 2ab b2 k=−∞ 1 1 2 √ exp − (k − )2 dk 2 b 2π a− 1 2 b (6) = Q(a) + exp = Q(a) + exp 1−Q Q (7) (8) (9) 2 − ab b Since Q(−x) = 1 − Q(x).9938 .5 10 × 6 × log10 e = 20.2572 σψdB 2−2ab b2 c = Q(a) + e 24. 23.r where k = a + b ln R . γ = 3 d0 = 1 k = 0dB σ = 4dB R = 100m Pt = 80mW Pmin = −100dBm = −130dB Pγ (R) = Pt K a= d0 d γ = 80 × 10−9 = −70.7575 σ 10γ log10 e b= = 3.
Due to these forces. we have maximum coverage when γ = 6 and σ = 4.7728 Since Pr (r) ≥ Pmin for all r ≤ R.8 0.6302 4 0. and thus σ decreases as a function of σ. γ = 4 and σ = 8 can be seen from the table or the ﬁgure to be 0.7170 6 0. Therefore. Since the average power at the boundary of the cell is ﬁxed.85 Coverage 0.6786 0. the probability of nonoutage is proportional to Q −1 . C decreases as a function of σ.25. .6 12 10 8 4 6 σ ψ 6 5 3 4 2 γ dB Figure 4: Problem 25 The value of coverage for middle point of typical values i.7 0.7728 0.9 0. The same can also be seen from this ﬁgure: 0.7728. γ/σψdB 4 8 12 2 0. hence more received power at r < R. we have minimum coverage when γ = 2 and σ = 12.75 0. C increases with γ.e.65 0.7728 0. By a similar argument.8255 0. because it forces higher transmit power.8977 0.95 0.8587 0.
Delay for LOS component = τ0 = 23 ns Delay for First Multipath component = τ1 = 48 ns Delay for Second Multipath component = τ2 = 67 ns τc = Delay for the multipath component to which the demodulator synchronizes. . hr = 4m. Tm = 19 ns. d = 100m ∴ Tm = 2. When τc = τ1 . Tm = 44 ns. For the 2 ray model: d ) l c x+x τ1 = c τ0 = x+x −l = c Tm = (ht + hr )2 + d2 − c (ht − hr )2 + d2 ∴ delay spread(Tm ) = when d (ht + hr ) 1 2ht hr c d ht = 10m. Tm = max τm − τc m So. d = vt 2 r + r = d + 2h d Equivalent lowpass channel impulse response is given by c(τ.67 × 10−9 s 3.Chapter 3 1. when τc = τ0 . t) = α0 (t)e−jφ0 (t) δ(τ − τ0 (t)) + α1 (t)e−jφ1 (t) δ(τ − τ1 (t)) G α0 (t) = λ4πd l with d = vt φ0 (t) = 2πfc τ0 (t) − φD0 τ0 (t) = d/c φD0 = t 2πfD0 (t)dt v fD0 (t) = λ cos θ0 (t) θ0 (t) = 0 ∀t√ √ λR G G α1 (t) = 4π(r+r l) = λR 2hl2 with d = vt 4π(d+ √ φ1 (t) = 2πfc τ1 (t) − φD1 2 τ1 (t) = (r + r )/c = (d + 2h )/c d φD1 = t 2πfD1 (t)dt v fD1 (t) = λ cos θ1 (t) h θ1 (t) = π − arctan d/2 ∀t 2.
Sample Code is given: clear P0 = 1e11.x). int_pr = sum(pdf)*ss.z.*besseli(0. ss = z0/1e7. . we use Matlab and I0 (x) = besseli(0. z = [0:ss:z0].0952 7.4.*(sqrt(s2)/sigma2)). sigma2 = (1e11)/2.*exp((z. s2 = 1e11.0311 2σ 2 = −80dBm. P0 = −95dBm.^2+s2)/(2*sigma2)). power in LOS component = s2 = 80dBm = 10−11 2 10−5 P r[z 2 ≤ 10−11 ] = P r[z ≤ √ ] 10 Let z0 = 10−5 √ 10 z0 −(z 2 +s2 ) 2σ 2 = 0 z e σ2 − I0 zs dz. fc = 109 Hz 10 τn. P0 = −90dBm. z0 = sqrt(1e11).min = 3×108 s ∴ min fc τn = 1010 3×108 = 33 1 5. For Rayleigh fading channel Poutage = 1 − e−P0 /2σ 0. Use CDF strategy. 2σ 2 = 80dBm = 10−11 Target Power P0 = 80 dBm = 10−11 Avg. P r(Pr < P0 ) = 0. pdf = (z/sigma2).01 = 1 − e−P0 /Pr ∴ Pr = −60dBm 8.3457 To evaluate this. Fz (z)= P [x2 +y 2 ≤ z 2 ] = x2 +y 2 ≤z 2 −(x2 +y)2 1 e 2σ2 dxdy = 2πσ 2 2π z 0 0 −z 2 −r2 1 e 2 rdrdθ = 1 − e 2σ2 (z ≥ 0) 2πσ 2 2σ df z (z) z − z22 = 2 e 2σ → Rayleigh dz σ For Power: √ −z 2 z] = 1 − e 2 2σ 1 −z fz (z)= 2 e 2 → Exponential 2σ 2σ Fz 2 (z)=P [Z ≤ P r(Pr < P0 ) = 1 − e−P0 /2σ 2 6. For Rayleigh fading channel 2σ 2 = −80dBm. σ2 z≥0 = 0. P r(Pr < P0 ) = 0.
cdfR). plot(z.2 < T Y = y] fy (y) dy Pr. where the tail of Nakagamidistribution is always above the Ricean distribution.5 We need to plot the two CDF curves for K = 1. figure.*besseli(0.01*sum(pdfR(1:i)). which is Gaussian with mean W.2 < T ] since Z1 . end plot(z. hold on plot(z. Also.a2 σ 2 . prob(γ<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. m ≥ 0. we set the Nakagami m parameter to be (K + 1)2 /(2K + 1).’b’).^(2*m1))/(gamma(m)*Pr^m)). pdfN = ((2*m^m*z. and [Pr. CDF of Nakagamim distribution is Nakagamim FZ (z) = z 0 pNakagamim (z) Z where pNakagamim (z) = Z 2mm z 2m−1 −mz 2 exp . Pr = 1.01:3].’b’).5. pdfR = ((2*z*(K+1))/Pr).pdfN.2 < T ] = P [Pr. Γ(m)P rm Pr z ≥ 0.*exp(K((K+1). the Nakagamim approximation becomes better as K increases. hold on plot(z.2 Y = y] ∞ Poutage= −∞ Q W + by − T aσ 2 y2 1 √ e− 2σ2 dy 2πσ .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). Z2 independent = Q (c) Pout = −∞ ∞ W −T σ 2 2 = Q σ P [Pr. for a ﬁxed value of K and x. CDF of Ricean distribution is Ricean FZ (z) = 0 z pRicean (z) Z K(K + 1) Pr where pRicean (z) = Z 2z(K + 1) (K + 1)z 2 exp −K − I0 Pr Pr 2z .1 ≤ T . cdfN(i) = 0.1 Y = y ~ N W + by. variance σ 2 (b) Poutage = P [Pr.*exp((m/Pr)*z.9. (a) W = average received power Zi = Shadowing over link i Pr i = Received power in dBW. K = 10.1 < T ] P [Pr. Pr.^2))/Pr).*(z.1 Y = y] ⊥ [Pr. z≥0 For the Nakagamim approximation to Ricean distribution.01*sum(pdfN(1:i)).(2*sqrt((K*(K+1))/Pr))*z). for i = 1:length(z) cdfR(i) = 0. Sample code is given: z = [0:0.cdfN. 10. m = (K+1)^2/(2*K+1).1 < T ∩ Pr.pdfR).^2). As seen from the curves.
8 0.8 0.5 1 1.2 0.1 0.8 0.8 0.5 6 Figure 1: The CDF and PDF for K = 1.9 0.2 1 0.1 0.6 1.5 2 2.6 0.5 1 1.4 0.6 0.3 0.3 0.5 0.9 1 K = 10 Ricean Nakagami−m 0.7 0.5 3 0 0 0.4 Ricean Nakagami−m 1.5 3 0 0 0.5 1 1.5 4 4.7 1. 5.5 1 0.5 2 2.2 0. 10 and the Tail Behavior .5 1 1.5 3 0 0 0.3 0.8 1.2 0 3.5 3 K=1 1 1.5 0.4 0.3 0.6 0.5 K=5 2 K = 10 Ricean Nakagami−m 0.5 Tail Behavior 5 5.9 Ricean Nakagami−m Ricean Nakagami − m 1.6 0.6 0.4 0.5 2 2.4 1 1.4 0.4 0.5 2 2.2 0.2 0.7 0.5 0.5 1 1.8 0.5 2 2.1 0 0 0.4 0.1 0.8 0.6 0.7 Ricean Nakagami −m 0.5 3 2 x 10 −31 K = 10 1.5 0.8 1.9 0.6 0.5 1 1.5 3 0 0 0.2 0.K=5 1 1 Ricean Nakagami−m 0.5 2 2.2 0 0 0.
1.i). rqtemp(M. beta(M)=0.1316. Wm=2*pi*fm. Sample code for both the stochastic technique(preferred) and the Jake’s technique are included. %beta(i)=pi*i/M beta(n)=pi*n/M.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. end end %summarize ritemp(i) and rialpha ri=sum(ritemp)+rialpha. subplot(3.5*(N/21). %ritemp(i. %choose N=30 N=30. With correlated fading we get Pout = 0. = 5. clear all. With independent fading we get Pout = Q 5 8 2 = 0.let y σ= u ∞ = −∞ 1 √ Q 2π 1 √ 2 W − T + buσ aσ 2 e −u2 2 ∞ du = −∞ 1 √ Q 2π + byσ aσ 2 e −y 2 dy 2 (d) Let a = b = . %We choose 1000 samples/sec ritemp(M. Jakes: Summing of appropriate sine waves %Jake’s Method close all. σ = 8. fm=[1 10 100].1000*t+1)=2*sin(beta(n))*cos(Wn(n)*t). rialpha(1. Conclusion : Independent shadowing is prefarable for diversity.001:2 %Wn(i)=Wm*cos(2*pi*i/N) Wn(n)=Wm(i)*cos(2*pi*n/N).2001)=0. %find the envelope average mean=sum(r)/2001.0708. rqtemp(n. . %Because we choose alpha=0.2001)=0. Wn(M)=0. %summarize rqtemp(i) rq=sum(rqtemp).2001)=0. There are many acceptable techniques for this problem.20001)=2*cos(beta(i))*cos(Wn(i)*t) %rqtemp(i.1000*t+1)=2*cos(Wm(i)*t)/sqrt(2). M=0. %r=sqrt(ri^2+rq^2) r=sqrt(ri.^2+rq.20001)=2*sin(beta(i))*cos(Wn(i)*t) ritemp(n.^2). for i=1:3 for n=1:1:M for t=0:0.1000*t+1)=2*cos(beta(n))*cos(Wn(n)*t). 11.
ylabel(’Envelope(dB)’). %plot the figure and shift the envelope average to 0dB plot(time.001:2. titlename=[’fd = ’ int2str(fm(i)) ’ Hz’]. end . xlabel(’time(second)’). title(titlename).time=0:0.(10*log10(r)10*log10(mean))).
4 0.4 1. function [Ts.4 1.8 1 1.4 0. with sampling frequency f_s >= 1000Hz.t] % .z_dB : [dB] [1xN double] Rayleigh fading signal % % Required parameters if f_s < 1000.2 0.8 2 0 0.6 0.fd = 1 Hz 5 Envelope(dB) 0 −5 −10 10 Envelope(dB) 0 −10 −20 10 Envelope(dB) 0 −10 −20 0 0.2 1. z_dB] = rayleigh_fading(f_D.’s are ﬁltered by the Doppler Spectrum and summed.6 0. % % Input(s) % . t]. % [Hz] Minumum required sampling rate . set to 1000 if smaller.2 0.6 0.6 1. % during the time perios [0.2 1.8 1 fd = 100 Hz 1.4 0. t.6 1. end. z_dB] = rayleigh_fading(f_D.2 0. % Output(s) % .t : simulation time interval length.2 1.f_s : [Hz] sampling frequency.8 1 fd = 10 Hz 1.6 1.f_D : [Hz] [1x1 double] Doppler frequency % . f_s = 1000. f_s) % generates a Rayleigh fading signal for given Doppler frequency f_D.4 1.Ts : [Sec][1xN double] time instances for the Rayleigh signal % .V. Can also just do a Rayleigh distribution with an adequate LPF. although the above technique is prefered. f_s) % % function [Ts.8 2 Figure 2: Problem 11 Stochastic: Usually two guassian R. t. time interval [0.8 2 0 0.
% Finally. generate the Rayleigh distributed signal envelope z = sqrt(abs(r_I). % Number of samples % Ts contains the time instances at which z_dB is specified Ts = linspace(0. z_dB = 20*log10(z). CGfi_p = Gfi_p(1. % Return correct number of points z_dB = z_dB(1:length(Ts)).N). let them be 0 idx1 = find(f>=f_D). if mod( N.N/2). this is used to shape the Gaussian line spectra omega_p = 1. idx2 = find(f<=f_D).:)+i*Gfi_p(2.t.*sqrt(S_r)). % 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. CGfq_p = Gfq_p(1. S_r(idx1(1)) = S_r(idx1(1)1).f_s. 2) == 1 N = N+1.*sqrt(S_r)). % Generate r_I(t) and r_Q(t) using inverse FFT: r_I = N*ifft(CGfi.^2)). S_r(idx2) = 0. % Take care of samples outside the Doppler frequency range. % Quadrature : Gfq_p = randn(2.N). CGfq = [ conj(fliplr(CGfq_p)) CGfq_p ]. end.^2+abs(r_Q).:). r_Q = i*N*ifft(CGfq. % this makes sure that the average received envelop can be 0dB S_r = omega_p/4/pi. . S_r(idx2(length(idx2))) = S_r(idx2(length(idx2))+1).^2).:). f = linspace(f_s. S_r(idx1) = 0.N/2).:)+i*Gfq_p(2./(f_D*sqrt(1(f/f_D). % Generate fading spectrum. CGfi = [ conj(fliplr(CGfi_p)) CGfi_p ].N = ceil( t*f_s ).
tz = eρ − 1 √ = 0. So we have √ 2 LZ (above) = LZ (below) = 2πfD ρe−ρ And.8 1 10Hz 1.6 1. tz = 0.8 2 0 0.2 1.8 1 l00 Hz 1.0013s ρfD 2π 2 P0 = 15dBm.8 2 Figure 3: Problem 11 12.6 0.2 0.4 1.4 0.4 1. Pr = 30dBm fD = 10Hz P0 = 0dBm.8 2 0 0. tz = 0.4 0.1 Hz 0 −10 −20 −30 10 0 −10 −20 −30 20 0 −20 −40 −60 0 0.2 0.2 0. 3. 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.0072s P0 = 25dBm. In the reader.6 1. tZ (above) = p(z > Z) LZ (above) 2 2 p(z > Z) = 1 − p(z ≤ Z) = 1 − (1 − e−ρ ) = e−ρ 1 tZ (above) = √ 2πfD ρ .0264s 13. we ﬁrst need to go above it.2 1.6 0.8 1 1. but for a large enough T it will not eﬀect the result.44 in reader because to go below a level again.4 1. It is easy to convince that the level crossing rate above a level will have the same expression as eq.4 0.6 0.6 1. There will be some discrepancy at the end points.2 1.
1) = (N(i)*T)/Pi(i).19.1) = 0. ρ = N6 = 15.72 × 10−44 Nj → level crossing rate at level Aj N1 = 0. ρ = N4 = 82.0045s.085 .1) = (N(i)*T)/Pi(i).77.72e44].176 . : −10dB ≤ γ ≤ 0dB. for i = 1:8 if i == 1 p(i.09. T = 10e3.03.38 82.361 .085 = 0. Note: The problem has been solved for Ts = 10µs Pr = 10dB fD = 80Hz R1 R2 R3 R4 R5 R6 R7 R8 : −∞ ≤ γ ≤ −10dB.85 57. .325 = 0. ρ = N3 = 57. Notice that as fD increases.1)+p(i. else p(i.325 . : 30dB ≤ γ ≤ ∞.03 0]. p(i.3) = 1(p(i. N8 = 0.361 = 0. p(i.2) = (N(i+1)*T)/Pi(i).95e3 .38. elseif i == 8 p(i. tZ (above) reduces. : 20dB ≤ γ ≤ 30dB. : 10dB ≤ γ ≤ 15dB.85. 0.80 Hz are 0.1)+p(i. ρ = N7 = 0.1 10 1 10 100.5 10 102 10 103 10 N2 = 19.176 = 0. ρ = N5 = 73. : 15dB ≤ γ ≤ 20dB.54 × 10−5 = 3.5 10 10 10 101.54e5 3. p(i.2)).09 .0224s. p(i.19 73. Pi = [9. : 5dB ≤ γ ≤ 10dB.042 = 4.2)). 14. ρ= ρ= MATLAB CODE: N = [0 19.The values of tZ (above) for fD = 10.50.3) = 1(p(i.95 × 10−3 = 0.042 4.0028s respectively.2) = (N(i+1)*T)/Pi(i). 0. ρ= 0 10 0.77 15.1)+p(i.3) = 1(p(i. π1 π2 π3 π4 π5 π6 π7 π8 = 9. : 0dB ≤ γ ≤ 5dB.2) = 0.2)). p(i. p(i.
(a) Outdoor.0023 % 0. v = fD λ/ cos(θ).9934 1.9957 0.0068 0.022µsec) ρ = 49. Consider that 10 µsec ⇒ d = ct = 3km diﬀerence between length of ﬁrst and last path (b) Scattering function S(τ. 16 . (a) 0. 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. ρ) = 0 else The antenna setup is shown in Fig. we have ﬂat fading.0033 % 0.9909 0. Since d 4ht hr λ dc . the distance travelled by the LOS ray is d and the distance travelled by the ﬁrst multipath component is 2 d 2 2 + 64 Given this setup. θ = 45o .0020 0.33µs. ρ) = F∆t [Ac (τ. 15 So we have d 2 + 82 − d = 0.62 + d2 + 2d(6.0000 ρ = 70Hz α1 δ(τ ) α2 δ(τ − 0. ∆t)] = 1 W rect 1 Wρ for 0 ≤ τ ≤ 10µsec The Scattering function is plotted in Fig.33m/s.0036 % 0. Since Tm 16.end end % p = % % 0 % 0.0199 0. (c) Tm = 0.0005 0.0047 0. power falloﬀ is proportional to d−2 . We can use either of these rays and the corresponding fD value to get v = 23.022 × 10−6 3 × 108 2 2 ⇒4 d2 + 82 4 = 6. θ = 0 and for the multipath component.0023 % 0. B −1 .022µs. (b) dc = dc = 768 m.6) ⇒ d = 16.9801 0.9964 0.5Hz S(τ. B −1 = 0. since delay spread ≈ 10 µsec.1m fD = v cos(θ)/λ.0000 0 0 0. 15 From the ﬁgure.0066 % 0 15. For the LOS ray.0023 % 0.9973 0.9921 0.
since receiver power is evenly distributed relative to delay.94 symbols/sec R 17. that based on the choice of either Tm . Notice. Selective .1msec = 100µsec Bd ≈ .89µsec Ac (τ )dτ 0 = 50 Hz 1 Tm (d) βu > Coherence BW ⇒ Freq. (b) Bc ≈ T1 = 10kHz m ∆f > 10kHz for u1 ⊥ u2 (c) (∆t)c = 10s (d) 3kHz < Bc ⇒ Flat 30kHz > Bc ⇒ Freq.1Hz Answers based on µTm or σTm are ﬁne too. no dominant LOS path (f) tR = eρ √ −1 ρFD 2π 2 with ρ = 1. (a) Tm ≈ . in the worst case two symbols will span the fade duration. if we choose our data rate such that a single symbol spans the average fade duration. So our code can correct for the lost symbols and we will have errorfree transmission.8m 8m d meters Figure 4: Antenna Setting 0. So t1 = 72. fD = W 2 → tr = . µTm or σTm .022 us 0 t0 t0 = (d/3e8) t1 t t1 = 2 sqrt[(d/2)^2+8^2]/3e8 Figure 5: Time Axis for Ray Arrival ∞ τ Ac (τ )dτ (c) Avg Delay Spread = 0 ∞ = 5µsec Ac (τ )dτ ∞ 0 (τ −µT m )2 Ac (τ )dτ ∞ RMS Delay Spread = Doppler Spread = W 2 0 = 2. Selective Fading ≈ Can also use µT m or σT m instead of Tm = 105 ⇒ βu > = 105 kHz (e) Rayleigh fading. the remaining answers will be diﬀerent too. So.0137 sec (g) Notice that the fade duration never becomes more than twice the average.
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 0
S N0 B
As B → ∞ by L’Hospital’s rule C= 2. B = 50 MHz P = 10 mW N0 = 2 ×10−9 W/Hz N = N0 B C = 6.87 Mbps.
S 1 N0 ln 2
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) Cuser1→base station = 0.933B = 18.66M bps (b) Cuser2→base station = 3.46B = 69.2M bps 4. (a) Ergodic Capacity (with Rcvr CSI only)= B[
6 i=1 log2 (1
+ γi )p(γi )] = 2.8831×B = 57.66 Mbps.
(b) pout = P r(γ < γmin ) Co = (1pout )Blog2 (1 + γmin ) For γmin > 20dB, pout = 1, Co = 0 15dB < γmin < 20dB, pout = .9, Co = 0.1Blog2 (1 + γmin ), max Co at γmin ≈ 20dB. 10dB < γmin < 15dB, pout = .75, Co = 0.25Blog2 (1 + γmin ), max Co at γmin ≈ 15dB. 5dB < γmin < 10dB, pout = .5, Co = 0.5Blog2 (1 + γmin ), max Co at γmin ≈ 10dB. 0dB < γmin < 5dB, pout = .35, Co = 0.65Blog2 (1 + γmin ), max Co at γmin ≈ 5dB. −5dB < γmin < 0dB, pout = .1, Co = 0.9Blog2 (1 + γmin ), max Co at γmin ≈ 0dB. γmin < −5dB, pout = 0, Co = Blog2 (1 + γmin ), max Co at γmin ≈ 5dB. Plot is shown in Fig. 1. Maximum at γmin = 10dB, pout =0.5 and Co = 34.59 Mbps. 5. (a) We suppose that all channel states are used 1 =1+ γ0
4 i=1
1 pi ⇒ γ0 = 0.8109 γi
1 1 − > 0 ∴ true γ0 γ4 S(γi ) 1 1 = − γ0 γi S
3.5
x 10
7
3
2.5
Capacity (bps)
2
1.5
1
0.5
0
0
0.2
0.4 Pout
0.6
0.8
1
Figure 1: Capacity vs Pout 1.2322 S(γ) 1.2232 = 1.1332 S 0.2332 C = B (b)
1 σ = E[1/γ] S(γi ) σ = γi S 4
γ γ γ γ
= γ1 = γ2 = γ3 = γ4
log2
i=1
γi γ0
p(γi ) = 5.2853bps/Hz
= 4.2882 0.0043 S(γ) 0.0029 = 0.4288 S 4.2882 γ γ γ γ = γ1 = γ2 = γ3 = γ4
C = log2 (1 + σ) = 2.4028bps/Hz B (c) To have pout = 0.1 or 0.01 we will have to use all the subchannels 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 zerooutage case in part b . To have pout that maximizes C with truncated channel inversion, we get max 6. (a) SN Rrecvd Assume all channel states are used 1 =1+ γ0
4 i=1
C = 4.1462bps/Hz B
pout = 0.5
10dB Pγ (d) 5dB = = 0dB Pnoise −10dB
w.p. w.p. w.p. w.p.
0.4 0.3 0.2 0.1
1 pi ⇒ γ0 = 0.4283 > 0.1 γi
∴ not possible
Now assume only the best 3 channel states are used 0.9 =1+ γ0
3 i=1
1 pi ⇒ γ0 = 0.6742 < 1 γi γ γ γ γ = γ1 = γ2 = γ3 = γ4 = 10 = 3.1623 =1 = 0.1
∴ ok!
1.3832 S(γ) 1.1670 = 0.4832 S 0
C/B = 2.3389bps/Hz (b) σ = 0.7491 C/B = log2 (1 + σ) = 0.8066bps/Hz (c)
C B max = 2.1510bps/Hz obtained With pout = 0.1 + 0.2 = 0.3
by using the best 2 channel states.
7. (a) Maximize capacity given by C= max
S(γ): S(γ)p(γ)dγ=S γ
B log 1 +
S(γ)γ S
p(γ)dγ.
Construct the Lagrangian function L=
γ
B log 1 +
S(γ)γ S
p(γ)dγ − λ
S(γ) p(γ)dγ S
Taking derivative with respect to S(γ), (refer to discussion section notes) and setting it to zero, we obtain, 1 1 S(γ) γ0 − γ γ ≥ γ0 = 0 γ < γ0 S Now, the threshold value must satisfy
∞ γ0
1 1 − γ0 γ
p(γ)dγ = 1
Evaluating this with p(γ) =
1 −γ/10 , 10 e
we have
∞ γ0 ∞
γ0 10
1 = = =
1 10γ0
e−γ/10 dγ −
1 10
∞ γ0
e−γ/10 dγ γ
(1) (2) (3)
1 −γ0 /10 1 e − γ0 10
e−γ dγ γ
1 −γ0 /10 1 e − EXPINT(γ0 /10) γ0 10
where EXPINT is as deﬁned in matlab. This gives γ0 = 0.7676. The power adaptation becomes S(γ) = S
1 0.7676
−
1 γ
0
γ ≥ 0.7676 γ < 0.7676
0649 nats/sec/Hz. So for perfect channel knowledge at transmitter and receiver we compute γ0 = 0.75 + log 1 + S2 No B + I 0. (c) AWGN capacity C/B = log(1 + 10) = 2. C/B = log(1 + 0.0150 nats/sec/Hz. Note that I computed all capacites in nats/sec/Hz. Capacity with AWGN is always greater than or equal to the capacity when only the reciever knows the channel. In order to get the capacity values in bits/sec/Hz. However the capacity when the transmitter knows the channel as well and can adapt its power. C = B log 1 + S N0 B + I = 10. This gives S1 = 12.05 is computed as 1 10 ∞ γ0 e−γ/10 dγ = 0. (a) If neither transmitter nor receiver knows when the interferer is on.25S2 = S. (e) Capacity using channel inversion is ZERO because the channel can not be inverted with ﬁnite average power. This can be shown using Jensen’s inequality.3979 nats/sec/Hz.95 (4) (5) (6) 1 1 10 EXPINT(γ0 /10) nats/s/Hz.7676) e−γ/10 dγ = 2.25 . With channel known only to the receiver C/B = 0.3162) = 0. This gives us the capacity with truncated channel inversion as C/B = log 1 + = log 1 + = 1. they must transmit assuming worst case.95 ∗ 0. as if the interferer was on all the time.25mW and C = 53. (d) Capacity when only receiver knows γ C/B = 1 10 ∞ 0 log (1 + γ) e−γ/10 dγ = 2. (f) Channel Mean=5 dB = 0.21Kbps.5463 1 1 10 ∞ 1 −γ/10 dγ γ0 γ e ∗ 0.5129. i. C = B max log 1 + subject to 0. the capacity numbers simply need to be divided by natural log of 2. Threshold for outage probability 0.7Kbps.22765 which gives capacity C/B = 0. At low SNR.36 nats/sec/Hz.7676 log (γ/0. This is because I took the natural log. 8. S2 = 3. can be higher than AWGN capacity specially at low SNR. S1 No B 0. (b) Suppose we transmit at power S1 when jammer is oﬀ and S2 when jammer is oﬀ.95 which gives γ0 = 0.25mW . With AWGN.(b) Capacity can be computed as C/B = 1 10 ∞ 0.2748 nats/sec/Hz.e.3162.2510 nats/sec/Hz. the knowledge of fading helps to use the low SNR more eﬃciently.75S1 + 0.
Pt In the ﬁrst strategy C/B = log2 1 + Pint +Pnoise t In the second strategy C/B = log2 1 + PP−Pint noise Assuming that the transmitter transmits x[k] added to its message always. Next we assume that 0.25. our assumption was wrong again.0625 < γ0 < .48 > 0. we can ﬁnd capacity as: C= j:γj ≥γ0 B log2 γj γ0 This gives us. Now we assume that 0.25 4 ⇒ γ0 = . then 2 =1+ γ0 1 1 + 1 4 ⇒ γ0 = .25 So.8889 < 1 This time our assumption was right.(c) The jammer should transmit −x(t) to completely cancel oﬀ the signal. Suppose transmit power is Pt .4 Mbps.25 < γ0 < 1.0625. then 3 =1+ γ0 1 1 1 + + 1 . γ4 = 0. 9. .25 4 .25. our assumption was wrong.0625 So. and so strategy 2 is usually better. then we have 4 =1+ γ0 1 1 1 1 + + + 1 . power remaining for actual messages is Pt − Pint The ﬁrst or second strategy may be better depending on Pt Pint + Pnoise Pt − Pint ⇒ Pt − Pint − Pnoise Pnoise 0 Pt is generally greater than Pint + Pnoise . S = 10mW N0 = .0625 ⇒ γ0 = .0625 To compute γ0 we use the power constraint as: 1 1 − γ0 γj =1 + j First assume that γ0 < 0. Now.001×10−6 10×106 Hj 2 S N0 B = 1. γ2 = . γ3 = 4. So we get that only two subbands each of bandwidth 10 MHz are used for transmission and the remaining two with lesser SNR’s are left unused.1798 > 0.001 µW/Hz B = 10 MHz Now we compute the SNR’s as: γj = This gives: γ1 = 12 10−3 0. interference power is Pint and noise power is Pnoise . C = 23.
25. We show this for the case of a discrete fading distribution C = Σ log 1 + (1 + N0 B j)2 P (1 + j)2 Pj N0 B j Pj − P j L= i log 1 + − dj ∂L ∂Pj (1 + j)2 Pj ⇒ N0 B (1 + j)2 P letγj = N0 B Pj ⇒ P 1 1 denote = γ0 λP Pj ∴ P subject to the constraint ΣPj P 11.0625 To compute γ0 we use the power constraint as: 1 1 − γ0 γj =1 + j . γ4 = 0.001×10−6 10×106 = 0 = 1 (1 + j)2 −1 λ N0 B = 1 1 − λP γj = 1 1 − γ0 γj = 1 Hj 2 S N0 B = 1. γ2 = .001 µW/Hz B = 10 MHz Now we compute the SNR’s as: γj = This gives: γ1 = 12 10−3 0. S = 10mW N0 = .H(f) 2 1 0. γ3 = 4.25 fc20 fc10 fc fc+10 fc+20 f (in MHz) Figure 2: Problem 11 10.5 0.
0625 So. 12. then we have 4 =1+ γ0 1 1 1 1 + + + 1 .25 < γ0 < 1.4 Mbps.25 4 ⇒ γ0 = .0625 ⇒ γ0 = . then 2 =1+ γ0 1 1 + 1 4 ⇒ γ0 = . So we get that only two subbands each of bandwidth 10 MHz are used for transmission and the remaining two with lesser SNR’s are left unused.25 So.25. the problem can be stated as H(fi )2 P (fi ) max log2 1 + N0 s.0625. we can ﬁnd capacity as: C= j:γj ≥γ0 B log2 γj γ0 This gives us.1798 > 0. i P (fi )≤P i denote γ(fi ) = H(fi N0 )2 P (f i) L= i log2 1 + γ(fi ) P (fi ) P + λ( P (fi )) denote xi = P (fi ) P . Now.8889 < 1 This time our assumption was right.First assume that γ0 < 0.98Mbps . For the case of a discrete number of frequency bands each with a ﬂat frequency response.25 4 .0625 < γ0 < . our assumption was wrong again.t. then 3 =1+ γ0 1 1 1 + + 1 . Next we assume that 0.48 > 0. Now we assume that 0. C = 23. (a) C=13. the problem is similar to problem 10 ⇒ xi = 1 1 − γ0 γ(fi ) where γ0 is found from the constraints 1 1 − γ0 γ(fi ) = 1 and 1 1 − ≥ 0∀i γ0 γ(fi ) i 13. our assumption was wrong.
5 . gammac = a(find(a)). Pnoise = N0*Bc.:) = (1/Gammabar(i))*exp(hsquare/Gammabar(i)). pgammac = pgamma(i. for i = 1:length(Gammabar) pgamma(i. gamma = hsquare*(P/Pnoise). pgammac = pgamma(i.c] = max(a>0). C = 0.*pgammac)*ss. for i = 1:length(Gammabar) a = gamma. Bc = 4e6. for j = 1:length(gamma0v) gamma0 = gamma0v(j).c:length(gamma)). Pj_by_P = (1/gamma0)(1. .125]. end (b) C=13.001e6.125]. Pt = 30e3. sumP(j) = sumP(j) + sum(Pj_by_P.MATLAB Gammabar = [1 .001. N0 = .*(gamma>gamma0ch). gammac = a(find(a)).001e6. end gamma0v = [1:.c] = max(a>0).01:2].c] = min(abs((sumP1))). Pnoise = N0*Bc.*(gamma>gamma0).*pgammac). P = 30e3.5 .001. Bc = 4e6. gamma0ch = gamma0v(c). N0 = . end end [b. [b. ss = . sumP(j) = 0. [b. hsquare = [ss:ss:10*max(Gammabar)]. for i = 1:length(Gammabar) a = gamma.c:length(gamma))./gammac). C = C + Bc*ss*sum(log2(gammac/gamma0ch).27Mbps MATLAB Gammabarv = [1 . ss = .
01:1]. end Ctot = sum(C). Pj_by_P = (1/gamma0)(1.*pgammac)*ss. pgammac = pgamma(c:length(gamma)).*pgammac).c] = max(a>0). sumP(j) = sum(Pj_by_P. [b. pgamma = (1/Gammabar)*exp(hsquare/Gammabar). gamma = hsquare*(P/Pnoise). gammac = a(find(a)).01:./gammac).P = Pt/3. [b.c] = max(a>0). a = gamma. hsquare = [ss:ss:10*Gammabar]. gamma0ch = gamma0v(c). end [b. for j = 1:length(gamma0v) gamma0 = gamma0v(j).c] = min(abs((sumP1))). a = gamma. gammac = a(find(a)). for k = 1:length(Gammabarv) Gammabar = Gammabarv(k).*(gamma>gamma0ch). gamma0v = [.*(gamma>gamma0). . pgammac = pgamma(c:length(gamma)). C(k) = Bc*ss*sum(log2(gammac/gamma0ch).
Chapter 6 1.7 (a) Pe = P r(0 detected. 0 sent — 0 sent)p(0 sent) dmin dmin dmin + 0.25 2.3Q √ =Q √ = 0.0203 Pe = 3.7Q N0 2 N0 2 Clearly A > B.9587 Pe = 5.3Q .87 × 10−6 In part b) a=0. p0 = 0. a = [0:.1. B = 1 2Ts ⇒ Ts = 1 2B = 5 × 10−5 s P (b) SN R = N0bB = 10 Since 4QAM is multilevel signalling P Es 2Es SN R = N0bB = N0 BTs = N0 B BTs = 1 2 E ∴ SNR per symbol = Ns = 5 0 E SNR per bit = Nb = 2. for a given A we can ﬁnd B E A (d) Take Nb = N0 = 10 0 In part a) Pe = 3. (a) For sinc pulse.3Q . N0 = .00001:1].3.a > 0 0. 1 sent — 1 sent)p(1 sent) + P r(1 detected.7Q √ 2N0 2N0 2N0 = 2A 2 2A = Q N0 dmin (b) p(m = 0m = 1)p(m = 1) = p(m = 1m = 0)p(m = 0) ˆ ˆ A + a A − a 0. 2 MATLAB CODE: A = 1. t1 = .a > 0 N0 2 N0 2 Solving gives us ’a’ for a given A and N0 (c) p(m = 0m = 1)p(m = 1) = p(m = 1m = 0)p(m = 0) ˆ ˆ B A = 0. p1 = 0.42 × 10−6 Clearly part (b) is the best way to decode.7Q = 0.53 × 10−6 In part c) B=0.7*Q(A/sqrt(N0/2)).5 (a symbol has 2 bits in 4QAM) 0 E (c) SNR per symbol remains the same as before = Ns = 5 0 SNR per bit is halved as now there are 4 bits in a symbol Eb N0 = 1. .
d] = min(diff). Once again.67Q √3a 2N0 2 Ps. a(d) c 3. s(t) = ±g(t) cos 2πfc t r = r cos ∆φ ˆ where r is the signal after the sampler if there was no phase oﬀset.Q √min 2N0 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 of neighbors = 4.p1 + 0.exact = 1 − √ 2( M − 1) √ 1− Q M 3γ s M −1 . We will use the approximation Pe ∼ (average number of nearest neighbors).p1 + 0.p0 ] + 6 6 = A2 Tb c = ( p1 + p0 = 1 always ) d 5.t2=.p0 ] + 6 1 [0.3*Q(a/sqrt(N0/2)).4Q 1×4+4×3+4×2 Q 9 √3a 2N0 = 2. Pe = 4 1 − (c) Pe ∼ (d) Pe ∼ 6.p1 + 1. Pe however increases as numerator is reduced due to multiplication by cos ∆φ Pe = Q 4.p1 + 0. the threshold that ˆ minimizes Pe is 0 as (cos ∆φ) acts as a scaling factor for both +1 and 1 levels.p0 ] + [0.5Q √2a 2N 0 (b) 16QAM. A2 c Tb 0 dmin cos ∆φ √ 2N0 cos2 2πfc tdt = A2 c Tb 0 1 + cos 4πfc t 2 T b sin(4πfc Tb ) = A2 + c 8πfc 2 →0 as fc 1 =1 2 x(t) = 1 + n(t) Let prob 1 sent =p1 and prob 0 sent =p0 Pe = 1 [1. [c.5 Pe = 4. diff = abs(t1t2).p0 ] 6 1 1 [p1 + p0 ] = 6 6 2 2 [0. 2×3+3×2 Q 5 1 4 Q √2a 2N0 = 3Q √2a 2N0 √2a 2N0 √2a 2N0 = 2.
01:25].Ps_approx. (a) ∞ Ix (a) = 0 e−at dt x2 + t2 2 since the integral converges we can interchange integral and derivative for a¿0 ∂Ix (a) ∂a x2 Ix (a) − ∂Ix (a) ∂a ∞ = 0 ∞ −te−at dt x2 + t2 (x2 + t2 )e−at dt = x2 + t2 2 2 ∞ 0 = 0 e−at dt = 2 1 2 π a . Ps_approx = ((4*(sqrt(M)1))/sqrt(M))*Q(sqrt((3*gamma)/(M1))). hold on semilogy(gamma_db. gamma = 10. this is very unlikely and this overcounting becomes negligible. 7. At high SNR.approx 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. The nearest neighbor bound overestimates the error rate because it overcounts the probability that the transmitted signal is mistaken for something other than its nearest neighbors. M = 16.Figure 1: Problem 5 √ 4( M − 1) √ = Q M 3γ s M −1 Ps. semilogy(gamma_db. Ps_exact=1exp(2*log((1((2*(sqrt(M)1))/(sqrt(M)))*Q(sqrt((3*gamma)/(M1)))))). The approximation error decreases with SNR because the approximate formula is based on nearest neighbor approximation which becomes more realistic at higher SNR. MATLAB CODE: gamma_db = [0:. Ps_exact).^(gamma_db/10). See ﬁgure.’b:’). 8.
= e P (u)u 1 2 π a .10 0 Problem 2 − Symbol Error Probability for QPSK 10 Ps −100 10 −200 Approximation Exact Formula 10 −300 0 5 10 15 20 25 30 10 0 Problem 2 − Symbol Error Probability for QPSK 10 P s −1 10 −2 Approximation Exact Formula 10 −3 0 1 2 3 4 5 SNR(dB) 6 7 8 9 10 Figure 2: Problem 7 (b) Let Ix (a) = y.F. Q(a) = − 1 2 1 2 π a = e−x 2 2a ∴ e−x a y = solving we get y = (c) − π −x2 u e du a √ π ax2 e erf c(x a) 2x ∞ 0 √ 2x −ax2 2x 2 e erf c(x a) = Ix (a) e−ax = π π a=1 2 2x −ax2 ∞ e−at erf c(x) = e dt π x2 + t2 0 = √ 1 Q(x) = erf c(x/ 2) = 2 2 π 1 π π/2 0 π/2 0 e−at dt x2 + t2 2 e−x e−x 2 /sin2 θ dθ dθ 2 /2sin2 θ . we get y − x2 y = − comparing with y + P (a)y = Q(a) P (a) = −x2 I.
C = Θ = 22.39 × 10−13 = 0.0074s = 7400µs > 15µs outage or outage combined with average prob of error can be a good measure.8488W x = 1. 10.82dB Pγ. at 100 mph Tc = 0.dB ∼ N (µPγ . at 10 mph Tc = 0.dB ≥ x) Pγ. 11.074s Ts ∴ outage probability is a good measure. c = dmin . ∞ Mγ (s) = = esγ p(γ)dγ 1 esγ e−γ/γ dγ γ 0 ∞ 0 = 1 1 − γs √ √ 12.9 = −1. SN R = 25 √ √ Pe = Q( 2γ) = Q( 50) = 7.9.dB − µPγ x − µPγ P ≥ 8 8 x − µPγ ⇒Q 8 x − µPγ ⇒ 8 ⇒ µPγ = −128.9 = 0. 1 with fading Pe = 4γ = 0.2816 (b) 13. (a) When there is path loss alone.74s Ts ∴ outage probability is a good measure.01 b So the system can’t be used for data at all.39 Es Can also use formula from reader √ √ dmin 2 (b) Ps = αm Q βm γs = 2Q = 2Q( .5 E c = dmin = 2Es (1 − cos 22. It can be used for very low quality voice.076 .b = √ s .3122 × 10−14 = −138. d = 1002 + 5002 = 100 6 × 103 Pe = 1 e−γb ⇒ γb = 13.687 × 10−13 data requires Pe ∼ 10−6 voice requires Pe ∼ 10−3 so it can be used for both.5) = . (a) Law of Cosines: √ c2 = a2 + b2 − 2ab cos C with a.5672dB = 1. σdB = 8 P (Pγ. 8). Ts = 15µsec 1 1 at 1mph Tc = Bd = v/λ = 0. βm = .9 = 0.3122 × 10 Pγ Pt √ = Gλ 4πd 2 ⇒ 4.1224 2 Pγ −14 N0 B = 13.076γs ) 2No αm = 2. P = 100W N0 = 4W.1224 ⇒ Pγ = 1.
gamma_b_bar = 10. line(i)). ’b’] for i = 1:size(m.^(avg_SNR/10). 2. 4. ’m = 4’).33 × 10−3 15. line = [’k’.[].2dB 14. ylabel(’Average bit error probability ( P_b ) ’). 2. ’r’. 2) Pb_bar(i.[].076γs . ’m = 2’.65 Mγ (s) = ∴ Pb = For M = 4. ∞ Pb = 1 Pb (γ) = e−γ 2 Pb = But from 6. semilogy(avg_SNR.∞ (c) Pe = ∞ 0 Ps (γs )f (γs )dγs = 0 √ αm Q( βm γs )f (γs )dγs π 2 Using alternative Q form = αm π 1+ 0 gγs (sin φ)2 gγs 1+gγs −1 dφ with g = βm 2 m = α2 1 − the integral = 1− . . %Script used to plot the average probability of bit error for BPSK modulation in %Nakagami fading m = 1. hold on. Pb_bar(i.038γs = 1 .gamma_b_bar(j). m = [1 2 4]. γ = 10 0 Pb (γ)p(γ)dγ 1 2 ∞ 0 1 e−γb pγ (γ)dγ = M 2 −m −m 1− 1 γ 1+ 2 m sγ m P b = 3.038γs 1+. ⇒ γb = 250.0.2dB 250 Also will accept γb (16P SK) = 822 ⇒= 5. legend(’m = 1’.5 b Penalty = 3289. end xlabel(’Average SNR ( gamma_b ) in dB’).2) for j = 1:size(gamma_b_bar.1:20]. title(’Plots of P_b for BPSK modulation in Nagakami fading for m = 1. where we have used an integral table to evaluate (d) Pd = Ps 4 1 (e) BPSK: P b = 4γ = 10−3 .1). 16PSK: From above get γs = 3289.pi/2. 4’). %Initializations avg_SNR = [0:0. end figure(1).j) = (1/pi)*quad8(’nakag_MGF’.:).m(i).5 = 11.
1/2 5 10 γΣ = 35 40 w.5 × 10 mW = 88.4 m 17.^2)) ).p. 2/3 5 w. Pb = 2γb ⇒ γb = 500 −9 −12 mW ⇒ P No B = 3 × 10 target = γ b N0 B = 1.p. m.24 dBm Now.function out = nakag_MGF(phi. 4 i=1 1 1 − γ0 γi pi = 1 4 i=1 1 ⇒ =1+ γ0 pi γi ./(m*(sin(phi). w.53×10−3 1. 1/3 30 w. SNR = 10dB M 1 2 4 BER 2. Gamma_b_bar is the average SNR per bit.^(m).p. 1/2 10 w. w. g).33×10−2 5.p. For DPSK in Rayleigh fading. gamma_b_bar. γΣ = γ1 +γ2 . We ﬁrst assume γ0 < 5.p.03×10−3 1 16. (a) γ1 = γ2 = In MRC. m is the Nakagami parameter %and g is given by Pb(gamma_b) = aQ(sqrt(2*g*gamma_b)). γ ≥ γ0 0 γ < γ0 1 γ0 Notice that we will denote γΣ by γ only hereon to lighten notation.01) = 2.73 × 10−8 Ptarget −Pr σ mW = Pt λ 2 4πd 0 w.28 dBm = ⇒ d = 1372.p. consider shadowing: Pout = P [Pr < Ptarget ] = P [Ψ < Ptarget − Pr ] = Φ ⇒ Φ−1 (. Ptarget −Pr σ 3. w.327 = Pr = −74.p. %phi can be a vector. %This function calculates the mNakagami MGF function for the specified values of phi.p. 1/6 1/6 1/3 1/3 (b) Optimal Strategy is waterﬁlling with power adaptation: S(γ) = S 1 − γ . out = (1 + gamma_b_bar. So.
the capacity of the channel is given by: 4 C=B i=1 log2 (1 + γi )pi C = 451.66 Kbps Notice that we have denote γΣ by γ to lighten notation. (a) s(k) = s(k − 1) z(k − 1) = gk−1 s(k − 1) + n(k − 1) z(k) = gk s(k) + n(k) From equation 5. 4 C=B i=1 log2 γi γ0 pi C = 451.9365 < 5 So we found the correct value of γ0 . receiver knowledge. 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)nk−1 + g( k − 1) s (k − 1)nk + nk nk−1 but s(k) = s(k − 1) s(k)s (k − 1) = sk 2 = 1 (b) nk = sk−1 nk nk−1 = sk−1 nk−1 z = gk gk−1 + gk nk−1 + gk−1 nk p1 p2 A B φx (s) = = + (s − p1 )(s − p2 ) s − p1 s − p2 p1 p2 A = (s − p1 )φx (s)s=p1 = p1 − p2 p1 p2 B = (s − p2 )φx (s)s=p2 = p2 − p1 (c) Relevant part of the pdf φx (s) = ∴ px (x) = (d) Pb = prob(x < 0) = p1 p2 (p2 − p1 ) p1 p2 L−1 (p2 − p1 ) (normalized) p1 p2 (p2 − p1 )(s − p2 ) 1 (s − p2 ) = 0 −∞ p1 p2 ep2 x (p2 − p1 ) ep2 x dx = − .63 .91 Kbps (c) Without. 18.⇒ γ0 = 0.x < 0 p1 p2 − p1 .
01.0001.001. 0.138 × 10−5 1 − (ρc / 2)2 1 − ρc / 2 Since its Rayleigh fading.9998 P f loor 21. ﬂoor can be seen about γ b = 60dB when BD T = 0. we can assume that σTm ≈ µTm = 100ns Pf loor ≤ 10−4 which gives us σTm 2 ≤ 10−4 Ts σTm Ts ≥ = 10µsec Pf loor So. ISI: Formula based approach: Pf loor = σTm Ts 2 1 1− = 2 √ √ (ρc / 2)2 (2 − 2)K/2 √ √ exp − = 2.0001 where J0 is 0 order Bessel function of 1st kind. Pb = 1 1 + γ b (1 − ρc ) 2 1 + γb γ b (1 − ρc ) + 1 2(γ b + 1) 1 2(γ b + 1) when BD T = 0. ∴ Ts = 2 40×103 = 5 × 10−5 sec DQPSK 2 Gaussian Doppler power spectrum.01. ρc = e−(πBD T ) BD = 80Hz Rician fading K = 2 ρc = 0. ﬂoor can be seen between γ b = 0 to 60dB 20. γ b 0 to 60dB ρc = J0 (2πBD T ) with BD T = 0. Data rate = 40 Kbps Since DQPSK has 2 bits per symbol. . Tb ≥ 5µs. Ts ≥ 10µs. Rb ≤ 200 Kbps. 0.(e) p2 − p1 = 1 1 γb + 1 + = 2N0 [γ b (1 − ρc ) + 1] 2N0 [γ b (1 + ρc ) + 1] N0 [γ b (1 − ρc ) + 1][γ b (1 + ρc ) + 1] ∴ Pb = (f) ρc = 1 ∴ Pb = 19.001. ﬂoor can be seen about γ b = 40dB when BD T = 0.
22. BPSK d = θTm /Ts . so ρc = J0 (2πBD Ts ) = 1 − (πfD Ts )2 ≥ 0.9999 ⇒ Ts ≤ 39. we have 50.89µs. Rb ≥ 50.7Kbps QPSK d = 4 × 10−2 Ts = 75µsec R = 2/Ts = 26.26 Kbps ≤ Rb ≤ 200 Kbps (or 2 Mbps). Let Ts ≥ 10 µT . Then ⇒ ρc ≥ 0.9999 But ρc for uniform scattering is J0 (2πBD Ts ). From ﬁgure 6.5 with Pb = 10−3 . ISI will be negligible. θTm = 3µs d = 5 × 10−2 Ts = 60µsec R = 1/Ts = 16. Combining the two.26 Kbps. As long as Ts 2bits 1 R ≤ symbol Ts symbols = 2Mbps sec Doppler: BD = 80 Hz Pf loor = 10−4 ≥ 1 1− 2 √ 2 ρc / 2 √ 1 − ρc / 2 2 µT .Thumb .7Kbps MSK d = 3 × 10−2 Ts = 100µsec R = 2/Ts = 20Kbps .Rule approach: µt = 100 nsec will determine ISI.79µs Tb ≤ 19.
Ps = 2Q( γs ) ≤ 10−3 .1917 γ 1−e − γ0 γ 2 1−e − γ0 3 = 0. pγΣ (γ) = M 1 − e−γ/γ γ γ = 10dB = 10 as we increase M.^. M = [1 2 4 8 10]. Ps = 10−3 √ QPSK.8276. γ 2 = 31. (M(i)1). end 3. ∞ M −1 Pb = = 0 ∞ 0 1 −γ e pγΣ (γ)dγ 2 1 −γ M e 1 − e−γ/γ 2 γ ∞ M −1 e−γ/γ dγ M −1 = = = M 2γ M 2γ M 2 e−(1+1/γ)γ 1 − e−γ/γ M −1 n M −1 n dγ 0 M −1 n=0 M −1 n=0 (−1)n e−(1+1/γ)γ dγ (−1)n 1 = desired expression 1+n+γ .. fori=1:length(M) pgamma(i. γs ≥ γ0 = 10. M =1 γ − 0 Pout = 1 − e γ 1 M =2 γ − 0 Pout = 1 − e γ 1 M =3 γ − 0 Pout = 1 − e γ 1 = 0.0197 e−γ/γ 2. MATLAB CODE gamma = [0:. the mass in the pdf keeps on shifting to higher values of γ and so we have higher values of γ and hence lower probability of error.*(exp(gamma/gamma_bar)). γ 3 = 100. M Pout (γ0 ) = i=1 1−e − γ0 i γ γ 1 = 10..6228.6613 γ 1−e − γ0 2 = 0. gamma_bar = 10.:) = (M(i)/gamma_bar)*(1exp(gamma/gamma_bar)).1:60].Chapter 7 1.
03 0.06 pγ (γ) 0. See MATLAB CODE: gammab_dB = [0:.0129 P b (20dB) 0.^(gammab_dB/10).04 0.0076 0.1:20]. M= 2. γ1 < γ} γ < γτ P r{γτ ≤ γ1 ≤ γ} + P r{γ2 < γτ .08 0.7 × 10−5 2.0455 0. γ1 < γ} γ > γτ If the distribution is iid this reduces to pγΣ (γ) = 5.0050 9.1 M=1 M=2 M=4 M=8 M = 10 0. pγΣ (γ) = P r{γ2 < γτ . gammab = 10.05 Σ 0. ∞ Pγ1 (γ)Pγ2 (γτ ) γ < γτ P r{γτ ≤ γ1 ≤ γ} + Pγ1 (γ)Pγ2 (γτ ) γ > γτ Pb = 0 1 −γ e pγΣ (γ)dγ 2 1 −γr /γ γe 1 −γr /γ γe pγΣ (γ) = 1 − e−γT /γ 2 − e−γT /γ γ < γT γ > γT ∞ γT Pb = = 6.0. .09 0.01 0 0 10 20 30 γ 40 50 60 Figure 1: Problem 2 4. γT 1 1 e−γ/γ e−γ dγ + 1 − e−γT /γ 2 − e−γr /γ 2γ 2γ 0 1 −γT /γ + e−γT e−γT /γ 1−e 2(γ + 1) e−γ/γ e−γ dγ no diversity SC(M=2) SSC M 2 1 2(γ+1) 1 2(γ+1) Pb M −1 m m=0 (−1) 1+m+γ 1 − e−γT /γ + e−γT e−γT /γ M −1 m P b (10dB) 0.02 0.7 × 10−4 As SNR increases SSC approaches SC 7.07 0.
’). Pbs(j) = Pbs(j) + (M/2)*((1)^m)*f*(1/(1+m+gammab(j))). γΣ = 1 N0 M i=1 ai γi M 2 i=1 ai 2 ≤ 1 N0 a2 i a2 i 2 γi = 2 γi N0 . Pbs(j) = Pbs(j) + (M/2)*((1)^m)*f*(1/(1+m+gammab(j))).Pbs. end end semilogy(gammab_dB.Pbs.Pbs. end end semilogy(gammab_dB. for j = 1:length(gammab) Pbs(j) = 0 for m = 0:M1 f = factorial(M1)/(factorial(m)*factorial(M1m)). hold on 8.’b:’).avg (DPSK) 10 −3 10 −4 10 −5 10 −6 10 −7 0 2 4 6 8 10 γavg 12 14 16 18 20 Figure 2: Problem 7 for j = 1:length(gammab) Pbs(j) = 0 for m = 0:M1 f = factorial(M1)/(factorial(m)*factorial(M1m)).’b’) hold on M = 3.10 0 M=2 M=3 M=4 10 −1 10 −2 Pb. Pbs(j) = Pbs(j) + (M/2)*((1)^m)*f*(1/(1+m+gammab(j))). end end semilogy(gammab_dB.’b. hold on M = 4. for j = 1:length(gammab) Pbs(j) = 0 for m = 0:M1 f = factorial(M1)/(factorial(m)*factorial(M1m)).
take N = 3. 10. + γN . Denote N (x) = 2 √1 e−x /2 2π . (b) In MRC. So γΣ = 10N Pb = .0013. Pb = 6. 1 ≤ i ≤ N N = 1. (a) γi = 10 dB = 10. M = 4 Pb = .2e −1.4412 So.5 (Mγ −1) = . Q (x) = −N (x) ∞ Pb = Q(∞) = 0. P (0) = 0 Q( 2γ)dP (γ) 0 √ 1 d 2 1 Q( 2γ) = −N ( 2γ) √ = − √ e−γ √ dγ 2 γ 2 γ 2π ∞ 1 1 √ e−γ √ P (γ)dγ Pb = 2 γ 2π 0 M P (γ) = 1 − e−γ/γ k=1 ∞ (γ/γ)k−1 (k − 1)! 1 0 ∞ 1 1 √ e−γ √ dγ = 2 γ 2π M 1 2 M k=1 2 0 1 (γ/γ)k−1 1 √ e−γ √ e−γ/γ dγ = 2 γ (k − 1)! 2π k=1 Denote A = 1+ 1 γ −1/2 1 1 √ (k − 1)! 2 π ∞ e 0 1 −γ 1+ γ γ −1/2 γ γ k−1 dγ M −1 = m=0 1 1 √ m! 2 pi ∞ 0 e−γ/A γ −1/2 2 γ γ m dγ let γ/A2 = u M −1 = m=0 1 1 √ m! 2 pi M −1 ∞ 0 e−u u−1/2 A uA2 γ m A2 du = Pb = A + 2 1−A − 2 m=1 M −1 m=1 2m − 1 m A2m A 22m γ m A2m+1 22m γ m 2m − 1 m . γ = 10.12 ×10−8 ≤ 10−6 . Equality holds if ai = cγi where c is a constant 9.2e Σ −1. .Where the inequality above follows from CauchySchwartz condition. γΣ = γ1 + γ2 + .5 (M −1) γ = . .2e−15/3 = 0.2e−5N ≤ 10−6 ⇒ N ≥ 2.
pb_mrc =(((1Gamma)/2).0021 0. Gamma=sqrt(gammab.01:50*gammab]. gammab = 10./(1+gammab)).^0+2*((1+Gamma)/2).45 × 10−5 0.^2). MATLAB CODE: gammatv = [. gamma2 = [gammat+. gammab_dB = [10]. anssum(i) = sum(tointeg1)*.^(gammab_dB/10). for i = 1:length(gammatv) gammat = gammatv(i).^1). tointeg2 = Q(sqrt(2*gamma2)). . pb_egc = .0057 0.*exp(gamma1/gammab)).01:gammat]. gammab = 100. no diversity two two two two branch branch branch branch SC SSC EGC MRC 1 2 1 1− 2 1− Pb 1− √ Q( 2γ)pγΣ dγ √ Q( 2γ)pγΣ dγ √ Q( 2γ)pγΣ dγ √ Q( 2γ)pγΣ dγ γb 1+γ b P b (10dB) 0./(gammab+1)).67 × 10−5 1.01:50*gammab].0016 P b (20dB) 0. end 13.0030 0. γ 1 1 √ √ 2 γ B Aγ (3) B =1+ 1 (1 + γ)2 overall P b = 12.186 × 10−4 2.*(((1+Gamma)/2).*exp(gamma2/gammab)).01. gamma = [0:.0233 0.01+sum(tointeg2)*.01:.5*(1sqrt(1(1.*((1/gammab)*(2exp(gammat/gammab)).84 × 10−5 As the branch SNR increases the performance of all diversity combining schemes approaches the same.1:10]. 1 2 DenoteN (x) = √ e−x /2 2π Q (x) = [1 − φ(x)] = −N (x) ∞ ∞ Pb = Q( 2γ)dP (γ) = 0 0 1 1 √ e−γ √ P (γ)dγ 2γ 2π ∞ 0 ∞ 0 1 1 √ e−γ √ dγ = 2γ 2π 0 1 1 √ e−γ √ e−2γ/γ dγ = 2γ 2π 2γ γ dγ = 1 γ ∞ 1 1 √ Γ 2 π 1 2 1+ 2 γ = 1 2 (1) (2) 1 1 √ e−γ √ 2γ 2π πγ −γ/γ e 1 − 2Q γ where A = 1 + 2 .01:.11.*((1/gammab)*(1exp(gammat/gammab)).^2)). tointeg1 = Q(sqrt(2*gamma1)).0025 3. gamma1 = [0:.
5). p(γ) where 0 p(γ)e−xγ dγ = we will use MGF approach 1 π ∞ 0.5 dB Pb(γ) 10 −3 10 −4 10 −5 0 2 4 6 8 10 γ 12 14 16 18 20 Figure 3: Problem 13 √ 14.1041 > Pout. If each branch has γ = 10dB Rayleigh −γ/(γ/2) γΣ = overall recvd SNR = γ1 +γ2 ∼ γe(γ/2)2 γ ≥ 0 2 BPSK ∞ Pb = Q( 2γ)pγΣ dγ = 0. 10−3 = Pb = Q( 2γb ) ⇒ 4. γ = 101. P b.5 = 5. γ = 10 k−1 0 /γ) MRC Pout = 1 − e−γ0 /γ M (γ(k−1)! = 0.0055 0 17.EGC 16.75.01γ)2 = 0.M RC 15.M RC = 0.0021P b.0827 k=1 √ √ ECG Pout = 1 − e−2γR − πγR e−γR (1 − 2Q( 2γR )) = 0. 1 π π/2 0 1 sin2 φ = 1+ 3 γ 2 sin2 φ −2 Mγ − 1 sin2 φ dφ.0016 < 0.0025 4 m 18. Pb = Nakagami2 fading 1−π 2 3 2 m=0 l+m m 1+π 2 .10 −1 MRC EGC 10 −2 dB penalty ~ ./(gammab+1)). Gamma = sqrt(gammab. π= γ 1+γ Mγ − Pb = MATLAB CODE: gammab = 10^(1.01γ √ x π/2 0 Pb = Π2 Mγi − i=1 1 sin2 φ dφ = = 1 π/2 (0.12 × 10−9 .01γ sin φ)2 dφ π 0 (0.
^2))))./(2*(sin(phi). phi = [0.001.^(2)). (gammab..*((1+./(2*(sin(phi). for m = 0:2 f = factorial(2+m)/(factorial(2)*factorial(m)). pb_nakagami = (1/pi)*sum(sumvec)*. sumvec = ((1+(gammab. sumvec = (1+(gammab. end pb_rayleigh = ((1Gamma)/2)^3*sumf.001:. 2 Pb = Q 3 α = 2/3.001:pi/2]. phi = [0. 19./(1*(sin(phi).^(gammab_dB/10).^(6).^(1))..^2)))). for i = 1:length(gammabvec) gammab = gammabvec(i). pb_nakagami(i) = (1/pi)*sum(sumvec)*.1:20].001:pi/2]. Pb = 1 π π/2 1+ 0 γ 2 sin2 φ −2 1+ γ sin2 φ −1 dφ gammab_dB = [5:. sumf = sumf+f*((1+Gamma)/2)^m. end 10 −2 10 −3 10 Pbavg 10 −4 −5 10 −6 10 −7 5 10 γavg (dB) 15 20 Figure 4: Problem 19 20. gammabvec = 10.001:.^2)))).sumf = 0. Mγ g − 2 sin φ π/2 0 2γb (3) sin g = 3 sin2 = π 8 π 8 −1 gγ 1+ sin2 φ −M α Pb = π gγ 1+ sin2 φ dφ .001.
for k = 1:length(M) for i = 1:length(gammabvec) gammab = gammabvec(i).001:pi/2]. g = 3*sin(pi/8)^2.^2)))).MATLAB CODE: M = [1 2 4 8].1:20].^(gammab_dB/10). end end 10 0 10 −5 Pb avg 10 −10 10 −15 5 10 γavg (dB) 15 20 Figure 5: Problem 20 . alpha = 2/3. gammabvec = 10. gammab_dB = [5:. phi = [0.001. pb_nakagami(k.001:. sumvec = ((1+((g*gammab).i) = (alpha/pi)*sum(sumvec)*./(1*(sin(phi).^(M(k))).
Q(z) = Q2 (z) = Ps (γs ) = 1 π 1 π 4 π 4 π Ps = = 0 π/2 exp − 0 π/4 z2 dφ sin2 φ z2 dφ 2 sin2 φ π/2 . K = 2. gammab = 10.0553 = 1.^(gammab_dB/10).21. .1670 1.z > 0 gγs dφ − sin2 φ gγs dφ sin2 φ exp − 0 ∞ 1 1− √ M 1 2 1− √ M exp − 0 π/4 exp − 0 Ps (γΣ )pγΣ (γΣ )dγΣ 1 1− √ M 1 1− √ M 1 1− √ M 1 1− √ M π/2 0 2 0 π/2 0 2 0 π/4 0 0 ∞ ∞ 4 π 4 π exp exp gγΣ sin2 φ gγΣ sin2 φ g sin2 φ pγΣ (γ)dγΣ dφ − pγΣ (γ)dγΣ dφ But γΣ = γ1 + γ2 + . .5 15 MATLAB CODE: gammab_dB = 10.z > 0 . + γM = Σγi = 4 π 4 π 22. Rayleigh: Mγs (s) = (1 − sγ s )−1 Rician: Mγs (s) = MPSK 1+k 1+k−sγ s ΠM Mγi − i=1 dφ − dφ π/4 ΠM Mγi − i=1 g sin2 φ exp ks γ s 1+k−sγ s (M −1)π/M Ps = Three branch diversity Ps = g = sin2 π 16 1 π 0 M γs − g sin2 φ dφ → no diversity (M −1)π/M 1+ 0 gγ sin2 φ −1 (1 + k) sin2 φ kγ s g exp − (1 + k) sin2 φ + gγ s (1 + k) sin2 φ + gγ s 2 dφ = 0. .5 16−1 MQAM: Formula derived in previous problem with g = P s = 0.
gammab_dB = [5:.^2). sumvec1=((1+((g*gammab).5/(161).^.001:.5/(161).. .001:pi*(15/16)]./(sin(phi2).001:..*exp((K*gammab*g). (1+K)*sin(phi). for k = 1:length(M) for i = 1:length(gammabvec) gammab = gammabvec(i).*exp((K*gammab*g). g*gammab)). sumvec2=((1+((g*gammab).^2+g*gammab))). pb_mrc_psk = (1/pi)*sum(sumvec)*. g = 1./((1+K)*. 10 0 10 −1 10 −2 10 −3 10 avg −4 Ps 10 −5 10 −6 10 −7 10 −8 10 −9 5 10 15 20 Figure 6: Problem 22 23. g = 1.001.001 ..001:pi/4].g = sin(pi/16)^2. sin(phi1).*(((((1+K)*sin(phi1).^2+g*gammab)). (1+K)*sin(phi2)./((1+K)*sin(phi).*((((.^2))). phi2 = [0..001:pi/2]... phi1 = [0..^2+. (1)).. MATLAB CODE: M = [1 2 4 8].. 1+K)*sin(phi1)./((1+K)*sin(phi2)..*((((. (4/pi)*(1(1/sqrt(16)))^2*sum(sumvec2)*. sumvec=((1+((g*gammab).001./(sin(phi1)./((.^2+g*gammab))). alpha = 2/3../(sin(phi).*exp((K*gammab*g).^2+.001:.^2). pb_mrc_qam = (4/pi)*(1(1/sqrt(16)))*sum(sumvec1)*.^(1)).^(1)). phi1 = [0.^2).^2))).. gammabvec = 10.^2).^(gammab_dB/10)..^2))).001:pi/2].^2).^2)./((1+K)*sin(phi)....1:20]. phi = [0. g*gammab))..001:./((1+K)*sin(phi2).^2+g*gammab))).
.^(M(k)))./(1*(sin(phi1).^2)))).^2))))./(1*(sin(phi2).001 .001:..i) = (4/pi)*(1(1/sqrt(16)))*sum(sumvec1)*. sumvec1 = ((1+((g*gammab). pb_mrc_qam(k. (4/pi)*(1(1/sqrt(16)))^2*sum(sumvec2)*. end end .phi2 = [0.001. sumvec2 = ((1+((g*gammab)..001:pi/4].^(M(k))).
8685 −0. H = U ΣV T −0.e.7164 1.7034 0 0 0 0.1302 Rank(A) (1 + λi ) .6508 −0. λ = λ.4793 0. (c) IM + AAH = IM + QΛQH = Q(I + Λ)QH AH positive semideﬁnite ⇒ λi ≥ 0∀i ∴ 1 + λi > 0∀i ∴ IM + AAH positive deﬁnite (d) det[IM + AAH ] = det[IM + QΛQH ] = det[Q(IM + ΛM )QH ] = det[IM + ΛM ] = Πi=1 Rank(A) T = AAH (1 + λi ) det[IN + AH A] = det[IN + QΛQH ] = det[Q(IN + ΛN )QH ] = det[IN + ΛN ] = Πi=1 AAH and AH A have the same eigenvalue ∴ det[IM + AAH ] = det[IN + AH A] 2. i. eigenvalues are real AAH = QΛQH (b) X H AAH X = (X H A)(X H A)H = X H A ≥ 0 ∴ AAH is positive semideﬁnite.4272 −0. (a) (AAH )T = (AH )T .AT = (AT ) AT = AAH ∴ (AAH )H For AAH .Chapter 10 1.7152 0 Σ= 0 0 0.2513 0.1298 U = −0.5896 −0.6855 −0.
11 H −H .3311 −0.3876 .8894 .86 for i = 2. Check the rank of each matrix rank(HI ) = 3 ∴ multiplexing gain = 3 rank(H2 ) = 4 ∴ multiplexing gain = 4 5.09 D= . .2198 0.14 where : dij = ij ij × 100 H .0145 −0.4188 −. For example: .05 .13 .68 for i = 3 γ3 = .0708 V = −0.9017 3.2407 −. −0.8478 . H = U ΣV T Let 1 0 U = 0 1 0 0 1 0 V = 0 1 0 0 Σ= 1 0 0 2 1 0 0 ∴H= 0 2 0 0 0 0 4.2191 0.2423 −.7813 F = −. 0 2 P γi = λio B = 94.10 (b) precoding ﬁlter M = V −1 shaping ﬁlter F = U −1 −.3460 −.9078 .8540 .3458 0. C= i=1 RH log2 1 + λi ρ Mt Constraint Vi = ρ λi = constant ∴ ∂C ρ 1 1 ρ = − =0 λi ρ ∂λi Mt ln 2 (1 + ) Mt ln 2 (1 + λi ρ ) Mt Mt ⇒ λi = λj ∴ when all RH singular values are equal.23 . 6.8473 −.4629 −.5195 −. this capacity is maximized.5 for i = 1.68 is clearly too small for data transmission .4263 −0.5708 0.2373 . N Assume γ2 > γ0 > γ3 since Pi P = else 6.08 .0251 −.3887 M = −.13 show H ≈ U ΛV is acceptable.6849 0.3622 Thus Y = F(H)MX + FN = U ∗ U ΛVV ∗ X + U ∗ N = ΛX + U ∗ N (c) 1 1 1 1 γo − γi for γi > γo .7116 −0. (a) Any method to .6109 0.4727 −.
i ρi ≤ρ 8.78)(100) = 78. C= ρi : max i ρi ≤ρ B log2 (1 + λi ρi ) i √ where λi are the RH nonzero singular values of the channel matrix H and ρ is the SNR constraint. N0 B (1) This gives a capacity of 630. it will perform an SVD decomposition of H as H = U ΣV HRX H H = (U ΣV )RX (U ΣV )H By Hadamard’s inequality we have that for A ∈ n×n det(A) ≤ Πn Aii i=1 with equality iﬀ A is diagonal. then the SNR with beamforming would be equal to the largest SNR in part(c).9 kbps Pi P (d) With equal weight beamforming.2 1 1 = 1 ⇒ γ0 − γ1 − γ2 = 1 ⇒ γ0 = 1. say = Ω then det(IM R + HRX H H ) = det(I + ΩΣ2 ) ∴ C = max Bσi log2 (1 + λi ρi ) √ where λi are the singular values. We choose RX to be diagonal. 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. The SNR (3) is then given by: SN R = cH H H Hc = (. The resulting capacity is given as C= i:γi ≥γ0 B log2 (γi /γ0 ) .35 kbps. γi = λi ρ Then the optimal power allocation is given as Pi = P 0 1 γ0 − 1 γi γi ≥ γ0 γi < γ 0 (2) for some cutoﬀ value γ0 .4324 C = B log2 1 + γ1 P1 + log2 1 + γ2 P2 P P = 775.73 P1 P2 P = . The capacity of the channel is found by the decomposition of the channel into RH parallel channels. If we had chosen c to equal the eigenvector corresponding to the best eigenvalue.5676 P = . the beamforming vector is given by c = √1 [1 1 1]. where RH is the rank of the channel matric H. The SNR achieved with beamforming is smaller than the best channel in part (c). 7. C = max B log2 det[IM γ + HRX H H ] RX : Tγ (RX ) = ρ If the channel is known to the transmitter.
γ0 = 3 1+ 3 1 i=1 γi which gives γ0 = 2. . 9. γ4 = 40. . γ3 = 40.. γ1 = 80. 4 γ0 = 1 1 + 4 γi i=1 which gives γ0 = 3. . C = 13.4732 bits/sec/Hz B For 1 1 1 −1 1 1 −1 1 H= 1 −1 1 1 1 −1 −1 −1 RH = 4. . h1Mt . hMr 1 .. hence the assumption was correct. . .. .. .8236 < mini γi . hence the assumption was correct. . hMr Mt M r ×Mt . γ1 = 40. .. We ﬁrst assume that γ0 is less than the minimum γi which is 40. γ2 = 40.For 1 1 H= 1 1 1 −1 1 1 −1 −1 1 1 1 1 1 −1 RH = 3..6780 < mini γi . γ3 = 40. C = 12. H = . γ2 = 40.7720 bits/sec/Hz B h11 . . We ﬁrst assume that γ0 is less than the minimum γi which is 40.
. hiMt ] Mt →∞ Mt 1 Mt →∞ Mt lim 2 Mt hi1 .Mt. Mt = 1.i=j lim 1 = lim [hi1 . . hiMt = hij j=1 2 = Ej hij = σ2 = 1 ∀i 1 Gij Mt Mt →∞.’dBW’. M] = svd(H). . . . rho_dB = [0:25]. . hjMt hik hjk k=1 ∴ lim ∴ lim B log2 det IM M →∞ M →∞ 1 HH T M ρ HH T + M = IM = B log2 det [IM + ρIM ] = B log2 [1 + ρ] det IM = M B log2 [1 + ρ] 10.’complex’).Denote G = HH T 1 Gii = lim Mt →∞ Mt 1 lim [hi1 .0. Mr = 1. clc. . i = j Mt hj1 .Mr) . rho = 10. j. We ﬁnd the capacity by randomly generating 103 channel instantiations and then averaging over it. hiMt ] Mt →∞ Mt 1 = lim Mt →∞ Mt = Ek hik hjk = Ek (hik )Ek (hjk ) = 0 ∀i. for j = 1:min(Mt. . MATLAB CODE clear.^(rho_dB/10). [F. for k = 1:length(rho) for i = 1:100 H = wgn(Mr. L. We assume that distribution is uniform over the instantiations.
/gamma0)). MATLAB CODE clear. . end Cergodic(k) = mean(C). gamma = rho(k)*sigma_used./gammatemp1)). while gamma0 > gammatemp1(length(gammatemp1)). gamma0 = 1e3. end 25 Mt = Mr = 3 Mt = 2 Mr =3 Mt = Mr = 2 Mt = 2 Mr =1 Mt = Mr = 1 20 15 Cergodic 10 5 0 0 5 10 ρ (dB) 15 20 25 Figure 1: Problem 10 11. end C(i) = sum(log2(gammatemp1. gamma0 = length(gammatemp1)/(1+sum(1. gammatemp = gammatemp(1:length(gammatemp)1). Mt = 1.sigma(j) = L(j. gammatemp1 = gammatemp. end sigma_used = sigma(1:rank(H)). clc. gammatemp1 = gammatemp.j). We ﬁnd the capacity by randomly generating 104 channel instantiations and then averaging over it. %% Now we do water filling\ gammatemp = gamma. We assume that distribution is uniform over the instantiations.
L. while pout > . for j = 1:min(Mt.’dBW’.1. M] = svd(H). end if Cout(k)<0. rho_dB = [0:30].Mt. [F. pout = sum(C<Cout(k))/length(C).0. C(i) = sum(log2(1+gamma/Mt)).Mr) sigma(j) = L(j.^(rho_dB/10). end end 25 Mt = Mr = 3 Mt = 2 Mr =3 Mt = Mr = 2 Mt = 2 Mr =1 Mt = Mr = 1 20 15 Coutage 10 5 0 0 5 10 15 ρ (dB) 20 25 30 Figure 2: Problem 11 .’complex’).j). end sigma_used = sigma(1:rank(H)). rho = 10. for k = 1:length(rho) for i = 1:1000 H = wgn(Mr. end Cout(k) = mean(C). pout = sum(C<Cout(k))/length(C).01 Cout(k) = Cout(k).Mr = 1. Cout(k) = 0. gamma = rho(k)*sigma_used.
Σx = = u Hvx u Hv 2 x H 2 2 = vH H H u u Hv x 2 2 = v H H H Hv x = v H QH Qv x ≤ λmax x 2 with equality when u.1818 = −0.17 in reader.4190 −0.7 When both the transmitter and the receiver know the channel. we get that the maximum singular value of H is 1.4641 −0. u and v correspond to the principal singular vectors (or the singular vectors corresponding to the maximum singular value of H).5294 and vopt −0.4480 and the singular vectors corresponding to this value are −0.2 0. v are the principal left and right singular vectors of the channel matrix H ∴ SN Rmax = λmax 14.4480 opt Since. Using Matlab.3 0.6 0.3 0.9 H = 0.5 0.7101 uopt = −0.12.1 0. x 2 = λmax ρ N 0. y = (uT Hv)x + uT n . for beamforming. Mr P (u n < X) = P i=1 Mr ui ni < 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. Notice that the singular values of H are the square root of the eigen values of HH H (Wishart Matrix).8896 T It is easy to check that uT uopt = 1 and vopt vopt = 1 and that opt uT ∗ H ∗ vopt = 1.1 0. during beamforming from eq. 10.
5090 u2 = h 0.5090 It is easy to check that uT u2 = 1 and that 2 uT ∗ h = 1. SNR is simply (1. noise power is not increased. we get that the maximum singular value of h is 1. When the channel is not known to the transmitter.6941 = 0.2477 and the singular vector corresponding to this value is 0. For a given transmit SNR of ρ.and for a given transmit SNR of ρ. u2 has norm 1. uopt has norm 1. 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. the received SNR is given as SNRrcvd = ρ(uT Hv2 )2 2 since. SNR is simply (1. Solving the equation that relates diversity gain d to multiplexing gain r at high SNR’s we get d = (Mr − r)(Mt − r) . For. noise power is not increased.2477)2 = 1.0968. 15. we should have d ≥ 3. eq. or at least d = 3.5090 0.5567. For. 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 (u h) is maximized.2477 2 Alternatively. from MRC concept we know that: 0.4480)2 = 2.5090 hH where h is the L2 norm of h. ρ = 1. (a) ρ = 10 dB = 10 Pe = ρ−d So to have Pe ≤ 10−3 .6941 u2 = 0. ρ = 1. Using Matlab. the received SNR is given as SNRrcvd = ρ(uT Hvopt )2 opt since.
038bps/Hz 17. H= P = 10mW N0 = 10−9 W/Hz B = 100 KHz (a) When H is known both at the transmitter and at the receiver.24 hold at ﬁnite SNR’s too and we are given that we can use base 2 for logarithms.5757 −. No credits for this part: If we are allowed to assume that equations 10. we have to ﬁnd the cutoﬀ value γ0 .⇒ 3 = (8 − r)(4 − r) Solving for r we get r = 3. 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. Pe = ρ−d = 10−5 16. We can then do waterﬁlling over the two parallel channels available to get capacity.3328 −.5946 . r = 3.35. First assume that γ0 is less than both γ1 and γ2 .8041 .08 N0 B .8507 γ1 = γ2 = Finding γ0 Now.10) = 4. So.92 N0 B λ2 P 2 = 11.8507 .8041 −. Finding the γi ’s λ2 P 1 = 75.3 .23 and 10.5 .242 ∴ C/B = log2 (1 + λρ) = log2 (1 + 1. which gives us r=3 .2 = −. Then 1 1 − γ0 γ1 + 1 1 − γ0 γ2 =1 . we can ﬁnd the data rate as R = r log2 (ρ) = 9.8713 0 0 . so r ≤ 4 and so r = 3.5757 −.64 We have that r ≤ min{Mr . √ According to SVD of h λ = 1.35 or 8. we take the nearest integer which is smaller than the calculated value of r. we can ﬁnd d as d = (Mr − r)(Mt − r) = (8 − 3)(4 − 3) = 5 For this value of d. But we know that r has to be an integer. Mt }.7 .2422 .5946 .96 bits/s/Hz (b) With.
2 1 1 =1+ + γ0 γ1 γ2 1 ⇒ γ0 = 1 1 = 1. Finding γ0 or γK We now ﬁnd the cutoﬀ γ0 . M (γ) = 1 + γK Finding K −1. γ2 }. γK = γ0 /K = 5.08. R is given as R = B log2 γ1 γK + log2 γ2 γK .1742 Finding Rate R Therefore the total rate. so our assumption was correct.283 ln(5Pb ) S(γ) S K= γK = γ0 /K. 1 1 − γ0 γ1 K ⇒ + 1 1 − γ0 γ2 K =1 2 1 1 =1+ + γ0 γ1 K γ2 K 1 ⇒ γ0 = = 1. Notice that γ1 and γ2 have already been calculated in part (a) as γ1 = 75.5 = .81 1 + γ1 + γ2 ⇒ which is less than both γ1 and γ2 values so our assumption was correct.92 and γ2 = 11. First assume that γ0 < {γ1 .4649 1 1 1 1 + K γ1 + γ2 which is less than both γ1 and γ2 values. Finding capacity Now we can use the capacity expression as 2 C= i=1 B log2 γi γ0 = 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.
γs = γb . [F. Now we can use the expression for Pb for BPSK Pb = Q 2γb = Q √ 2 × 75. If we are willing to decrease the rate at which we transmit. Comparing with previous part Comparing with part (b). 18.Mr) . we can see that the rate R decreases by 397. M] = svd(H). we get the rate using BPSK to be R=100 Kbps .⇒ R = B4. for j = 1:min(Mt.36 Kbps and the Pb improves as Pb is now 3. λmax is 0. Once this is done the SNR at the combiner output is simply λmax ρ. given in the Ques. one strong channel which gives a much less value of Pb .0.92 .’dBW’.’complex’). the transmitter and the receiver use the principal left and right eigen vectors of the Wishart Matrix HHH . (a) clear.e.^(rho_dB/10). Finding Rate R Since we are using BPSK and are given that B = 1/Tb .4 × 10−35 ∼ 0 whereas earlier it was 10−3 . rho = 10. Mr = Mt. 3. where λmax is the maximum eigen value of the Wishart Matrix HHH and ρ is NPB 0 Finding γs As given in the question.7592 and ρ was calculated to be 100. clc. Mt = 4.97 This gives that R=497. for k = 1:length(rho) for i = 1:1000 H = wgn(Mr. L. (d) Therefore we see that we can tradeoﬀ rate for robustness of the system.36 Kbps (Obviously less than ergodic capacity).4 × 10−35 Using Matlab Credit is given for either value.Mt. rho_dB = [0:20]. Finding Pb When using BPSK. (c) Since now we use beamforming to get diversity only.92 = 0 Using the approx. So we get that γs = 75. we can get more diversity advantage i.
^(rho_dB/10). Mt = 4.j). gammatemp1 = gammatemp.’dBW’. L.0.8320 MATLAB CODE clear.sigma(j) = L(j. gamma0 = 1e3. end sigma_used = sigma(1:rank(H)). Mr = Mt./gamma0)). gamma0 = length(gammatemp1)/(1+sum(1. gammatemp = gammatemp(1:length(gammatemp)1). %% Now we do water filling\ gammatemp = gamma. for k = 1:length(rho) for i = 1:1000 H = wgn(Mr. end 19. Mr = Mt. M] = svd(H). clc. gamma = rho(k)*sigma_used. rho = 10. end Cout(k) = mean(C).’complex’). clc. rho = 10.j). gammatemp1 = gammatemp. end sigma_used = sigma(1:rank(H)). for k = 1:length(rho) for i = 1:1000 . end Cergodic(k) = mean(C). using Matlab we get Cout = 7. rho_dB = 10. [F./gammatemp1)).^(rho_dB/10). gamma = rho(k)*sigma_used. rho_dB = [0:20].Mr) sigma(j) = L(j. end (b) clear. Mt = 4. for j = 1:min(Mt. C(i) = sum(log2(1+gamma/Mt)). while gamma0 > gammatemp1(length(gammatemp1)).Mt. end C(i) = sum(log2(gammatemp1.
Cout(k) = 0. C(i) = sum(log2(1+gamma/Mt)). end end 20.j).Mt. As µ increases. . for j = 1:min(Mt.0. [F.Mr) sigma(j) = L(j.’complex’).’dBW’. MATLAB CODE clear. pout = sum(C<Cout(k))/length(C). end sigma_used = sigma(1:rank(H)). the span of cdf becomes narrower and so capacity starts converging to a single number.1 Cout(k) = Cout(k). clc. gamma = rho(k)*sigma_used. L. end Cout(k) = mean(C). pout = sum(C<Cout(k))/length(C). M] = svd(H). while pout > . end if Cout(k)<0.01.25 Cergodic Mt = Mr = 1 C M =M =1 out t r Cergodic Mt = Mr = 4 Cout Mt = Mr = 4 20 15 C 10 5 0 0 2 4 6 8 10 ρ (dB) 12 14 16 18 20 Figure 3: Problem 18 H = wgn(Mr.
6 FX(x) 0.’complex’). j = 1:min(Mt.4 0. for i = H = [F. Empirical CDF’s of Capacity w/o Tx CSI 1 M=4 M=6 M=8 0.Mr) sigma(j) = L(j.8 0.Mt = 8.Mt. rho_dB = 10.9 0. for 1:1000 wgn(Mr.x] = ecdf(C). C(i) = sum(log2(1+gamma/Mt)). L.7 0. Mr = Mt. rho = 10.’dBW’.0. end sigma_used = sigma(1:rank(H)). gamma = rho*sigma_used.3 0.2 0.1 0 6 7 8 9 10 11 12 x 13 14 15 16 17 18 Figure 4: Problem 20 .j). end [f.^(rho_dB/10). M] = svd(H).5 0.
fc = 100 MHz 1 =f H(f ) fc+B Heq (f ) = Noise PSD = N0 W/Hz. (a) For the ﬁrst channel: ISI power over a bit time = A2 Tb /Tb = A2 For the 2nd channel: 2 (n+1)T ISI power over a bit time = Ab ∞ nTb b e−t/Tm dt = 2e−1/2 A2 n=1 T . B = 50 KHz. See Fig 1 2B = 100 KHz fcB fc = 100 MHz Figure 1: Band of interest. 2. As seen from the noise power values. Using this we get Noise Power = = N0 Heq (f )2 df fc −B fc +B N0 f 2 df fc −B 3 (fc +B) f 3 (fc −B) fc +B (1) (2) (3) (4) (5) = N0 = N0 (fc + B)3 − (fc − B)3 3 = 1021 N0 W Without the equalizer.Chapter 11 1. the noise power will be 2BN0 = 105 N0 W. there is tremendous noise enhancement and so the equalizer will not improve system performance.
8Kbps If baseband signal =100KHz: pulse width = 10µs Data rate = 2/10µs + 10µs = 100Kbps 3.5µs = 181.w. (a) h(t) = τ = 6 µ sec 1 H(f ) t e− τ 0 t t≥0 o. 1 τ 1 + j2πf 1 + j2πf τ Heq (f ) = (b) B −B Sx (f )H(f )2 Heq (f )2 df B 2 −B N0 Heq (f ) df B 2 −B Sx (f )H(f ) df SNReq = SNRISI 2BN0 .S(t) A T /2 m t h (t) 1 1 T m t h (t) 2 t Figure 2: Problem 2a (b) No ISI: pulse interval = 11/2µs = 5. (6) Heq (f ) = ∞ H(f ) = 0 e− τ e−j2πf t dt (7) (8) = Hence.5µs ∴ Data rate = 1/5.
we have a0 =1. −3Ts τ −Ts τ z −1 + e z −1 n z −2 + e z − Ts τ z −3 + . 1 Thus. . . a two tap ﬁlter is suﬃcient. a1 < 1 1 2 .. say F (z) = 1 − a1 z then 1 = 1 + a1 z −1 + a2 z −2 + . + ωN z −N ) F (z) at z = ejω If F(z) is of length 2 and monic. 4. Now. we need to use some approximation to come up with the ﬁlter tap Ts coeﬃcient values.9364 = −0.28 dB (c) h [n] = 1 + e H(z) = 1 + e ∞ −Ts τ δ [n − 1] + e −2Ts τ −2T s τ δ [n − 2] + .h (t) 1 1 10us t X (t) 1 1us t h (t)*X(t) 1 1 12us t Figure 3: Problem 2b Assume Sx (f ) = S..0039 as the tap coeﬃcient values. ωi = ci where {ci } is the inverse Z. . Any other reasonable way is also accepted. (1) . 1 1− Ts e− τ z −1 = n=0 e− Ts τ = z−e = ⇒ Heq (z) = 1 H(z)+N0 .617 × 10−6 = 0. . c2 = a2 . we get Heq (z) = 1−e− τ z −1 . 1 F (z) where c1 = a1 . . . −B ≤ f ≤ B ⇒ N0 S 2 2B + 8π B 3 3 τ2 B −B 2BS H(f )2 df 2BN0 = 2 1 + 4π B 2 3 τ2 2B 1. a1 = −0.. For Ts = 30 ms..transform of 1/F(z) Show that this choice of tap weights minimizes 1 − (ω−N z N + . If we assume N0 0 (the zeroforcing assumption).
3.5464 P3 = 2.5SN Pb ≤ 0. So if we had the opportunity to cancel any (2N+1) coeﬃcients we will cancel the ones that are closest to z 0 .2831 Pi = P 1 γ0 − 0 1 Kγi γi ≥ γ0 /K γi ≤ γ0 /K (10) We can see that all subchannels will be used and P1 = 2.7207 thus R = 2B 6. (a) Heq (f ) = 1 H(f ) for ZF equalizer 1 2 0. n Ts FZ (f ) = 1 TS F n=−∞ f+ = 1 ∴ folded spectrum of f(t) is ﬂat.5 3: fc ≤ f < fc + 10M Hz α3 = 2 4: fc + 10M Hz ≤ f < fc + 20M Hz α4 = 0.w.1225 and γ0 = 3.6523 P2 = 2.0125µsec 1.2193M bps T (d) We use M=4 non overlapping subchannels.52 × 10M Hz + 42 × 10M Hz] = 0. (9) (b) S=10mW Signal power N = N0 [12 × 10M Hz + 22 × 10M Hz + 0.It is easy to see that the coeﬃcients become smaller and smaller. 2.5γ/M −1 or M ≤ 1 + −ln(5PR) b for Pb = 10−3 M ≤ 14. 4 γ1 = 1000 γ2 = 250 γ3 = 4000 γ4 = 62.5 Heq (f ) = 4 0 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. .2e−1.25 P α2 i Power optimization: γi = N0 B for i = 1.73dB (c) Ts = 0. Hence we get that ωi = ci minimizes (1).5 for Pb = 10−3 K = 0.6788 P4 = 2.0588 = 16. (a) F{f (t)} = T 0 ∞ 4 i=1 log(Kγi /γ0 ) = 419. each with B=10MHz bandwidth 1: fc − 20M Hz ≤ f < fc − 10M Hz α1 = 1 2: fc − 10M Hz ≤ f < fc α2 = 0. 5.w.9711M bps f  < 1/T o. The result can be similarly proved for length of F(z) greate than 2 or nonmonic.3228 (M ≥ 4 thus using the formula is reasonable) R = log2sM = 307.2125mW ∴ SN R = 47.
i=k Xi f ((k − i)T + t0 ) ISI (c) N +k ISI = i=−N +k. t < Ts 1 Noise whitening ﬁlter : G (1/z ) m . gm (t) = g (−t) = g(t) = sinc(t/TS ). ISI → ∞ as N → ∞ 7.i=k Xi sin (π(k − i) + t0 /T π) π(k − i) + t0 /T π N = sin(πt0 /T ) i=−N.i=0 N 1 πt0 /T − πi = sin(πt0 /T ) = 2 sin(πt0 /T ) π i=1 N −1 1 + πt0 /T − πi πt0 /T − πi n2 n − t2 /T 2 0 n=1 Thus.(b) yk = y(kT + t0 ) ∞ = i=−∞ N Xi f (kT + t0 − iT ) Xi f ((k − i)T + t0 ) i=−N N +k = = Xk sinc(t0 ) + i=−N +k.
5Ts −0.8. 0. VW J J ∴ = = ∂J ∂J ..5Ts ∴ 0 ≤ Jmin ≤ 1 ..5Ts ∞ n=−∞ H(ω = Ts N0 df FΣ (f ) + N0 9.5Ts Ts 1df = 1 −0.. ∂w0 ∂wN wT Mv w − 2 {Vd w } + 1 ∂J = 2Mv wT − 2Vd ∂w ∂J = 0 ⇒ 2Mv wT = 2Vd ∂w T −1 H ⇒ wopt = Mv Vd 10..5Ts N0 df FΣ (f ) + N0 ∴ Jmin ≥ 0 N0 =1 N0 0. Jmin = 1 − ∞ j=−∞ cj f−j B(z) = C(z)F (z) F (z)F (z −1 ) = F (z)F (z −1 ) + N0 X(z) = X(z) + N0 1 B(z) ∴ b0 = dz 2πj z 1 X(z) = dz 2πj z[X(z) + N0 ] = ∴ Jmin = 1 − T 2π π/T −π/T T 2π π/T −π/T X(ejωT ) dω X(ejωT ) + N0 X(ejωT ) X(ejωT ) + N0 dω = = T 2π T 2π π/T −π/T π/T N0 jωT ) X(e T −1 + N0 dω N0 dω + 2πn/T )2 + N0 −π/T −0.5Ts Jmin N0 FΣ (f ) + N0 N0 FΣ (f ) + N0 ∴ Jmin = ≥ ≤ ≤ Ts 0 −0.
and the noise power is given by the integral of N (f ) from 50 kHz to 50 kHz: 50kHz 50kHz N= f =−50kHz N (f )df = 2N0 f =0kHz Heq (f )2 df = 2N0 (1 + 4 + 9 + 16 + 25)(10kHz) = 1.5 we get N = 2N0 (. frequency selective fading (c) Require Tb = Tm + T ⇒ R = 1 Tm +T = 49997. FΣ (f ) = = 1 Ts 1 Ts ∞ F n=−∞ ∞ f+ n Ts n −j2π f + T s 1 + 0. For α = .69))(10kHz) = 0.5/Ts N0 df FΣ (f ) + N0 N0 FΣ (f ) + N0 Jmin = exp Ts ln −0.44 + 1. But 2/T is also acceptable (Null to Null bandwidth) (b) τ T is more likely since T = 10−9 sec As long as τ > Tb .44 + .5/Ts df 12.44 + 1 + 1.5bps 1 (d) Heq (z) = F (z) for ZF equalizer 1 ⇒ Heq (z) = d0 +d1 z −1 +d2 z −2 Long division yields the ﬁrst 2 taps as w0 = 1/α0 2 w1 = −α1 /α0 13.0516 mW .64 + .11.78 + 2.134 mW For α = 1 we get N = 2N0 (.5/Ts Jmin = Ts DF equalizer : −0.3e n −j4π f + T s MMSE equalizer : 0.25 + . get ISI and so.04))(10kHz) = 0. so theoretically inﬁnite.5/Ts 0. (a) 1 2 1 Hzf (f ) = = 3 H(f ) 4 5 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 .1mW N0 (c) The noise spectrum at the output of the ﬁlter is given by N (f ) = (H(f )+α)2 .56 + .5e n=−∞ + 0. (a) G(f) is a sinc(). and the noise power is given by the integral of N (f ) from 50 kHz to 50 kHz.
5 msec = 162 Kbps Case2: fD = 1000 Hz ⇒ retrain every 0. . 15. Thus. In fact it has to be retrained at least every channel correlation time. ∴ RW = p) ∴ GK = RWk − p = −E[εk Yk ] ˆ ˆ ˆ where Yk = [yk+L . The bit time is diﬀerent for LMSDFE/RLS. which also depends on the spectrum of the input signal (which is not given here). yK−L ]T and εk = dk − dk Approximately (1) can be rewritten as Wk+1 = Wk + ∆εk Yk . . one bit sent. N = 4 LMSDFE: 2N +1 operations/iteration ⇒ 9 operations/iteration RLS: 2. Beneﬁts of training (a) Use detected data to adjust the equalizer coeﬃcients.5(N )2 + 4. . but the signal power decreases as well. The factor α should be chosen to balance maximizing the SNR and minimizing distortion.4 Kbps 16. In the adaptive method.5N operations/iteration ⇒ 58 operations/iteration Each iteration. But time to convergence is faster for RLS. . (1) ˆ ˆ ˆ where ∆ is some small positive number and GK is the gradient of MSE = Edk − dk 2 is RWk − p (Notice that 11. the signal power also goes to zero. (e) As α → ∞.5 msec RLMSDFE = 0 bps RRLS = 72. However. the noise power decreases. Tb (LMSDFE) <Tb (RLS). Case 1: fD = 100 Hz ⇒ (∆tc ) ≡ 10 msec. must retrain every 5 msec. Can work without training information (b) eliminate ISI. we start with some initial value of tap coeﬃcients W0 and then use the steepest descent method Wk+1 = WK − ∆GK . The equalizer must be retrained because the channel decorrelates. LMSDFE: R = RLS: R = 107 9  1000 bits 5 msecs = 911 Kbps 107 50 bits 58 . 14. the frequency response Heq (f ) decreases for all f .37 was a solution of gradient =0 . the noise power goes to 0 because Heq (f ) → 0 for all f .(d) As α increases.
0) (0. City has 10 macrocells each cell has 100 users ∴ total number of users = 1000 Cells are of size 1 sqkm √ maximum√distance traveled to traverse = 2km 2 ∴ time = 30 = 169.0) Figure 1: Problem 2 D2 = (j20)2 + (i20)2 − 2(j20)(i20) cos(2π/3) √ ⇒ D = 2a i2 + j 2 + ij = 3R i2 + j 2 + ij 3.69s ∴ number of users increases by 10000 and handoﬀ time reduces by 1/100 2.1) v (0. (a) R=1km D=6km √ 2 3D 1 cluster N = AAcell = 3√3R2/2 = 3 (D/R)2 = 1 62 = 12 3 /2 number of cells per cluster = N = 12 .1) 60 (1. diamond shaped cells.7s In the new setup number of cells = 105 microcells total number of users = 1000 × 1002 users √ 2×10 time = 30×103 =1. R= 100m Dmin = 600m D = 2KR D 600 K = 2R = 2×100 = 3 N = K2 = 9 (a) number of cells per cluster = N = 9 (b) number of channels per cell = total number/N = 450/9 = 50 4. See Fig 1 u (1.Chapter 15 1.
G = 100 ξ=1 λ = 1. interference is reduced by a factor of 3 Nc = 76.167 a2 = 3 1 N > a2 Cu = 50 SIR0 a1 2/γ ⇒ N ≥ 9. γ = 2 BP SK √ Pb = 10−6 → Pb = Q( 2γb ) ⇒ γb = SIR0 = 4. R=10m D=60m γI = 2 γ0 = 4 M = 4 for diamond shaped cells SIRa = R−γI R−2 = = 32400 −γ0 MD 4D−4 R−4 = 324 4D−4 R−2 =9 SIRc = 4D−2 SIRa > SIRb > SIRc SIRb = 6.7534 B = 50M Hz each user100KHz = Bs γ 1 SIR = M D M=6 for hexagonal cells R a1 = 0.4879 ∴ N = 10 7.7349 = 76 .2450 = 26 With sectorization.5 With no sectorization SIR = ξ 3G (Nc 1 − 1)(1 + λ) = 4.SHADED CELLS USE SAME FREQUENCY D Figure 2: Problem 3 (b) number of channels in each cell = 1200/12 = 100 √ (c) i2 + j 2 + ij = 2 3 ⇒ i = 2. j = 2 5.7534 Nc = 26.
247Nc .078Nc (c) Nc = 35 α = 0. j ∈ {1.078Nc ) Po ut = p(SIR < SIR0 ) (a) Pout = p Nc −1 i=1 Nc −1 i=1 G Xi +N G Xi +N < SIR0 =p Nc −1 i=1 Xi +N > G SIR0 Nc −1 (b) X = i=1 Xi then X ∼ Bin(α.^n(i)*(1alpha).247Nc √ x = n 0.. Nc − 1) p(x + N > G/SIR0 ) = Nc −1 p(n + N > G/SIR0 x = n)p(x = n) N =0 p(x = n) = Nc − 1 αn (1 − α)Nc −1−n n Nc −1 p(x + N > G/SIR0 ) = N =0 Nc −1 p(N > G/SIR0 − nx = n)p(x = n) p N =0 Nc −1 = = N =0 N − 0.078Nc + (Nc − 1)α(1 − α)) p(x + N > G/SIR0 ) = Q (e) p= 0.^(Nc1n(i)).078Nc G SIR0 G SIR0 − n − 0.0973 MATLAB for i = 1:length(n) pn(i) = (factorial(Nc1).5 SIR0 = 5 G = 150 p = 0. .247Nc √ > 0.*factorial(Nc1n(i)))). *alpha. sump = sump + . deﬁne γk = nk + ρ gk Pk k=j G SIR0 − (0. K} .8. end sump = 0. for i = 1:length(n) f = ((G/sir0)n(i). end (d) If x can be approximated as Gaussian then x ∼ G((Nc − 1)α.078Nc + (Nc − 1)α(1 − α) gkj pj k. SINR = α = p(Xi = 1) N ∼ G(0.078*Nc)).247Nc + (Nc − 1)α) 0.247Nc √ 0.078Nc p(x = n) Q − n − 0.247*Nc)/(sqrt(.247Nc + (Nc − 1)α..5*erfc((f)/sqrt(2))*pn(i)./(factorial(n(i)). . .0969 (very accurate approximation!) 9. 0. (Nc − 1)α(1 − α)) x + N ∼ G(0. 0.
.. The optimal power control policy is given to be P = (I − F )−1 u 10.2*(D(i)+R/2)^(gamma)+.2*(D(i)R/2)^(gamma)+. +.5*D(i))^2). then a power control policy exists.2*(D(i)+R)^(gamma)). R = 1.2*(D(i)R)^(gamma)+. ASEbest(i) = log(1+Pdes/Pintbest)/(pi*(. K} u= γ nK γ1 n1 γ2 n2 . end . . .. . Matlab D = 2:.. ASE(i) = log(1+Pdes/Pint)/(pi*(. Pdes = R^(gamma). Pintworst = 6*((D(i)R)^(gamma))..5*D(i))^2).where.. for i = 1:length(D) Pint = 6*(. gk is channel power gain from user k to his base station nk is thermal noise power at user k’s base station ρ is interference reduction factor (ρ ∼ 1/G) gkj is channel power gain from j th interfering transmitter to user k’s base station pk is user k’s Tx power pj is user j’s Tx power deﬁne a matrix F such that Fkj = k. j ∈ {1.2*(D(i))^(gamma). Pintbest = 6*((D(i)+R)^(gamma))..01:10. ASEworst(i) = log(1+Pdes/Pintworst)/(pi*(.5*D(i))^2). K g1 g2 gK 0 γk gkj ρ gk k=j k=j If Perron Ferbinius eigenvalue of F is less than 1 . gamma = 2.
SIR = 261. D = 8R .25 0. n = 1. Matlab Pt = 5.2 ASE(D) 0.3 0. number of users that share band = 2(2(n1)+1) ∴ each user gets 100KHz = Bu (n) 2(2n−1) interference is only from ﬁrst tier SIR(n) = N0 2 Bu (n) Pt K + 2 Pt K d0 R 3 3 > 25 √ d0 R2 +D(n)2 using Matlab .0.05 0 2 3 4 5 6 D 7 8 9 10 Figure 3: Problem 10 11.1 0. Pt = 5W B = 100KHz N0 = 10−16 W/Hz Pr = Pt K d0 d 3 d0 = 1. n = 4.35 5 Case ASE Best Case ASE Worst Case ASE 0. R = 1000. K = 100 (a) D=2R 2 users share the band available Each user gets 50KHz R=1Km BASESTATION USER Figure 4: Problem 11a 1 1 (b) P b = 4γ ⇒ 10−3 = 4γ ⇒ γb = 250 b b If D(n) = 2nR.15 0.9253. sigma_2 = 1e16.
sir = Pdes/(Npower+Pint). theta = 0:ss:1. for i = 1:length(alpha) capvec = log2(1+(K*P*(1+2*alpha(i)*cos(2*pi*theta)). end (c) ASE = (R1 +R2 )/B 2km×2km R1 = R2 = Bu (1) log(1 + SIR(1)) Bu (1) = 50KHz. ss = . Pdes = Pt*K*(d0/R)^3.6801bps/Hz/km2 12. For low SNR values (α less than a value) c decreases with increase in α as interference is increased which cannot be easily decoded due to low SNR. sir = Pdes/(Npower+Pint). B=100KHz N0 = 10−9 W/Hz K = 10 P = 10mW per user (a) 0 ≤ α ≤ 1 α is channel gain between cells. K = 10. interference can be decoded and subtracted easily so capacity grows with α as high SNR’s (beyond an α value) . D = 2*n*R.001. K = 100.01:1. Pdes = Pt*K*(d0/R)^3. Pint = 2*(Pt*K*(d0/sqrt(R^2+D^2))^3). Npower = sigma_2*Bu.5899 ASE = 0. SIR(1) = 5. .^2)/(sigma_2*B)). MATLAB CODE: B = 100e3. sigma_2 = 1e9. d0 = 1. alpha = 0:. Npower = sigma_2*Bu. C(i) = (1/K)*sum(capvec)*ss.D = 2*n*R. See Matlab If α is large. Pint = 2*(Pt*K*(d0/sqrt(R^2+D^2))^3). Bu = (100/(2*(2*n1)))*1e3. P = 10e3. K = 100. d0 = 1. Bu = (100/(2*(2*n1)))*1e3. while sir < 250 n = n+1.
4 0. for i = 1:length(K) capvec = log2(1+(K(i)*P*(1+2*alpha*cos(2*pi*theta)).88 0. alpha = .96 0. end B = 100 KHz 6 5 4 C(K)/B 3 2 1 0 0 5 10 15 K 20 25 30 Figure 6: Problem 12b (c) as transmit power P ↑. K = 1:. C(i) = (1/K(i))*sum(capvec)*ss.02 1 0. capacity C ↑ but gets saturated after a while as the system becomes interference limited.86 0. MATLAB CODE: BB = 100e3.84 0. sigma_2 = 1e9. MATLAB CODE: . theta = 0:ss:1.9 1 Figure 5: Problem 12a (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.5.001.7 0. ss = .94 C(α)/B 0.^2)/(sigma_2*B)).1:30.6 0.82 0 0.92 0.5 α 0.end B = 100 KHz 1.3 0.2 0.8 0.98 0. P = 10e3.9 0.1 0.
6 0.B = 100e3. ss = . alpha = . sigma_2 = 1e9. P = [0:.5.8 C(P)/B 0. for i = 1:length(P) capvec = log2(1+(K*P(i)*(1+2*alpha*cos(2*pi*theta)). K = 10.4 0. end B = 100 KHz 1. theta = 0:ss:1.4 1.^2)/(sigma_2*B)).2 1 0.1:100]*1e3.001.2 0 0 10 20 30 40 50 P (in mW) 60 70 80 90 100 Figure 7: Problem 12c . C(i) = (1/K)*sum(capvec)*ss.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.