2
7
3
1
4
5
6
2
7
1
5
2
7
6
2
7
4
5
3
3
1
1
6
7
4
7
4
5
3
2
7
n
o
BP-BB transformation
(quadrature demodulation)
sI (t)
sBB (t)
2cos(2f0 t)
x1 (t)
sBP (t)
2sin(2f0 t)
sQ (t)
sBP (t)
Lowpass
filter
sI (t)
2cos(2f0 t)
sBB (t)
2sin(2f0 t)
x2 (t)
Lowpass
filter
sQ (t)
x
1+x1+
for x 1 .
2
b) Determine the resulting delay difference = 2 1 between the two paths.
c) In the case of free-space propagation, the received power is given by
2
0
.
PR = PT GBS GM S
4d
Under this assumption, derive an analytical expression for the signal rRF (t)
received at the MS.
d) Determine the equivalent complex baseband signals s(t) and r(t) that correspond
to the RF-signals sRF (t) and rRF (t).
Hints: 2 cos(x) cos(y) = cos(x + y) + cos(x y) and
2 sin(x) cos(y) = sin(x + y) + sin(x y).
e) Calculate the transmitted and the received power.
Hint: The power of a given RF-signal is the same as the power of its equivalent
complex baseband signal.
f) Determine an analytical expression for the power ratio PR /PT as a function of
the horizontal distance d.
g) For very large d, the expression for PR /PT can be approximated using the relation
ex 1 + x
for |x| 1 .
Determine the power ratio for large and for very large horizontal distances d.
00
GBS 11
00
11
11
00
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
00
11
1
0
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
hBS
d1
(LO
S
d2
pat
h)
1
0
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
GMS
= 1
000000000000000000000000
111111111111111111111111
1
0
0
1
0
1
0
1
0
1
hMS
0
1
0
1
0
1
0
1
0
1
d1
(hBS + hMS )
(hBS
d2
hBS
hMS )
hMS
hMS
d
50
60
70
80
90
hBS=30 m
hMS=1.9 m
GBS=10 dBi
GMS= 0 dBi
100
110
120
10
100
1000
Horizontal distance between the antennas in meter
10000
d
l
The size of the overall (rectangular) coverage area is given by the length l, while the
size of each cell is determined by the distance d between neighbouring base stations
(BSs). A mobile station (MS) in a certain cell can be served by the corresponding
BS if the received power PR is larger than a minimum power PR,min . In general, the
received power can be expressed as a function of the distance D between the MS and
BS according to
0
,
PR = PT GBS GM S
4D
where is called propagation exponent. The antenna gains GBS and GM S are assumed
to be one in this exercise. Hence, they can be neglected subsequently.
BS
a) Calculate the minimum required transmit power PT,min
for a single BS as a
function of d and . Consider for your calculations that the condition PR PR,min
must be fulfilled at each point in a cell.
NW
b) Based on the result in a), calculate the minimum possible network power PT,min
as a function of d and .
Distance
City of Newark,
L = ... (refer to the diagram on next page)
: d = 10 km
: PT = 0.5 W
: GT = GM S = 3 dBi
:
:
:
f0 = 900 MHz
T = 3.7
s
Es /N0 req 1.5 dB
Boltzmanns constant
: k = 228.6 dB Ws/K
a) Calculate the available Es /N0 avail in dB.
90
100
PR/PT in dB
110
120
130
Lineofsight
Rural area
Suburban area
Newark
Philadelphia
Tokyo
140
150
160
10
Horizontal distance d in km
6. Channel Modeling
Optimization/comparison of digital communication systems or of certain system
components normally requires many computer simulations. In order to predict the
performance of a system in a particular application, we need to emulate the physical
channel on a digital computer according to an appropriate channel model.
Transmitter
Channel
Receiver
Concerning a particular wireless radio channel, the effects of propagation path loss
and shadowing (which causes long-term fading) are usually investigated by extensive
measurements, while the effects of multipath propagation and Doppler frequency shift
can be well emulated on a digital computer. Therefore, the task of channel modeling is
mainly to describe the short-term fading process of a wireless radio channel. For this
purpose, the wide-sense-stationary uncorrelated-scattering (WSSUS) channel model is
suitable and widely used.
In this exercise, we want to design a Rice fading channel emulator which may be easily
implemented on a digital computer.
s(t)
r(t)
f (t)
f1 = KK+1
LOS path
n(t)
f2(t)
K : Ri e fa tor
K +1
f~(t)
S
attered
omponents
(Rayleigh pro
ess)
E ff~(t)g = 0
E fjf~(t)j2g = 1
a) Given the above block diagram, determine the average power of the LOScomponent and of the scattered components. What is the average power of the
Rice fading channel?
Since the LOS component has a fixed value determined by the Rice factor, the main task
of the channel emulator is to generate the scattered components with an appropriate
Doppler power density spectrum. In the following, we assume that the power of the
scattered components has a 2-D isotropic distribution with respect to the receive
antenna. The Doppler power density spectrum of the scattered components can be
described by Jakes spectrum:
S(fD )
S(fD ) =
fDmax
-fDmax
fDmax
fD
s 1
fDmax 1 f
2
fD
if |fD | fDmax
Dmax
otherwise.
Suppose that we want to emulate the scattered components by using a low-pass filter as
shown in the following figure, where w(t) is a complex-valued zero-mean white Gaussian
noise process.
w(t)
f(t)
H(fD )
LP
b) How should the frequency response of the low-pass filter look like?
c) Is f(t) also a zero-mean white Gaussian random process?
d) Which difficulty may arise when implementing such a channel emulator?
Instead of using a low-pass filter, the so-called Jakes Model is popularly applied to
produce a Rayleigh fading process with desired Doppler power density spectrum.
To start with, let us first consider the effect of Doppler shift on a baseband signal in
a single propagation path. Suppose that a baseband signal with a constant value 1 is
transmitted over the following link. We assume that there is no propagation path loss.
Quadrature modulation is used, and the carrier frequency is given by fc .
Tx
Rx
e) Please give the equivalent complex baseband representation of the received signal.
f) What is the relationship between the transmitted and received baseband signal?
10
11
Pulse
Shaping
gT x (t)
s(t)
r(t)
Matched
Filter
y(t)
kT
y[k]
gRx (t)
We assume that E{|x[k]|2 } = 1 and that the pulse shaping filter is given by
(
A, 0 t T,
gT x (t) = A rect(t) =
0, elsewhere,
where A is an arbitrary constant and T denotes the symbol duration.
a) How would you design the matched filter? Please give a formula for gRx (t) and
sketch it with necessary labels.
b) Sketch the waveform of gT x (t) gRx (t). ( denotes convolution)
c) After sampling, the output signal y[k] is a superposition of the useful signal and
the noise sample, and, hence, may be written as y[k] = ys [k] + yn [k]. Please give
a mathematical expression for ys [k] and yn [k], respectively.
d) What is the principle of integrate & dump?
e) Let Es denote the average energy used for transmitting one symbol, what is the
relationship between Es and the average power of ys [k]?
f) Given that the one-sided power spectral density of n(t) is N0 , please calculate
the average power of yn [k] and the resulting signal-to-noise ratio.
Now, let us consider a bandpass transmission with quadrature modulation over an
AWGN channel. We assume that the pulse shaping filters for the inphase part and the
quadrature part both have the impulse response given by gT x (t), and the one-sided
noise spectral density within the signal bandwidth is given by N0 .
g) Please provide an appropriate diagram for the transmission model. (cf. Exer. 1)
h) Please calculate the SNR after matched filtering and symbol-rate sampling, and
express it in terms of Es and N0 .
i) Given the above results, please derive an equivalent discrete-time channel model.
How should we set the noise power?
12
Im
d2
d1
d0
Re
d3
with X = {d0 , d1 , d2 , d3 }.
g) Compare 2-PSK and 4-PSK with regard to bandwidth and power efficiency.
13
2
Re(d0 g(t))
Im(d0 g(t))
2
2
2
2
t/T
2
Re(d1 g(tT))
Im(d1 g(tT))
2
2
2
2
t/T
2
Re(d3 g(t2T))
Im(d3 g(t2T))
2
2
t/T
t/T
2
Re(s(t))
Im(s(t))
2
2
t/T
2
2
t/T
2
t/T
2
2
t/T
14
Im
Im
d2
d2
d0
d1
d0
d1
Re
Re
d3
d3
Im
d2
d0
d1
Re
d3
15
9. Power-Bandwidth Diagram
We consider the power-bandwidth diagram introduced in the lecture.
a) Why is the modulation scheme 4-PSK so popular in wireless communication
systems?
b) Add to the power-bandwidth diagram below the points resulting for M -ASK
(refer to the diagram on page 16). Assess the efficiency of M -ASK.
c) Suppose, you have the task to develop a transmission system for a space probe.
Which modulation scheme would you choose?
d) Which modulation scheme would you choose for HDTV (high-definition television)?
e) Consider an OFDM system which modulates 1/3 of all carriers with 64-QAM,
1/3 with 16-QAM, and 1/3 with 4-QAM. Determine the bandwidth efficiency of
such a system. (OFDM: orthogonal frequency division multiplexing)
10
Shannon
bound
16QAM
64QAM
32PSK
16PSK
8PSK
4QAM
= 4PSK
10
2PSK
8DPSK
4DPSK
2DPSK
Ps = 10
10
5.0
0.0
5.0
10.0
Eb/N0 in dB
15.0
20.0
25.0
16
10
-1
10
-2
10
-3
10
-4
10
-5
10
M=2
M=4
M=8
M=16
M=32
M=64
-6
10
-7
10
-8
10
-9
10
-10
10
-10.0
-5.0
0.0
5.0
10.0
15.0
Eb/N0 in dB
20.0
25.0
30.0
17
Impulse
Modulator
1
g (t)
T 0
rect(t/T )
y0 (t)
n(t)
exp(j2t/T )
s(t)
y0 [k]
kT
Detector
a
0 [k]
r(t)
Impulse
Modulator
1
g (t)
T 1
rect(t/T )
y1 (t)
y1 [k]
kT
Detector
a
1 [k]
exp(j2t/T )
Both and are non-negative constants, and the rectangular pulse is given by
(
1 0 t T,
rect(t/T ) =
0 elsewhere.
a) How should we choose and so that the two baseband pulses g0 (t) and g1 (t)
are orthogonal?
b) Given that g0 (t) and g1 (t) are orthogonal, are we able to recover a0 [k] and a1 [k]
with two parallel matched filters as shown above?
For wideband data transmission, the wireless radio channel often exhibits a frequencyselective property, which in turn causes intersymbol interferences (ISI). In order to
circumvent the task of equalization, OFDM can be used to convert a wideband ISI
channel into a set of parallel narrowband ISI-free sub-channels.
In the following, we consider an OFDM transmission system depicted on page 20.
Suppose that we want to transmit four BPSK symbols
a = [+1, 1, +1, +1],
and the OFDM block length N (number of subcarriers) is set to be 4.
c) Calculate d[k] (the sequence after IDFT), k = 0, 1, . . . , N 1.
d) What is the relationship between IDFT and the orthogonal multi-carrier
modulator as shown in the above picture?
18
The pulse shaping filter, the physical channel, the matched filter and the (symbol-rate)
sampler can be altogether represented by an equivalent discrete-time ISI channel with
L memory units (as shown on page 20). From now on, we assume that L = 2 and the
channel impulse response is given by
.
h = [h0 , h1 , h2 ] = [1, 0.5, 0.5j].
In order to avoid ISI, a cyclic prefix of length L is added to each OFDM symbol block.
e) How does the transmitted symbol sequence x[k] look like?
f) Assuming that the noise power is negligible, calculate the channel output y[k].
What results after removing the samples corresponding to the cyclic prefix?
g) Calculate c[n] (the sequence after DFT), n = 0, 1, . . . , N 1.
h) What is the relationship between DFT and the matched filter bank of the
multicarrier transmission system shown in last page?
IDFT:
19
k = 0, 1, . . . , N 1
g0 [k] = rect[k/N ]
d[0]
rect[k/N ]
1
g1 [k] = ej2k/N rect[k/N ]
d[1]
rect[k/N ]
d[k]
j2k/N
1/N
1]
d[N
DFT:
=
d[n]
N
1
X
d[k]ej2nk/N ,
k=0
g0 [k] = rect[k/N ]
n = 0, 1, . . . , N 1
k=0
d[0]
k=0
d[1]
k=0
1]
d[N
rect[k/N ]
1
g1 [k] = ej2k/N rect[k/N ]
rect[k/N ]
d[k]
ej2k/N
j2(N 1)k/N
gN
rect[k/N ]
1 [k] = e
rect[k/N ]
ej2(N 1)k/N
y[N 1]
Remove
Prefix
y[N 1]
y[L]
y[L 1]
y[0]
N = N + L
d[N 1]
d[0]
P/S
P/S
y[k]
x[k]
kT
Matched
Filtering
RF
RF
Pulse
Shaping
z 1
hL
c[N 1]
y[L]
c[0]
DFT
(FFT)
d[N 1]
1]
d[N
IDFT
(IFFT)
Cyclic
Prefix
h0
z 1
h1
S/P
S/P
d[0]
x[k]
a
[k]
a[k]
d[0]
d[N 1]
d[N L]
n[k]
y[k]
0
h
21
n
[0]
d[0]
c[0]
1
h
n
[1]
d[1]
c[1]
N 1
h
1]
d[N
n
[N 1]
c[N 1]
Matched
Filtering
y[N 1]
P/S
y[L]
c[0]
y[N 1]
d[N 1]
1]
d[N
c[N 1]
DFT
(FFT)
S/P
a
[k]
a[k]
S/P
d[0]
IDFT
(IFFT)
d[0]
Cyclic
Prefix
Remove
Prefix
y[L]
y[L 1]
y[0]
N = N + L
d[N 1]
d[0]
d[N 1]
d[N L]
P/S
x[k]
y[k]
kT
RF
RF
Pulse
Shaping
22
User n
0
1
2
3
4
5
6
23
a) Signal of user 0
a0 = [+1, 1, 1]
+1
a0 [k] rect(t kT )
t
T
2T
3T
Tc
2T
3T
Tc
2T
3T
Tc
1
t
1
s0 (t)
+1
t
1
24
b) Signal of user 1
a1 = [+1, 1, +1]
+1
a1 [k] rect(t kT )
t
T
2T
3T
Tc
2T
3T
Tc
2T
3T
Tc
1
t
1
s1 (t)
+1
t
1
25
t
Tc
2T
3T
Tc
2T
3T
2T
3T
s1 (t)
+1
t
1
+1
t
Tc
1
2
r(t)
s(t)
n(t)
1
T
1
T
g0 (T
1
T
g1 (T
t)
t)
gN
1 (T t)
y0 (t)
y1 (t)
yN 1 (t)
kT
kT
kT
y0 [k]
y1 [k]
yN 1 [k]
Detector
a
0 [k]
Detector
a
1 [k]
Detector
a
N 1 [k]
26
n1[k]
h1[k]
y1[k]
x[k]
h2[k]
n2[k]
y2[k]
x[k] is a BPSK modulated data symbol at time index k. h1 [k] and h2 [k] are
statistically independent Rayleigh fading channel coefficients with average channel gain
E{|h1 [k]|2 } = E{|h2 [k]|2 } = 1. n1 [k] and n2 [k] are statistically independent zero-mean
white complex Gaussian noise samples with identical variance n2 .
a) Give an expression for the channel outputs y1 [k] and y2 [k], respectively.
b) Calculate the average signal-to-noise ratio (SNR) at each receiver branch and
also the overall average SNR.
From now on, we omit the time index k in all expressions for the sake of simplicity.
c) Assuming that all channel coefficients are perfectly known at the receiver, how
should we detect the data symbol x according to the maximum-likelihood (ML)
criterion?
Let us suppose that the channel coefficients are h1 = 0.5, h2 = j, and the two received
signal samples are given by y1 = 0.6j, y2 = 0.4 0.5j.
d) Perform ML detection for the data symbol x.
e) What results if a maximum ratio combining (MRC) receiver is used?
f) Is an MRC receiver optimal in the sense of the ML criterion? Give reasons.
g) In which ways do the multiple receive antennas help improve the system
performance?
27
Now let us consider the following multiple-input single-output (MISO) system which
adopts the famous Alamouti transmission scheme.
x1, x2
h1
n1 , n2
x2, x1
y1 , y2
h2
x1 and x2 are two BPSK modulated data symbols to be transmitted. h1 and h2 are
statistically independent Rayleigh fading channel coefficients with average channel
gain E{|h1 |2 } = E{|h2 |2 } = 1, and are perfectly known at the receiver. n1 and n2
are statistically independent zero-mean white complex Gaussian noise samples with
identical variance n2 .
h) Give an expression for the channel outputs y1 and y2 , respectively.
i) Calculate the average signal-to-noise ratio (SNR) at the receiver side.
j) If a hard decision is to be made, state a formula for the ML estimate of the
symbol pair (x1 , x2 ).
Let us suppose that the channel coefficients are h1 = 0.5, h2 = j, and the two received
signal samples are given by y1 = 0.5j, y2 = 0.5 j.
k) Perform ML detection for the symbol pair (x1 , x2 ) by means of exhaustive search.
l) If we collect the transmitted data symbols in a vector x = [ x1 , x2 ]T , the channel
outputs can be concisely written as
y = Hx + n.
Find an appropriate expression for y, H and n, respectively.
m) Show that by computing z = HH y the transmitted symbol pair (x1 , x2 ) can
be easily estimated without performing exhaustive search. Is the resulting hard
decision optimal in the sense of ML criterion?
n) Estimate the symbol pair (x1 , x2 ) by the above mentioned detection scheme.
o) If the Alamouti transmission scheme is used, what is the relationship between
the energy per info bit (Eb ) and the energy per symbol (Es )? (BPSK assumed)
p) Will you expect the same performance from the two systems (discussed in this
exercise) in the sense of BER vs. Es /N0 ? How about BER vs. Eb /N0 ?
(Hint: ML detection assumed)
28
10
|h1[k]|2
10
10
10
10
100
200
300
k (time index)
400
500
400
500
400
500
10
|h2[k]|2
10
10
10
10
100
200
300
k (time index)
|h1[k]|2 + |h2[k]|2
10
10
10
10
100
200
300
k (time index)
29
10
1x1: ML detection
1x2: ML detection
2x1: Alamouti
10
10
10
10
15
Es/N0 in dB
20
25
10
1x1: ML detection
1x2: ML detection
2x1: Alamouti
10
10
10
10
15
Eb/N0 in dB
20
25
30
Training
Data
+1 +1
Training (midamble)
+1
1 +1 +1 +1
Data
1 +1 +1
Training
+1 +1
a) At first, let us consider the training sequence [+1 + 1 + 1 1]. Determine the
cyclic auto-correlation function (ACF) of the sequence.
b) Now, let us return to the above burst structure. How much overhead is
introduced? How has the midamble been constructed? What is the function of
the midamble? What is the function of the first two training bits and the last
two training bits?
c) Assume a binary modulation scheme (a[k] {+1, 1}) and a time-invariant
channel. Let the received sequence be
[+0.9, +1.6, 0.5, 1.0, 0.1, +1.6, +1.4, 0.2, +0.2, +1.4,
+ 1.4, 0.2, +0.2, +1.4, +1.8, 0.2, 1.5, +0.3, +1.0, +1.3].
d) How many trellis states are required for maximum-likelihood sequence estimation
(MLSE)?
e) Calculate the most probable data sequence by means of the Viterbi algorithm.