You are on page 1of 10

Homework #5

1. Write a function generating the Jakes time-varying flat Rayleigh fading channel coef-
ficients assuming bit rate Rb = 12.2Kbps, vehicle velocity v = 120Km/h, and carrier
frequency f0 = 2.115GHz. When sampling the process, start at time t = 500Tb .

(a). Plot the gain (dB) of channel coefficients over 2000 bit interval.
(b). Evaluate E{|c(t)|2 }.
(c). Numerically evaluate E{c(t)c∗ (t + ∆t)} and plot it as a function of ∆t.
(d). Plot J0 (2πfd ∆t) as a function of ∆t.

Note:

• You should write a Matlab function (we need to use it later). Pass v, fc , Tb to
the function. The function should return a vector, which contains the channel
coefficients corresponding to each bit interval. We have to assume that channel
is constant over one bit interval.
• Assume the maximum Doppler shift
• J0 (·) is the zeroth-order Bessel function of the first kind
• To generate J0 (2πfd ∆t) in Matlab, you may use function “besselj(0, 2πfd ∆t)”.
• After you get the function working properly, change velocity v and observe the
channel variations for each case.

1
ECE669 52 ECE669 53

Wireless channels (continued) Wireless channels (continued)

Fading channel simulators

• Flat (frequency nonselective) Rayleigh channels:


 Fading channel process c(t) is modeled as a
normalized, zero-mean complex-valued
wide-sense stationary Gaussian process.

ECE669 54 ECE669 55

Wireless channels (continued) Wireless channels (continued)

 Spaced-time correlation function Φ(∆t) • Continue time-varying function c(t) is usually


approximated by piece-wise constants.
Φ(∆t) = E{c(t)c∗(t + ∆t)},
• How channel changes from one interval to the other
is determined by the spaced-time correlation
∗ E{·}: statistical expectation. function.
∗ ∗: complex conjugate.
∗ Φ(0) = 1: due to the normalization of the channel.
ECE669 56 ECE669 57

Wireless channels (continued) Wireless channels (continued)

• Jakes’ model: • The channel fading process whose spaced-time


correlation is modeled by Jakes’ model can be
Φ(∆t) = J0(2πfd∆t) generated by
C0
c(t) = √ [Xc(t) + jXs(t)]
2S0 + 1
 fd: the maximum Doppler shift of the channel. S0

 J0(·): the zeroth order Bessel function of the first Xc(t) = 2 [cos(φn)cos(ωnt)] + 2cos(φN )cos(ωmt)
kind. n=1


S0

Xs(t) = 2 [sin(φn)cos(ωnt)] + 2sin(φN )cos(ωmt)
n=1

ECE669 58 ECE669 59

Wireless channels (continued) Wireless channels (continued)

 S0: the total number of sinusoids (Jakes suggests


• Relationship between S and S0:
that S0 = 8 sinusoids will give a pretty good
approximation)  S = 2(2S0 + 1) (we use S0 = 16 sinusoids for our
 Other parameters: simulations).
 φN = 0, φn = πn/(S0 + 1).
ωm = 2πfd,  C0 = 1 for normalization.
ωn = ωmcos(2πn/S).
• The corresponding statistics of Xc(t), Xs(t) and c(t):

E{Xc2(t)} = S0,
E{Xs2(t)} = S0 + 1,
E{|c(t)|2} = C0 = 1.
16

In Section II.D, we also use the same analysis result to study several
important special cases to provide insights and explore connections with well known
existing results. The following special cases are considered: quasi-static Rayleigh
fading channels, fast Rician fading channels and fully correlated fast Rayleigh fading
channels. In particular, the role of channel estimation errors, antenna correlation
and user signal correlation on detector performance are quanti ed. In addition,
we quantify the e ect of the number of training symbols on the performance of
our MAP receiver with a Kalman lter. Finally, we compare our MAP receiver
with two other adaptive receivers, the adaptive channel predictor and the adaptive
MMSE combiner.
Finally, in Section II.E, we brie y summarize the results.

II.A Signal Model


In this section, we rst discuss in detail how the fading process is mod-
eled and generated in simulation. Then we establish the digital signal model used
throughout this chapter. We end this section with a short discussion on the inter-
ference and a summary of the notations used.

II.A.1 Fading Processes


Let x(t) be the complex fading process for the desired user. Experiments
indicate that the complex fading coecient x(t) is a random quantity that changes
slowly over time [64]. So the mathematical nature of x(t) is a narrowband random
process which has correlation over time. In the case of Rayleigh fading, x(t) is
a complex Gaussian narrowband process, which can be modeled as the output of
a low pass lter excited by temporally white complex Gaussian noise. The low
pass lter is often referred as the shaping lter, because it determines the power
spectrum shape and the temporal correlation function of the fading process. In
the most widely used Jakes' model, x(t) is assumed to have the following temporal
17

correlation function,

E fx(t)xH (t ,  )g = J0(2Fd ); (II.1)

where J0(?) is the zeroth Bessel function of the rst kind and Fd is the physical
Doppler frequency 2. The corresponding power spectrum is X (f ) = q 1 f 2 ; jf j 
1 , ( Fd )
Fd . In current systems, typical Fd ranges from 5Hz to 300Hz, depending on the
speci c situation. For example, for a carrier frequency fc of 2GHz and a mobile
speed v of 30 m/sec (68 mile/hour), Fd = vfc c = 30 32 10  109 = 200Hz. If F is
8 d
bigger than 100Hz, it is often referred to as \fast fading". A common method to
generate Jakes' x(t) is to sum up several sinusoids [66, 67], as rst suggested by
Jakes [64]. However, this method in fact generates a deterministic process [68] and
the temporal correlation property of such a process is brought into question in [69].
In the following, we will discuss how to generate the fading process for simulation
with a digital receiver.
In the digital receiver we still use parameters in the above example. Now
suppose that the symbol duration is limited to Ts. If the baud rate Rbaud (= 1 )
Ts
is 40k/per second, then the fading rate normalized to data rate is fd = R d = F
baud
200 = 1 . Roughly speaking, the channel does not change much over 200
40  103 200
symbols. Since the fading is so slow at the symbol level, we can often neglect the
change of the fading process over one symbol duration Ts and assume that the
fading process remains constant over a symbol, i.e., x(t) = x(nTs ) = xn , for nTs 
t < (n + 1)Ts. According to Eq.(II.1), the correlation function is E fxnxHn,m g =
J0(2fd m), where fd = RFd = Fd Ts is the normalized Doppler frequency. The
baud
power spectrum of xn is x(f ) = q 1 f 2 ; jf j  fd.
1 , ( fd )
To generate xn, a straightforward way is to pass white complex gaussian
2
In fact, Doppler frequency alone does not account for all the dynamics of the channel. For example,
xed wireless links also slowly change over time due to the movement of nearby re ectors such as tree
leaves and pedestrians [65].
18

noise un through the following lter,


q
h(f ) = x(f ) = s 1 ; jf j  fd; (II.2)
4 f
1 , (f ) 2
d
which is the square root of the power spectrum of xn . This lter is highly nonlinear
and approximation has to be sought. Since the lter in Eq.(II.2) has in nite impulse
response, it is natural to try to nd an IIR lter to approximate this lter. In [70],
a third order AR lter is shown to t the spectrum above accurately. More recently,
32 order IIR lters have been developed in the wireless industry to generate the
fading process [71]. Throughout this thesis, we will use a general AR lter to
approximate an arbitrary shaping lter, i.e., our shaping lter is

h(f ) = el(,j2f ) ; (II.3)


X
l,1
1 , ie(l,i)(,j2f )
i=0
where (l , 1) is the order of the lter and i ; i = 0; : : : ; l , 1 are the coecients of
the lter. In the temporal domain, the lter is
X
l,1
xn+l = i xn+i + un; (II.4)
i=0
where un is a complex white Gaussian noise process. One advantage of this approach
is that eld data can be t into this parametric model. In fact, the correlation in
Eq.(II.1) is derived from mathematical models rather than measured from real
data. It is well possible that di erent correlation functions may arise when the
re ecting environment deviates from the assumed mathematical model. In order
to get optimal tracking performance, the receiver requires knowledge of temporal
variation characteristics of the channel such as Eq.(II.2) or Eq.(II.3). Therefore,
modeling of the channel is of much interest and an active research area (see [72] and
references therein). To retain the exibility of the analysis and to accommodate
various channel models, we will assume the order and coecients of the lter to be
known. In simulations in this thesis, the temporal correlation function is speci ed
19

using Jakes' model as shown by Eq.(II.1). Then the coecients of the shaping lter
are calculated using the Yule-Walker equation [27, 73].
Stacking Eq.(II.4) for m antennas in vector form, we have the following
model to characterize the dynamics of the vector channel response,
X
l,1
Xn+l = Fi Xn+i + Un;
i=0
where Fi ; i = 0; : : : ; l , 1 are m  m diagonal matrices whose diagonal entries are the
coecients of m shaping lters of order (l , 1). Un is a temporally white Gaussian
noise vector, whose elements are input to m shaping lters. Q1 = E fUnUHn g can
be non-diagonal, corresponding to correlated fading channels. We point out that in
its most general form, the spatial-temporal correlation function is a function of both
temporal interval  and distance d, i.e., E fx(t; s)xH (t,; s,dg = R(; d). However,
a common engineering practice is to assume that this correlation function can be
decoupled as the product of the temporal correlation and the spatial correlation [74,
75]. That is, R(; d) = R( )R(d). Though not necessarily true, this assumption can
reduce the complexity of the problem in many cases. The model above also implies
this assumption. Finally, we note that all the correlation functions involved are the
correlation of the complex fading coecients, not that of the fading amplitude. For
transformation between these two quantities, see [76].
Note that this model can accommodate even more general scenarios than
mentioned above. For example, by making Fi time dependent, i.e., Fi(n), we can
have a time-varying model. Another possible extension is that diagonal elements
in Fi are not necessarily identical, so the e ect of possible di erent fading modes
can be incorporated.

II.A.2 Vector Signal Model


Suppose that we use (t) as waveform for a BPSK signal of interest, where
R
(t) has support over [0,T]. Furthermore, 0T 2(t)dt = 1.
We assume Rayleigh at fading channels. So the received signal at a single
Rayleigh Faded Carrier Envelope

z The envelope of a received


Raylei gh Faded Ca rrie r Enve lope (dB)
fd = 5Hz
10

5 multi-path signal will typically


have a Rayleigh distribution
Le ve l above m ea n (dB)

-5

-10

-15

-20 z Typical time varying


-25
amplitude and phase of a
-30
0.0 1 .0 2.0
Tim e (s )
3.0 4.0 5.0
Rayleigh faded channel are
shown in the two graphs
Ra yleigh Faded Carrier Phase
fd = 5Hz

6.5
6.0
5.5
z Doppler Frequency (Fd)
5.0
refers to Doppler shifts of
Phase Angle ( rad)

4.5
4.0
3.5
3.0
2.5
different components of the
2.0
1.5
1.0
receive signal, if either the
0.5
0.0
0.00 1.00 2.00 3.00 4.00 5.00
transmitter, receiver or
Time ( s) reflecting objects are in
9
motion
Rayleigh Fading Simulator (Jakes' Model)

z Weighted sum of 9 sinusoids


z Complex output form to
model amplitude and phase
change
z Implemented as ROM look-
up and scaling factor
multiplier
z Frequency adjusted using
Phase Accumulator
z Doppler frequency (Fd)
selected at 5Hz (worst case)
for indoor channel modeling

10

You might also like