Course Hero has millions of student submitted documents similar to the one

below including study guides, practice problems, reference materials, practice exams, textbook help and tutor support.

Course Hero has millions of student submitted documents similar to the one below including study guides, practice problems, reference materials, practice exams, textbook help and tutor support.

1 Chapter Solutions to Exercises
1
Chapter 2 Solutions to Exercises
Exercise 2.1. 1. Let r(t) =
2 r0 + (vt)2 + 2r0 vt cos(). Then,
Er (f, t, r(t), , )) =
[(, , f ) exp{j 2f (1 r(t)/c)}] . r(t)
Moreover, if we assume that r0 v t, then we get that r(t) r0 + vt cos(). Thus, the doppler shift is f v cos()/c. 2. Let (x, y, z ) be the position of the mobile in Cartesian coordinates, and (r, , ) the position in polar coordinates. Then (x, y, z ) = (r sin cos , r sin sin , r cos ) (r, , ) = x2 + y 2 + z 2 , arctan(y/x), arccos(z/ x2 + y 2 + z 2 ) xy xy = 2 + y2 x zr z r = 2 r 1 (z/r)2 We see that is small for large x2 + y 2 . Also is small for |z/r| < 1 and r large. If |r/z | = 1 then = 0 or = and v <= r|| so v/r large assures that is small. If r is not very large then the variation of and may not be negligible within the time scale of interest even for moderate speeds v. Here large depends on the time scale of interest. Exercise 2.2. Er (f, t) = cos [2f (t r(t)/c)] 2 [d r(t)] cos [2f (t r(t)/c)] + 2d r(t) r(t)[2d r(t)] cos [2f (t + (r(t) 2d)/c)] 2d r(t) 2
Tse and Viswanath: Fundamentals of Wireless Communication =
3
2 sin [2f (t d/c)] sin [2f (r(t) d) /c] 2 [d r(t)] cos [2f (t r(t)/c)] + 2d r(t) r(t)[2d r(t)] (2.1)
where we applied the identity cos x cos y = 2 sin x+y 2 sin yx 2
We observe that the rst term of (2.1) is similar in form to equation (2.13) in the notes. The second term of (2.1) goes to 0 as r(t) d and is due to the dierence in propagation losses in the 2 paths. Exercise 2.3. If the wall is on the other side, both components arrive at the mobile from the left and experience the same Doppler shift. Er (f, t) = [ exp{j 2 [f (1 v/c)t f r0 /c]}] [ exp{j 2 [f (1 v/c)t f (r0 + 2d)/c]}] r0 + vt r0 + 2d + vt
We have the interaction of 2 sinusoidal waves of the same frequency and dierent amplitude. Over time, we observe the composition of these 2 waves into a single sinusoidal signal of frequency f (1 v/c) and constant amplitude that depends on the attenuations (r0 + vt) and (r0 + 2d + vt) and also on the phase dierence f 2d/c. Over frequency, we observe that when f 2d/c is an integer both waves interfere destructively resulting in a small received signal. When f 2d/c = (2k + 1)/2, k Z these waves interfere constructively resulting in a larger received signal. So when f is varied by c/4d the amplitude of the received signal varies from a minimum to a maximum. The variation over frequency is similar in nature to that of section 2.1.3, but since the delay spread is dierent the coherence bandwidth is also dierent. However there is no variation over time because the Doppler spread is zero. Exercise 2.4. 1. i) With the given information we can compute the Doppler spread: Ds = |f1 f2 | = fv | cos 1 cos 2 | c from which we can compute the coherence time Tc = c 1 = 4Ds 4f v | cos 1 cos 2 |
ii) There is not enough information to compute the coherence bandwidth, as it depends on the delay spread which is not given. We would need to know the dierence in path length to compute the delay spread Td and use it to compute Wc .
Tse and Viswanath: Fundamentals of Wireless Communication
4
2. From part 1 we see that a larger angular range results in larger delay spread and smaller coherence time. Then, in the richly scattered environment the channel would show a smaller coherence time than in the environment where the reectors are clustered in a small angular range. Exercise 2.5. 1. (hs hr )2 ) 2r 2 (hs + hr )2 r2 + (hs + hr )2 = r 1 + (hs + hr )2 /r2 r(1 + ) = 2r2 (hs + hr )2 (hs hr )2 h2 + h2 + 2hs hr h2 h2 + 2hs hr r s r =s 2r 2r 2hs hr = r r2 + (hs hr )2 = r 1 + (hs hr )2 /r2 r(1 +
r1 = r2 r2 r1
Therefore b = 2hs hr . 2. Er (f, t) = = = Re[[exp{j 2 (f t f r1 /c)] exp{j 2 (f t f r2 /c)]] r1 Re[[exp{j 2 (f t f r1 /c)][1 exp(j 2f (r1 r2 )/c)] r1 Re[[exp{j 2 (f t f r1 /c)][1 exp(j 2f /c b/r)] r1 Re[[exp{j 2 (f t f r1 /c)][1 (1 j 2f /c b/r)] r1 2f ||b [j exp(j ) exp[j 2 (f t f r1 /c)]] cr2 2f ||b sin[2 (f t f r1 /c) + ]] cr2
Therefore = 2f ||b/c. 3. 1 1 1 1 = = r2 r1 + (r2 r1 ) r1 [1 + (r2 r1 )/r1 ] r1 1 r2 r1 r1 1 r1 1 b 2 r1
Therefore if we dont make the approximation of b) we get another term in the expansion that decays as r3 . This term is negligible for large enough r as compared to /r2 .
Tse and Viswanath: Fundamentals of Wireless Communication
5
Exercise 2.6. 1. Let f2 be the probability density of the distance from the origin at which the photon is absorbed by exactly the 2nd obstacle that it hits. Let x be the location of the rst obstacle, then f2 (r) = P {photon absorbed by 2nd obstacle at r} =
x
P {absorbed by 2nd obstacle at r | not absorbed by 1st obstacle at x} P {not absorbed by 1st obstacle at x} dx
Since the obstacle are distributed according to poisson process which has memoryless distances between consecutive points, the rst term inside the integral is f1 (r x). The second term is the probability that the rst obstacle is at x and the photon is not absorbed by it. Thus, it is given by (1 )q (x). Thus,
f2 (r) =
x=
(1 )q (x)f1 (r x)dx
2. Similarly, we observe that fk+1 (r) is given by fk+1 (r) =
x
P {absorbed by (k + 1)th obst at r | not absorbed by 1st obst at x} P {not absorbed by 1st obstacle at x} dx
=
x=
(1 )q (x)fk (r x)dx
(2.2)
3. Summing up (2.2) for k = 1 to , we get:
fk (r) =
k=2 x=
(1 )q (x)
k=1
fk (r x) dx
Tse and Viswanath: Fundamentals of Wireless Communication Thus,
6
f (r) f1 (r) =
x=
(1 )q (x)f (r x)dx,
or equivalently,
f (r) = q (r) +
x=
(1 )q (x)f (r x)dx
(2.3)
4. Using (2.3), we get that F ( ) = (1 )Q( ) + F ( )Q( ), (2.4)
where F and Q denote the Fourier transform of f and q respectively. Since the q (x) is known explicitly, its Fourier transform can be directly calculated and it turns out to be: Q( ) = Substituting thin in (2.4), we get F ( ) = 2 . 2 + 2 2 . 2 + 2
Thus, F is of the same form as Q, except for a dierent parameter . Thus, |r| f (r) = e 2 5. Without any loss of generality we can assume that r is positive, then power density at r is given by x f (x)dx = e dx 2 x=r x=r 1 r e = . 2 A similar calculation for a negative r gives power density at distance r to be e Exercise 2.7.
|r|
2
.
Tse and Viswanath: Fundamentals of Wireless Communication Exercise 2.8. The block diagram for the (unmodied) system is: w(t)
7
Ak
/ (t)
O /x
/ R[]
/ h(t)
/ +
O /x
/ (t)
t=kT /
/ Bk
j 2f t c 2ej 2fc t 2e 1. Which lter should one redesign? One should redesign the lter at the transmitter. Modifying the lter at the receiver may cause {(t kT )}k no longer to be an orthonormal set, resulting in noise on the samples not to be i.i.d. By leaving {(t kT )}k at the receiver as an orthonormal set, we are assured the the noise on the samples is i.i.d. Let the modied lter be g(t). The block diagram for the modied system is: w(t)
Ak
/ g (t)
O /x
/ R[]
/ h(t)
/ +
O /x
/ (t)
t=kT /
/B k
j 2f t c 2ej 2fc t 2e (Solution to Part 3: Figure of the various lters at passband). We want to nd g(t) such that there is no ISI between samples. Before we continue to nd g(t), we depict the desired simplied block diagram for the system with no ISI: /+ / Bk Ak O wk For ease of manipulation, we transform the passband representation of the system to a baseband representation w(t)
Ak
/ g (t)
/ hb (t)
/+
/ (t)
t=kT /
/ Bk
H (f + fc ) [ W , W ] 2 2 0 otherwise H (f ) is assumed bandlimited between [fc W , fc + 2 where Hb (f ) = We let g (t) =
k
W ] 2
gk (t kT ), and redraw the block diagram:
Tse and Viswanath: Fundamentals of Wireless Communication w(t)
8
t=kT / Bk / We now convert the signals and lters from the continuous to discrete time domain: wk Ak
/ gk / (t) / hb (t) /+ / (t)
Ak
/ gk
/h k
/+
/B k
where hk = hb |t=kT . We justify interchanging the order of w(t) and (t), since we know the noise on the samples is i.i.d. G(z ) = H 1 (z ) gives the desired result. In summary, g (t) = k gk (t kT ) where gk is given by G(z ) = H 1 (z ), and H (z ) is given by the Z-Transform of hk = hb |t=kT Exercise 2.9. Part 1)
Figure 2.1: Magnitude of taps, W = 10kHz, time = 1 sec. Two paths are completely lumped together
Tse and Viswanath: Fundamentals of Wireless Communication
9
Figure 2.2: Magnitude of taps, W = 100kHz, time = 1 sec. Two paths are starting to become resolved.
Figure 2.3: Magnitude of taps, W = 1MHz, time = 1 sec. Two paths are more resolved.
Tse and Viswanath: Fundamentals of Wireless Communication
10
Figure 2.4: Magnitude of taps, W = 3MHz, time = 1 sec. Two paths are clearly resolved.
Tse and Viswanath: Fundamentals of Wireless Communication
11
Part 2) We see that the time variations have the same frequency in both cases (at fading in Figure 2.5 and frequency selective fading in Figure 2.7), but are much more pronounced in the case of at fading. This is because in frequency selective fading (large W) each of the signal paths corresponds to a dierent tap, so they dont interfere signicantly and the taps have small uctuations. On the other hand in the case of at fading, we sample the channel impulse response with low resolution and all the signal paths are lumped into the same tap. They interfere constructively and destructively generating large uctuations in the tap values. If the model included more signal paths, then the number of paths contributing signicantly to each tap would vary as a function of the bandwidth W , so the frequency of the tap variations would depend on the bandwidth, smaller bandwidth corresponding to larger Doppler spread and faster uctuations (smaller Tc ). Finally we could analyze this eect in the frequency domain. In frequency selective fading, the channel frequency response varies within the bandwidth of interest. There is an averaging eect and the resulting signal is never faded too much. This is an example of diversity over frequency. Exercise 2.10. Consider the environment in Figure 2.9. The shorter paths (dotted lines) contribute to the rst tap and the longer paths (dashed) contribute to the second tap. Then the delay spread for the rst tap is given by: fv | cos 1 cos 2 |, c and the delay spread for the second tap is given by: fv | cos 1 cos 2 |. c By appropriately choosing 1 , 2 , 1 and 2 , we can construct examples where the doppler spreads for both the taps are same or dierent. Exercise 2.11. Let H (f ) = 1 for |f | < W/2 and 0 otherwise. Then if h(t) H (f ) it follows that h(t) = W sinc(W t). Then we can write: {w[m]} = =
[w(t) 2 cos(2fc t)] h(t) |t=m/W
w( ) 2W cos(2fc )sinc(W (t ))d
t=m/W
=
w( ) 2W cos(2fc )sinc(m W )d
=
w( )m,1 ( )d
Tse and Viswanath: Fundamentals of Wireless Communication
12
Figure 2.5: Flat fading: time variation of magnitude of 1 tap. (x-axis is the time index m).
Figure 2.6: Flat fading: time variation of phase of 1 tap. (x-axis is the time index m).
Tse and Viswanath: Fundamentals of Wireless Communication
13
Figure 2.7: Frequency selective fading: time variation of magnitude of 1 tap. Note: scale of y-axis is much ner here than in the at fading case. (x-axis displays time with units of seconds. x-axis label of time index m is a typo. Should be time.)
Figure 2.8: Frequency selective fading: time variation of phase of 1 tap. (x-axis displays time with units of seconds. x-axis label of time index m is a typo. Should be time. )
Tse and Viswanath: Fundamentals of Wireless Communication
14
1 1
Tx Rx
2
v
2
Figure 2.9: Location of reectors, transmitter and receiver where m,1 ( ) = Similarly, 2W cos(2fc )sinc(m W ).
{w[m]} = [w(t) 2 sin(2fc t)] h(t) |t=m/W
=
w( ) 2W sin(2fc )sinc(W (t ))d
t=m/W
=
w( ) 2W sin(2fc )sinc(m W )d
=
w( )m,2 ( )d
where m,2 ( ) = 2W sin(2fc )sinc(m W ). Exercise 2.12. 1) Let n (t) denote (t nT ). Show that if the waveforms {n (t)}n form an orthogonal set, then the waveforms {n,1 , n, 2 }n also form an orthogonal set, provided (t) is band-limited to [fc , fc ]. n,1 , n, 2 are dened as n,1 (t) = n (t) cos 2fc t n, 2 (t) = n (t) sin 2fc t By denition {n (t)}n forms an orthogonal set
(t)m (t)dt = a [m n] n (f )m (f )df = a [m n] n
(2.5)
for some a R for some a R, by Parsevals Theorem(2.6)
where n (f ) is the Fourier Transform of n (t).
Tse and Viswanath: Fundamentals of Wireless Communication
15
We would like to show 1) < n,1 (t), m,1 (t) > [m n] m, n Z waveforms modulated by cos 2fc t remain orthogonal to each other 2) < n, 2 (t), m, 2 (t) > [m n] m, n Z waveforms modulated by sin 2fc t remain orthogonal to each other 3) < n,1 (t), m, 2 (t) > = 0 m, n Z waveforms modulated by cos 2fc t are orthog. to waveforms modulated by sin 2fc t. We will show these three cases individually: Case 1)
< n,1 (t), m,1 (t) > =
n,1 (t)m,1 (t)dt
=
1 (f )m,1 (f )df by Parsevals n,
(2.7)
where
n,1 (f ) =
n,1 (t)ej 2f t dt n (t) cos(2fc t)ej 2f t dt, from (2.5)
=
1 1 = n (f ) ( (f fc ) + (f + fc )) 2 2 1 = (n (f fc ) + n (f + fc )) 2 Substituting into (3) = 1 4 1 = 4
[ (f fc ) + (f + fc )][m (f fc ) + m (f + fc )]df n n (f fc )m (f fc ) + (f fc )m (f + fc ) + n n
=0
+ (f n
+ fc )m (f
=0
fc ) + (f n
+ fc )m (f + fc )df
The second and third terms equal zero since (t) is bandlimited to [fc , fc ] resulting in no overlap in the region of support of (f + fc ) and (f fc ), as seen in Figure 2.10(b). = 1 4
(f fc )m (f fc ) + (f + fc )m (f + fc )df n n
Tse and Viswanath: Fundamentals of Wireless Communication = 1 4
16
(f )m (f ) + (f )m (f )df n n
since integrals from to are invariant to shifts of the integrand along the x-axis.
1 2 (f )m (f )df 4 n a = [m n], by equation (2.6) 2 [m n]
=
(2.8)
Case 2) < n, 2 (t), m, 2 (t) >
=
n, 2 (t)m, 2 (t)dt
=
2 (f )m, 2 (f )df by Parsevals n, ( 1 1 ) [n (f fc ) (f + fc )]( )[m (f fc ) m (f + fc )]df n 2j 2j (f fc )m (f fc ) (f fc )m (f + fc ) + n n
=0
=
1 = 4
(f n
+ fc )m (f
=0
fc ) + (f n
+ fc )m (f + fc )df
= = = =
1 (f fc )m (f fc ) + (f + fc )m (f + fc )df n 4 n 1 (f )m (f ) + (f )m (f )df n 4 n 1 2 (f )m (f )df 4 n a [m n], by equation (2.6) 2 [m n]
(2.9)
Case 3) < n,1 (t), m, 2 (t) >
=
n,1 (t)m, 2 (t)dt
=
1 (f )m, 2 (f )df by Parsevals n,
Tse and Viswanath: Fundamentals of Wireless Communication 1 ) 4j 1 =( ) 4j =(
17
[ (f fc ) + (f + fc )][m (f fc ) m (f + fc )]df n n (f fc )m (f fc ) (f fc )m (f + fc ) + n n
=0
+ (f + fc )m (f fc ) (f + fc )m (f + fc )df n n
=0 1 =( ) (f fc )m (f fc ) (f + fc )m (f + fc )df n 4j n 1 =( ) (f )m (f ) (f )m (f )df n 4j n = 0 m, n Z
For (t) to be orthonormal, set should scale n (t) by 2.
a 2
= 1 in (2.8) and (2.9), which implies a = 2. We
Part 2) (t) = 4fc sinc(4fc t) is an example (t) that is not band-limited to [fc , fc ]. See Figure 2.10(c). For this example, there will be an overlap in the region of support n of (f + fc ) and (f fc ). See Figure 2.10(d). The cross terms (f fc )m (f + fc ) (f + fc )m (f fc ) will no longer = 0 and {n,1 , n, 2 }n will no longer by or and n thogonal. 2 take away messages: 1) The orthogonality property of a set of waveforms is unchanged if the waveforms experience a frequency shift, or in other words are multiplied by ej 2fc t . 2) WGN projected onto {n,1 , n, 2 }n will yield i.i.d. gaussian noise samples. Exercise 2.13. Let F[] denote the Fourier transform operator, denote convolution, u() the unit step function and 1/j if f > 0 0 if f = 0 H (f ) = 1/j if f < 0 with h(t) H (f ). Then we can write: [yb (t)ej 2fc t ] = = = 1 1 [yb (t)ej 2fc t (yb (t)ej 2fc t ) ] = F1 [Yb (f fc ) Yb (f fc )] 2j 2j 2 1 2 1 2 F [Y (f )u(f ) Y (f )u(f )] = F [Y (f )H (f )] = y (t) h(t) 2j 2 2 2 [ai (t)x(t i (t))] h(t) 2i
Tse and Viswanath: Fundamentals of Wireless Communication
18
fc
fc
freq
(a) Frequency range of (f ) band-limited from fc , fc
2fc
fc
fc
2fc
freq
(b) Frequency range of (f + f c) and (f f c). Notice no overlap in region of support.
2fc
2fc
freq
(c) Frequency range of (f ) not bandlimited from fc , fc
Tse and Viswanath: Fundamentals of Wireless Communication = = =a =
i
19
2 2
{ai (t) 2 [xb (t i (t))ej 2fc (ti (t)) ]} h(t)
i
1 2 1 2j
{ai (t)[xb (t i (t))ej 2fc (ti (t)) + x (t i (t))ej 2fc (ti (t)) ]} h(t) b
i
{ai (t)[xb (t i (t))ej 2fc (ti (t)) x (t i (t))ej 2fc (ti (t)) ]} b
i
{ai (t) [xb (t i (t))ej 2fc (ti (t)) ]} ai (t)xb (t i (t))ej 2fc i (t) ej 2fc t
i
=
The equality (a) follows because the rst term between the braces is zero for negative frequencies and the second term is zero for positive frequencies. Yes. Both equations together allow to equate the complex arguments of the and operators, thus allowing to obtain the baseband equivalent of the impulse response of the channel. Exercise 2.14. Exercise 2.15. Eects that make the tap gains vary with time: Doppler shifts and Doppler spread: D = fc i (t), Tc 1/D = 1/(fc i (t)) The coherence time is determined by the Doppler spread of the paths that contribute to a given tap. As W increases the paths are sampled at higher resolution and fewer paths contribute to each tap. Therefore the Doppler spread decreases for increasing W and its inuence on the variation of the tap gains decreases. Variation of {ai (t)}i with time. ai (t) changes slowly, with a time scale of variation much larger than the other eects discussed. However as W increases and it becomes comparable to fc assuming that a single gain aects the corresponding path equally across all frequencies may not be a good approximation. The reection coecient of the scatterers may be frequency dependent and for very large bandwidths we need to change the model. Movement of paths from tap to tap. i (t) changes with t and the corresponding path moves from one tap to another. As W increases fewer paths contribute to each tap and the tap gains change signicantly when a path moves from tap to tap. A path moves from tap to tap when i (t)W = 1 or i (t)/t W = 1/t. So this eect takes place in a time scale of t 1/(W i (t)). As W increases this eect starts taking place in a small time scale and it becomes the dominant cause of time variation in the channel tap gains.
Tse and Viswanath: Fundamentals of Wireless Communication The third eect dominates when t < Tc or equivalently when W > fc . Exercise 2.16.
N
20
h [m] =
i=1
ai (m/W )ej 2fc i (m/W ) sinc( i (m/W )W ) and i (m/W ) = i (m/W ) . Then, ai (m/W )ej 2fc i (m/W ) sinc( W i (m/W )W )
i=1
Let =
1 N
N i=1 i (0) N j 2fc
h [m] = e
Often in practice fc fc r/c >> 1 1 so it is a reasonable assumption to model j 2fc j e =e where U nif orm[0, 2 ] and is independent of everything else. Note that does not depend on m so a particular realization of is the same for all components of h. Since ej has uniformly distributed phase, its distribution does not change if we introduce an arbitrary phase shift . So ej ej ej . It follows that N j 2fc i (m/W ) sinc( W i (m/W )W ) i=1 ai (m/W )e N j 2fc i ((m+1)/W ) sinc( W i ((m + 1)/W )W ) i=1 ai ((m + 1)/W )e ej h = ej ej . . . N j 2fc i ((m+n)/W ) sinc( W i ((m + n)/W )W ) i=1 ai ((m + n)/W )e N j 2fc i (m/W ) sinc( W i (m/W )W ) i=1 ai (m/W )e N j 2fc i ((m+1)/W ) sinc( W i ((m + 1)/W )W ) i=1 ai ((m + 1)/W )e =d ej . . . N j 2fc i ((m+n)/W ) sinc( W i ((m + n)/W )W ) i=1 ai ((m + n)/W )e =d h Since this is true for all , under the previous assumptions h is circularly symmetric. Exercise 2.17. 1. h(, t) is the response of the channel to an impulse that occurs at time t , i.e., (t (t )). Replacing x(t) by (t (t )) in the given expression we obtain: a h(, t) = K
K 1
( i (t)).
i=0
The projection of the velocity vector v onto the direction of the path at angle has a magnitude: v = |v| cos .
1
r is the distance between transmit and receive antennas
Tse and Viswanath: Fundamentals of Wireless Communication
21
The distance travelled by the mobile in the direction in time t is v t, which is the reduction in the distance between transmitter and receiver for the path of angle . Then, |v| cos t (t) = (0) . c 2. Td 1/W means that most of the paths arrive in an interval much smaller than the sample time of the signal. Since the signal remains approximately constant over the interval Td , it can be pulled out from the summation in part (a). In this way we can lump together the inuence of all the paths into a single tap h0 [m]. We assume that (t) 1/W , For this we assume that 0 (0) = 0 and 0 (t) 1/W for the time scale of interest. Thus,
2
h0 [m] =
0
a ej 2fc (m/W ) sinc[ (m/W ) W ]d,
where we use the fact that a (t) = a , t. Finally we note that lim sinc(t) = 1
t=0
and since (m/W ) W
1 we obtain:
2
h0 [m] =
0
a ej 2fc (m/W ) d.
3. The independence assumption requires that dierent paths come from dierence scatters. For this to be true even for small variations in angle of arrival , it is necessary that the scatters be located far away from the receiver. How far depends on the size of the scatters, and the angle dierence over which we require the paths to be independent. The identically distributed assumption requires that the lengths of the paths from transmitter to receiver be comparable for all angle . This occurs when r R in the following gure. 4. The stationarity of h0 [m] can be seen from previous formula and the uniformity of the phase. To calculate R0 [n],
2 2 0
R0 [n] = E [
0
a1 a2 ej 2fc (1 (0)
vm+vn cW
vm cos 1 2 (0)+ cW cos 2 )
d1 d2 ]
Tse and Viswanath: Fundamentals of Wireless Communication
2 2 0 2
22
vm cos 1 + cW cos 2 )
=
0
E [a1 a2 ej 2fc (1 (0)2 (0)) ]ej 2fc (
vn
vm+vn cW
d1 d2
=
0
E [|a |2 ]ej 2fc ( cW cos ) d.
5.
R0 [n] = = =
S (f )ej 2f n df 4a2 W/(Ds 1 (2f W/Ds )2 )ej 2f n df d,
Ds /2W
Ds /2W 2 jnDs cos( )/W
2a e
0
where we use the substitution cos() = 2f W/Ds . 6. From the denition of PSD. Exercise 2.18. 1. The key dierence is that in Clarkes model, the attenuations of the signals are random and have the same distribution in all directions from the receiver, whereas in the present model, they are deterministic and direction dependent. The key similarity is that in both cases, the phases are i.i.d. in all directions. 2. The delay spread Td is (4 2) km /c = 6.7s. Therefore the channel is at if W 1/Td = 150 kHz. 3. We assume that only paths that arrive with delay in [ /W 1/(2W ), /W + 1/(2W )] contribute to tap . Let be the angle the path makes with the line between the transmitter and the receiver at the receiver. The paths that arrive with the desired delay lie in between angles 1 and 2 and between 1 and 2 , where 1 , 2 are such that the delay of the path is /W 1(2W ), i.e., 1 + r(1 ) = c( /W 1/(2W )) 1 + r(2 ) = c( /W + 1/(2W )) where r() = the angle . 5 4 cos is the distance between the Tx and the scatterer at
The total power received is E[|h |2 ] = 2
2
G
1
1 d r()2
(2.10)
Tse and Viswanath: Fundamentals of Wireless Communication 4. The received power in the delay range [, + d ] is 2G/(5 4 cos )|d|
23
where cos = [5 (c 1)2 ]/4 and sin d = c(c 1)/2d . Hence, the received power is cG (c 1) 1 ( 5(c 1) )2 4 for c between 2 and 4. This gives the power-delay prole. 5. The Doppler shift at angle is v cos /, where = c/fc . Thus the Doppler spread for the tap is +fc v/c| cos 1 cos 2 | . The power of the received signal in the range of Doppler shifts [f, f + df ] is 2G/(5 4 cos )|d| where cos = /v f and sin d = /vdf . Hence the Doppler spectrum is S (f ) = 2G/v (5 4f /v ) 1 (f /v )2 .
2
The Doppler spectrum of tap picks o a section of this corresponding to the range of Doppler shifts for the paths that contribute to this tap. 6. No, since the location of the scatterers should determine exactly the phase of the arriving path, so there cannot be any randomness of the phase once the location of the scatterers is xed. On the other hand, the phase varies rapidly as a function of the scatterer positions at a spatial scale of the order of (cms) while the large scale path loss and delay varies at the scale of kilometers. So what our assumptions are saying is that we are assuming that the scatters are approximately 1 km from the receiver, where the approximation is accurate up to the order of s . The randomness at the small scale validates the random phase assumption. Exercise 2.19. 1. fm = vfc c
D = 2 fm =
2vfc c
Tse and Viswanath: Fundamentals of Wireless Communication Tc = c 1 = 4D 8vfc
24
b) The antennas should be spaced at least by d = vTc = c/(8fc ) to get independenlty faded signals. 2. The signal arriving at the base station antenna from an angle relative to the direction of v experiences a Doppler shift f = v fc cos() c
ranges from arctan(R/d) to +arctan(R/d). The Doppler shift is maximum for = and minimum for = arctan(R/d) or = + arctan(R/d). Therefore the Doppler spread is D= v fc [1 cos(arctan(R/d))] v fc 2[sin(arctan(R/d)/2)]2 = c c
and the corresponding Tc : Tc = 1 c = 4D v fc 8[sin(arctan(R/d)/2)]2
3. The minimum base station antenna spacing for uncorrelated fading is d = v Tc = c/(fc 8[sin(arctan(R/d)/2)]2 ). In practice the base station antenna is located in a high tower with no obstructions in its vicinity, so most of the scattering takes place around the mobile. In this case we can assume R << d and approximate sin(arctan(R/d)/2) R/2d to get d = (cd2 )/(2fc R2 ). In this particular setting this means that the minimum antenna spacing at the base station must be in the order of d2 /R2 larger than that at the mobile to get independently faded signals.
Chapter 3 Solutions to Exercises
Exercise 3.1. We have Pe = Eh [Q( 2|h|2 SNR)], 1 2 x et /2 dtdx, = e 2 0 2xSNR t2 /(2SNR) 1 2 = et /2 ex dxdt, 2 0 0 1 2 2 et /2 (1 et /(2SNR) )dt, = 2 0 1 1 2 et (1+1/SNR)/2 dt, = 2 2 0 = 1 2 1 SNR 1 + SNR , (3.1) (3.2) (3.3) (3.4) (3.5) (3.6)
where the third step follows from changing the order of integration. Now, for large SNR, we also have SNR 1 1 , 1 + SNR 2SNR which implies Pe Exercise 3.2. 1 4SNR
2x
1. Let = SNR. For Rayleigh fading |h[0]|2 Exp(1) so we have: 2|h[0]|2 =
0
Pe = E Q
1 2 et /2 ex dtdx 2
25
Tse and Viswanath: Fundamentals of Wireless Communication
t2 /(2) 0
26
=
0
1 2 et /2 ex dxdt = 2
0
1 2 2 et /2 1 et /(2) dt 2 1 1 + 1/
=
1 1 2
1+
1 + t2 (1+1/) 1 e2 dt = 1 2 2
We can approximate Pe = and
1/(1 + x) = 1 x/2 + o(x) for x 0 1 . Then, 11 1 1 11+ + o(1/) = + o(1/) 2 2 4 lim Pe = 1 4
2. We will need the following result:
0
Q( y )dy =
0
y
1 2 et /2 dtdy = 2
0 0
t2
1 2 et /2 dydt = 2
0
t2 1 2 et /2 dt = 2 2
Let f () be the pdf of |h[0]|2 . Then,
Pe = E Q
2|h[0]|2
=
0
Q(
2x)f (x)dx
Assuming that f () is right continuous at 0, that f (0) > 0 and that f () is bounded (this last condtion enables us to use the bounded convergence theorem to exchange limit and integral):
lim Pe =
lim
Q( 2x)f (x)dx = lim
0 0
Q( y ) f
0
y 2
1 dy 2
f (0) = 2
f (0) Q( y )dy = 4
3. Let g () be the pdf of |h[ ]|2 , and assume that it is right continuous and strictly positive at 0, for = 1, . . . , L. Let f () be the pdf of i=1 |h[i]|2 . Then using the fact that the pdf of the sum of independent random variables equals the convolution of the corresponding pdfs we can write for x 0:
x
f2 (x) =
0
1
g1 (t)g2 (x t)dt = g1 (0)g2 (0)x + o(x)
limx0 o(x)/x = 0
Tse and Viswanath: Fundamentals of Wireless Communication
x x
27
f3 (x) = . . . f L ( x) =
0 0
f2 (t)g3 (x t)dt =
0
[g1 (0)g2 (0)t + o(t)]g3 (x t)dt = g1 (0)g2 (0)g3 (0)
x2 + o(x2 ) 2
x
x
L1
fL1 (t)gL (x t)dt =
0 L L1 =1
g (0)
tL2 + o(tL2 ) gL (x t)dt (L 2)!
=
=1
g (0)
x + o(xL1 ) = xL1 + o(xL1 ) (L 1)!
L =1
where we dened = [
g (0)]/(L 1)!. The probability of error is given by:
Pe = E Q
2 h 2
=
0
Q( 2x)fL (x)dx
Multiplying by L , taking limit for and assuming that we can exchange the order of limits and integrals we have:
lim Pe
L
= =
lim lim
Q( 2x) fL (x)dx = lim
0
L
0
Q( y )L1 fL
0
y 2
L1
1 dy 2 1 dy 2
fL Q( y )
y 2
y 2 L 1
0 0 0
y 2
L1
1 dy = 2
y Q( y ) 2
2
=L 2 = 2L
t 1 1 t2 /2 L1 2 e et /2 y L1 dydt y dtdy = L 20 2 2 0 y 2L 1 1 t 1 2 2 et /2 dt = L t2L et /2 dt L 2 2 2L 2
1 (2L)! =L = 2 2L L!2L
2L 1 L
1 4L
L
g (0)
=1
4. The K parameter of a Ricean distribution is dened as the ratio of the powers in the specular (or constant) component and the fading component. If the specular component has amplitude and the fading component is CN (0, 2 ) and we normalize the total power to be 1 we obtain: 1 = 2 + 2 = 2 (K + 1) =
K K +1
2 =
1 K +1
For Ricean fading with parameter K the pdf of |h[ ]|2 is given by (see Proakis (2.1-140)): f (y ) = (K + 1)e( K +1 +y)(K +1) I0
K
y K (K + 1) , y 0
Tse and Viswanath: Fundamentals of Wireless Communication which evaluated at y = 0 yields f (0) = (K + 1)eK . Using the result from part (3) we get:
28
lim Pe L =
2L 1 L
(K + 1)L eLK 4L
As K the expression above decays exponentially in K and the Ricean channel converges to the AWGN channel. Exercise 3.3. Since |h|2 is exponential, for large SNR, we have P(E ) SNR(1 ) . Therefore log P(E ) = (1 ). SNR log SNR lim Similarly, log P(E ) = (1 + ). SNR log SNR lim 1. By conditioning on E , probability of error can be written as: Pe = P(error|E )P(E ) + P(error|E c )P(E c ). Now, the for the second term, we see that |h|2 SNR > SNR whenever E c happens. Thus, because of the exponential tail of the Q function, the second term goes to zero exponentially fast and does not contribute to the limit. Also, upper bounding P(error|E ) by 1, we get log P(E ) log Pe lim , SNR log SNR SNR log SNR = (1 ). lim 2. Now, similarly conditioning on E , we get
c c Pe = P(error|E )P(E ) + P(error|E )P(E ), P(error|E )P(E ).
Now, we see that |h|2 SNR < SNR whenever E happens. Thus, the probability of error then can be lower bounded by a nonzero constant (e.g. Q(1)). Thus, we get log(Q(1)) + log(P(E )) log Pe lim , SNR SNR log SNR log SNR = (1 + ). lim
Tse and Viswanath: Fundamentals of Wireless Communication 3. Combining the last two part, we get log Pe = 1. SNR log SNR lim
29
Exercise 3.4. To keep the same probability of error, the separation between consecutive points should be the same for both PAM and QAM. Let this separation be 2a. Then, the average energy for a PAM with 2k points is given by: Eav (2k PAM) = = 1 2k1
2 2k1
(2i 1)2 a2 ,
i=1
a 2k (2 1). 3
Since a 2k QAM can be thought as two independent 2k/2 PAMs, we get that the average energy for a QAM with 2k points is: Eav (2k QAM) = 2Eav (2k/2 PAM), 2a2 k = (2 1). 3 Thus, the loss in energy is given by: 10 log which grows linearly in k . Exercise 3.5. Consider the following scheme which works for both BPSK and QPSK. We have y [m] = hx[m 1]u[m] + w[m], y [m 1] = hx[m 1] + w[m 1]. Substituting the value of hx[m 1] from the second equation into the rst, we get y [m] = y [m 1]u[m] + w[m] u[m]w[m 1], y [m] = hu[m] + w[m]. Where w[m] is CN (0, 2N0 ) and is independent of the input (because of symmetry of w[m 1]). The eective channel h is CN (0, a2 + N0 ) and is known at the receiver. Thus, the performance is same as the performance on a coherent channel with signal 2k + 1 2 ,
Tse and Viswanath: Fundamentals of Wireless Communication
30
to noise ratio given by (SNR + 1)/2. Thus, the probability of error for BPSK is given by 1 , 2(1 + SNR) and the probability of error for QPSK is given by: 1 . (1 + SNR) Exercise 3.6. 1. Let = SNR.
2x
Pe = E Q
2 h[0]
t2 /(2) 0
2
=
0
1 xL1 x 2 et /2 e dtdx (L 1)! 2 (3.7)
=
0 (a) 0
1 xL 1 x 2 et /2 e dxdt (L 1)! 2
L1
=
1 2 et /2 1 2 k=0 1+ 1+
L1 L1
t2 2
k
et /(2) dt k! 1 + t2 (1+1/) e2 dt 2
k
2
=
(b)
1 1 2 1 1 2
k=0 L1
1 k !(2)k 1 k !(2)k 1 2
=
k=0
1+
k
(2k )! k !2k
k
(3.8)
=
1 1 2 k=0
2k k
1+ 2
where = /(1 + ). Also (a) follows from the equivalence between the distribution functions of the Gamma and Poisson random variables, and (b) results from the formula for the even moments of the standard Gaussian distribution with proper scaling. 2. We start with the sucient statistics: rA = h[ ]x1 + wA , rB = h[ ]x2 + wB ,
() () () ()
= 1, 2, . . . , L = 1, 2, . . . , L (3.9)
where x1 = xA and x2 = 0 if xA is transmitted, and x1 = 0 and x2 = xB if xB is transmitted. As in the notes assume xA 2 = xB 2 = Eb .
Tse and Viswanath: Fundamentals of Wireless Communication
31
Since we are analyzing coherent reception, the receiver knows h and can further project onto h/ h obtaining the new sucient statistics: rA = h rA = h x1 + wA h h rB = rb = h x2 + wB h
(3.10)
where wA and wB are independent CN (0, N0 ) random variables. We can nally compare (rA ) and (rB ) to decide whether xA or xB was transmitted. The error probability is given by: Pe = E [P r ( {rA } > = E Pr {rB } | xB )] {wA wB } > h E b
=E Q
2L h
2
Eb (3.11) 2LN0
Since Lh CN (0, IL ) if follows that the formula of part (1) still holds by replacing Eb /(2LN0 ), which results in the desired answer. Exercise 3.7. and d2 ) 1. We have (for simplicity we denote the squared-distances as d1 SNR(|h1 |2 d1 + |h2 |2 d2 ) 2
q
SNR(xd1 +yd2 ) 0 2 2t2 /(d2 SNR) (2t2 /SNRd2 y )/d1
P[xA xB ] = Eh1 ,h2 Q 1 = 2 1 = 2 1 = 2 1 = 2
0 0 0 0 0
et
2 /2
ex ey dtdxdy ex dxey dyet )ey dyet
2 /2
dt
0 2t2 /(d2 SNR)
0
(1 e(2t
2 /SNRd
2 y )/d1
2 /2
dt
1e
2t2 /(d2 SNR) 0
e2t /d1 SNR 2 2 (1 e(1/d2 /d1 )2t /SNR ) et /2 dt, 1 d2 /d1
2 /2
2
= 0.5 + = 0.5 +
1 2 (d2 d1 )
et
(d1 e2t
2 /(d
1 SNR)
d2 e2t
2 /(d
2 SNR)
)dt,
0.5 d2 d1
d1 1 + 4/(d1 SNR)
d2 1 + 4/(d2 SNR)
Tse and Viswanath: Fundamentals of Wireless Communication 2. For the high SNR scenario, we get (using Taylor series) 1 1 + 4/(d2 SNR) Which implies P[xA xB ] = 3/(d1 d2 SNR2 ). Exercise 3.8. so 1. For QPSK over an AWGN channel Pcorrect = 1 Q Pe = 1 Pcorrect = 2Q 2a2 N0 Q 2a2 N0
2 2a2 N0
32
= 1 2/(d2 SNR) + 6/(SNR2 d2 ). 2
2
To compare the performance to that of a scheme that uses only real symbols we x the bit rate (2 bits per channel use) and total transmitted power. For 4-PAM with signal points located in {b, 3b} the symbol error probability is (see Proakis (5.2-42)): 3 Pe = Q 2 2b2 N0
The average transmitted power is b2 so normalizing the total transmitted power 5 to 1 we have a = 1/ 2, b = 1/ 5. The power loss of 4-PAM over QPSK is 5/2 = 4dB . 2. The conditional probability of error of QPSK conditioned on the channel realization h is: 2 2h2 2h2 2a Q 2a Pe|h = 2Q N0 N0 Averaging over the distribution of h 2 , noting that the second term of the above expression does not aect the high SNR error performance, upper bounding the Q function as done in the lectures, and using the characteristic function of the exponential distribution to evaluate the expectations we obtain: L 2h2 |h[ ]|2 a2 h 2 a2 E 2e N0 2Q 2a =2 E e N0 Pe = E Pe|h E N0 =1 =2 1 1 + a2 /N0
L
=2 1+
SNR 2
L
Tse and Viswanath: Fundamentals of Wireless Communication where SNR = 2a2 /N0 is the signal to noise ratio.
33
For 4-PAM we can nd the conditional error probability conditioned on h using the result of question 4.b) of homework 2 (which was derived for binary antipodal signaling but can easily be extended for 4-PAM) and scaling appropriately by b: 2h2 3 2b Pe|h = Q 2 N0 Taking expectation over h and bounding as before we get: Pe 3 2 1 1 + b2 /N0
L
=
3 2
1+
SNR 5
L
Therefore we get in both cases the same diversity gain, but the SNR performance is degraded by 4dB in the 4-PAM case. 3. Let A = {a(1+ j ), a(1 j ), a(1+ j ), a(1 j )}, B = {x C 2 : x1 A, x2 A}. Then the transmitted symbols are in D = {Ux : x B}, for some xed unitary matrix U C 2x2 . Let x = [x[1], x[2]]T be one of such symbols. Then the received signal is: y [ ] = h[ ]x[ ] + z [ ], = 1, 2 from which we can extract the sucient statistics: r[ ] = h[ ] y [ ] = |h[ ]|x[ ] + z [ ], = 1, 2 |h[ ]| = 1, 2 be the modied symbol as seen at
where z =d . Let v [ ] = |h[ ]|x[ ] for z the receiver.
We are interested in the pairwise error probability, that is the probability of detecting a symbol x2 when the transmitted symbol is x1 . The pairwise error probability conditioned on the channel realization h depends only on v1 v2 2 through the expression: 2 v2 v1 P(x1 x2 )|h = Q 2 N0 which can be rewritten and upper bounded as: 2 |x [1] x [1]|2 + |h[2]|2 |x [2] x [2]|2 |h[1]| 2 1 2 1 P(x1 x2 )|h = Q 2 N0
Tse and Viswanath: Fundamentals of Wireless Communication |h[1]|2 |x2 [1] x1 [1]|2 + |h[2]|2 |x2 [2] x1 [2]|2 4N0
34
exp
Taking expectation over h and using the characteristic function of the exponential distribution to evaluate it we obtain: P(x1 x2 ) = = |h[1]|2 |x2 [1] x1 [1]|2 + |h[2]|2 |x2 [2] x1 [2]|2 E exp 4N0 2 2 |h[1]| |x2 [1] x1 [1]| |h[2]|2 |x2 [2] x1 [2]|2 E exp E exp 4N0 4 N0 4 N0 4N0 2 1 + |x2 [1] x1 [1]| 1 + |x2 [2] x1 [2]|2 16N0 |x2 [1] x1 [1]|2 |x2 [2] x1 [2]|2
We see that the pairwise error probability depends only on the product distance |x2 [1] x1 [1]|2 |x2 [2] x1 [2]|2 . The rotation matrix U should be chosen in such a way as to maximize the minimum product distance between any pair of codewords in the constellation D. Exercise 3.9. Exercise 3.10. 1. The code can be represented by a permutation of the 16-point QAM. What is transmitted on the second subchannel can be obtained as a simple permutation of what is transmitted on the rst sub-channel. 2. Data rate = 2 bits/channel use (since all the information is contained in one of the QAMs itself). 3. Since the product distance is non-zero, the diversity gain is 2. The minimum product distance is given by by 64a4 where 2a is the minimum distance between the QAM symbols. Then, by normalizing the average receiver SNR to be 1 per time symbol, we get: 2 2 (4 20a2 )/16 = 1, a2 = 0.05. Therefore, the product distance is given by 0.32. 4. The power used for the rotation code is 4a2 per time and that of the permutation code is 20a2 , but for a fair comparison of power used, we need to normalize by the optimal product distance of the rotation code. Numerical results indicate that the rotation code outperforms a permutation code by a factor of 1.05.
Tse and Viswanath: Fundamentals of Wireless Communication Exercise 3.11. 1.
35
1 ]d 1 + SNR where d is the Hamming distance between the binary codewords xA and xB . The diversity gain of the code is the minimum Hamming distance dmin between the codewords . P {xA xB } [
2. Its 2. Same diversity gain as the repetition code but higher rate 3/2 rather than 1/2. 3. The probability a symbol gets decoded incorrectly is of the order of SNR1 . If dmin /2 errors are made, then there is a signicant probability (i.e., the probability does not decay with SNR) that an overall error is made, as an incorrect codeword may be closer to c than the transmitted codeword. We are ok if fewer than that is made. Hence, the diversity gain is dmin /2 . For the example, the diversity is only 1. 4. The typical error event for each symbol is when the channel is in a deep fade. If we declare an erasure whenever the channel is in a deep fade, then the typical error is that there are only erasures and no hard decision errors. We can decode whenever the number of erasures is less than dmin , since there is at most one codeword that is consistent with the erasure pattern. Hence the diversity gain is back to dmin , same as soft decision decoding. How do we know the channel is in deep fade? Heuristically, when |h |2 < 1/SNR. More rigorously, we can x a > 0 and use the threshold |h |2 < 1/SNR1 to decide on an erasure. This will give us a diversity gain of dmin (1 ). But we can choose arbitrarily close to zero so we can get close to the desired diversity gain. Exercise 3.12. 1. For repetition coding, probability of error is given by P |h|2 < 1/SNR . Now, let the singular value of decomposition of Kh be Kh = UU . Now, dene: h = U h. Then |h| = |h| and h has i.i.d. complex Gaussian entries with variance given by the singular values of Kh . Thus, the diversity order is given by the number non-zero singular values of Kh .
Tse and Viswanath: Fundamentals of Wireless Communication 2. We can write the time diversity channel as: x1 0 0 x2 0 y = [h1 , h 2 , , h L ] 0 0 ... 000 x1 0 1/2 1/2 = [h1 , h2 , , hL ]Kh Kh 0 0
1/2
36
0 0 + w, 0 xL 0 x2 0 .. . 0 0 0
0 0 + w, 0 xL
Now, [h1 , h2 , , hL ]Kh is i.i.d. Gaussian and is similar to a standard MISO channel. Thus the code design criterion is given by the determinant of K0.5 (XA h XB )(XA XB ) K0.5 . As Kh is a xed matrix and only contributes a constant h we can eliminate it from the code design criterion. Now, in our case XA and XB are restricted to be diagonal which implies that the dierence determinant is the the product distance. Thus, the criterion remains unchanged. Note, that here we have assumed Kh is full rank. Otherwise, one will have to work with the corresponding reduced problem by ignoring some of the hi s. 3. As seen in the previous part, the criterion remains unchanged. Exercise 3.13. The channel equation is y = hx + w. Let l = arg maxi |h[i]|. The selection combiner bases its decision on the lth branch only discarding the rest so the decision is based on y [ ] = h[ ]x + w[ ]. Let {xi }L be i.i.d. Exp(1) random variables, and x = maxi xi . Then the pdf of i=1 x, f (), for x 0 is given by: f (x) = L(1 ex )L1 ex = L[1 (1 x + o(x))]L1 [1 x + o(x)] = LxL1 + o(xL1 ) Noting that |h[ ]|2 has the above pdf, we can use the derivation of Ex. 3.2, part 3) replacing with L:
lim Pe L =
(2L 1)! L (2L 1)! =L L 4 L! 4 (L 1)!
We observe that this scheme still achieves a diversity gain of L but the error performance degrades by the factor L/ = L!/( L gi [0]) = L! with respect to that of i=1 optimal combining. Exercise 3.14.
Tse and Viswanath: Fundamentals of Wireless Communication
37
Exercise 3.15. 1. We obtain the same diversity gain over the MISO channel (assuming the same statistical characterization of the fading gains) as we have operationally converted the MISO channel into a parallel channel. Also, since the determinant of a diagonal matrix is the product of its diagonal elements, the determinant metric for the MISO channel is same as the product distance metric of the time diversity code. 2. For the rotation code the worst case pairwise error probability is given by 16 SNR2 , min ij where the optimal ij is given by 16/5, which gives 5/SNR2 as the upper bound for the rotation code. On the other hand, for Alamouti scheme, the worst case probability of error is given by: 16 . SNR det(XA XB )2
2
If u1 and u2 are the BPSK (+/ a) symbols used for the Alamouti scheme, then the average power per time symbol is given by 2a2 . The determinant is given by u2 + u2 , thus the worst case determinant is given by 4a2 . Thus, after normalizing 1 2 we get worst case probability of error is given by 4/SNR2 . Thus, the Alamouti scheme is better. For QPSK symbols, we are just using an additional degree of freedom which will change the power used for both the schemes, but the relative dierence in the worst case error performance (a factor of 1.25) remains the same. 3. See Figure 3.1. Exercise 3.16. 1. We have y = Ad + w, =
i
ai di + w.
Since A is orthogonal, all the ai s are orthogonal, thus for detecting di , we can project along ai : a y = ||ai ||2 di + a w. i i (3.12)
Since A is orthogonal, the noise a w is independent of other noise terms (and i hence the other projections).Thus, each of the di s can be decoded separately.
Tse and Viswanath: Fundamentals of Wireless Communication
A 2 Tx antennas MISO channel with i.i.d. Rayleigh fading
Probability of error for permutationn code Probability of error for the Alamouti scheme
10
1
38
10
0
probability of error
10
2
10
3
10
4
10
5
5
10
15
20
25
30
SNR in db
Figure 3.1: The error probability of uncoded QAM with the Alamouti scheme and that of a permutation code over one antenna at a time for the Rayleigh fading MISO channel with two transmit antennas: the permutation code is only about 1.5 dB worse than the Alamouti scheme over the plotted error probability range. 2. If ||am || = ||h||, then we can normalize the equation (3.12) and get a fading coecient of ||h|| which implies a full diversity gain for each symbol. 3. We have h X = dt At , which along with orthogonality and the full diversity property of A implies that h XX h = ||d||2 ||h||2 IL , h (XX ||d||2 IL )h = 0, for every h. Thus, XX must be ||d||2 IL . Exercise 3.17. Exercise 3.18. First, lets calculate the deep fade probability for Ricean fading which we will use for calculating both diversity order and product distance criterion. Let h = Aei + n,
Tse and Viswanath: Fundamentals of Wireless Communication
39
where A is a xed number and is uniform in [0, 2 ] and n is CN (0, 1). Since we are interested only in the |h|2 whose distribution does not depend on (because n is circularly symmetric), without any loss of generality, we can take = 0. Then we can write h as h = A + nR + jnI , where n = nR + jnI and nR and nI are i.i.d. Gaussian. Then P |h|2 < 1/SNR = P (A + nR )2 + (nI )2 < 1/SNR , P (A + nR )2 < 1/SNR and (nI )2 < 1/SNR , eA /4 1 , SNR SNR 2 eA /4 . SNR 1. Now, the deep fade event for a time diversity channel is when all the sub-channels are in deep fade. The probability of which is given by eLA /4 . SNRL Thus, the diversity order does not change for Ricean fading. 2. For a pairwise code design criterion, we want to calculate P
i
2 2
|hi |2 d2 < 1/SNR i
i
P |hi |2 d2 < 1/SNR , i eLA /4 . SNRL d2 d2 d2 12 L
2
Thus, we get the same product distance criterion for Ricean fading as well. Exercise 3.19. 1. Let H be the fading matrix for the MIMO channel. Then the channel model can be written as: Y = HX + W. Now, this channel model can be rewritten as a MISO channel with block-length nr N . Let X and h be X00 X = 0 ... 0 , 00X
Tse and Viswanath: Fundamentals of Wireless Communication h= H (1, 1) H (1, L) H (2, 1) H (nr , L) .
40
Then the received signal can be rewritten as y = hX + w , with y and w appropriately dened in term of Y and W. Then the probability of pairwise error can be written as: h SNRh(XA XB )(XA XB ) E Q 2 2. Since we have reduced the MIMO problem with i.i.d. Rayleigh fading to a MISO problem with i.i.d. Rayleigh fading, probability of pairwise error can be upper bounded as: P(XA XB ) = 4Lnr SNRLnr det (XA XB )(XA XB ) 4L SNRL det ((XA XB )(XA XB ) ) ,
nr
,
where the last step follows from the diagonal structure of XA and XB . 3. Thus, the code design criterion of maximizing the minimum determinant remains unchanged. Exercise 3.20. Using the same notation as in the text, and using a subindex to denote the receive antenna (either 1 or 2) we have: y1 [1] h11 h21 w1 [1] y1 [2] h h u1 w [2] 11 = 21 (3.13) + 1 y2 [1] h12 h22 u2 w2 [1] y2 [2] h h w2 [2] 22 12 where hij is the complex channel gain from transmit antenna i to receive antenna j . In vector notation we can rewrite (3.13) as y = Hu + w, where w CN (0, N0 I4 ). Noting that the columns of H (which we call hi ) are orthogonal, we can project the (slightly modied) received vector y onto the normalized columns of H to obtain the 2 sucient statistics: h (3.14) ri = i y = h ui + wi hi for i = 1, 2, where wi CN (0, N0 ) independent across i, and h = h1 = h2 is the eective gain.
Tse and Viswanath: Fundamentals of Wireless Communication
41
Exercise 3.21. 1. For spatial multiplexing, each stream comes from a BPSK constellation (+/ a). Then, to normalize the average transmit power to 1, we get a = 0.5. Then, the probability of pairwise error is upper bounded as: P(x1 x2 ) 16 , SNR ||x1 x2 ||4 16 , SNR2 (2a)4 4 . SNR2
2
For the Alamouti scheme, we should be using the same rate. Thus, each symbol should come from a four point PAM (+/ a, +/ 3a). Then to normalize the average transmit power we have: 1 2 2a2 + 18a2 = 1, 4 a2 = 1/10. Now, the probability of pair-wise error is upper bounded by: P(XA XB ) 42 SNR2 det ((XA XB )(XA XB ) ) 44 , SNR4 ((u1a u1b )2 + (u2a u2b )2 )2 256 , SNR4 44 a8 10000 . SNR4
2
,
For SNR > 50, Alamouti scheme outperforms the spatial multiplexing scheme. 2. For general R, the only thing that changes is the value of the minimum distance between the points. For a 2R/2 point PAM in case of spatial multiplexing, the minimum distance is given by: a2 = 2(2R 3 . 1)
Then, the probability of error is upper bounded by: P(x1 x2 ) 16 , SNR2 (2a)4
Tse and Viswanath: Fundamentals of Wireless Communication 4(2R 1)2 . 9SNR2
42
For the Alamouti scheme, the minimum distance is given by: a2 = 3 2(22R 1) .
Thus, the probability of error is given by P(XA XB ) 1 , SNR4 a8 16(22R 1)4 = . 81SNR4
The SNR threshold for general R turns out to be approximately 0.6623R which increases exponentially in R. Exercise 3.22. When using QAMs, the only thing that changes is the minimum distance. For the spatial multiplexing scheme, the QAM constellation size is 2R/2 . Thus, energy for the the 2R/2 QAM is given by 2a2 2R/2 1 . 3 3 4(2R/2 1)
Thus, to normalize the total transmit power per unit time, we get: a2 = .
Then, the probability of error is upper bounded by: P(x1 x2 ) 1 , SNR2 a4 16(2R/2 1)2 . 9SNR2 3 . 1)
For the Alamouti scheme, the minimum distance is given by: a2 = Thus, the probability of error is given by P(XA XB ) 1 , SNR4 a8 256(2R 1)4 . = 81SNR4 4(2R
Tse and Viswanath: Fundamentals of Wireless Communication
43
Exercise 3.23. 1. Let t = min(m, k ). Consider the repetition scheme in which we transmit the same symbol over a dierent antenna at each time until either we run out of antennas (in which case we can cycle again over the same antennas) or the block of lenght k ends. We further repeat the transmission n times over dierent blocks of length k . The antennas are used one at a time. If X is the matrix representing the transmitted codeword, with X(i, j ) denoting the signal transmitted through antenna i at time j (1 i m, 1 j kn) the codewords of the above repetition scheme are of the form: 1 0 0 1 0 0 1 0 0 0 1 0 0 1 0 0 1 0 X = . . .. . . . .. . .. ... . . . ... . x .. . . . .... .. .. . 0 0 1 0 0 1 0 0 1 where x is the scalar symbol we want to transmit and for simplicity we have assumed m = k . From the channel model, we see that the same symbol is transmitted over t n independent channel realizations, resulting in a diversity gain of t n = min(m, k ) n. We will see in the next part that this simple repetition code achieves the maximum possible diversity gain. 2. The channel model is: y[i]T = h[i]T X[i] + z[i] where i, 1 i n is the block index, y[i] C k is the received signal for block i, h[i] C m is the channel vector assumed constant for block i and independent across blocks, X[i] C mxk is the part of the codeword transmitted during the block i, and z[i] is i.i.d. white Gaussian noise. At the receiver the eective received codeword is: v = h[1]T X[1], h[2]T X[2], , h[n]T X[n]
T
C kn
Conditioned on the channel realization h[1], h[2], . . . , h[n], the pairwise error probability depends only on v1 v2 2 through the expression: 2 v2 v1 P(X1 X2 )|h = Q 2N0 which can be rewritten and upper bounded as: n T (X [i] X [i]) 2 h[i] 2 1 P(x1 x2 )|h = Q 2N0 i=1
Tse and Viswanath: Fundamentals of Wireless Communication
n
44
2
i=1
exp
h[i]T (X2 [i] X1 [i]) 4 N0
By taking expectations over h[i], i = 1, 2, . . . , n and using the fact that dierent blocks experience independent fadings we obtain: h[i]T (X2 [i] X1 [i]) E exp 4N0 i=1
n 2
P(x1 x2 )
Let A[i] = (X2 [i] X1 [i])(X2 [i] X1 [i]) 2 . Since A[i] is positive semidenite it can be expressed as A[i] = V[i][i]V[i] where V[i] is unitary and [i] = diag (2 [i], . . . , 2 [i]). 1 m Then we have: h[i]T (X2 [i] X1 [i])
2
= h[i]T V[i][i]V[i] h[i]T = h[i]T [i]h[i]T
m
=
=1
|h [i]|2 2 [i]
where h[i]T = h[i]T V[i] has the same distribution as h[i]T because the entries of h[i] are i.i.d. CN(0, 1) and V[i] is unitary. Therefore letting ti = rank (A[i]) min(m, k ),
n m n m
P(x1 x2 )
i=1 n ti
2 [i]|h [i]|2 E exp 4N0 =1 4N0 2 [i]
=
i=1 =1
1 1 + [i]/4N0
2
i=1 =1
where we have assumed 2 [i] 2 [i] 2i [i] > 0 for i = 1, 2, . . . , n. 1 2 t Finally we get the pairwise code design criterion: choose the code so that A[i] is full rank for all 1 i n and for all pairs of codewords (which assures a maximum diversity gain equal to n min(m, k )) and among those codes, choose the one that maximizes the minimum product n=1 ti=1 2 [i] = n=1 det(A[i]) i i over all pairs of codewords. Note that the repetition code proposed in a) with x = 1 achieves full diversity gain.
A[i] depends on the pair of codewords considered. To make the notation simpler we avoided using more indices to explicitly show this dependence.
2
Tse and Viswanath: Fundamentals of Wireless Communication
45
For pure time diversity we set m = 1 in which case the matrices A[i] reduce to complex scalars with eigenvalues |A[i]| = x2 [i] x1 [i] 2 and the pairwise error probability depends on the product distance between the pair of codewords n 2 i=1 x2 [i] x1 [i] . For pure spatial diversity we set n = 1 in which case there is only one matrix A for each pair of codewords. The design criterion reduces to choosing the code such that A is full rank for all pair of codewords, and among those codes that satisfy this, choose the one that maximizes the minimum det(A) for all pair of codewords. This criterion is the one obtained in class. Exercise 3.24. 1. Assume uniform scattering around the mobile. In this case the fading correlation is independent of the direction of movement, so we can assume that the mobile is moving in the same direction as where the second antenna would be located. The mobile reaches the location of the second antenna after traveling a distance d, after a time interval d/v where v is the speed of movement. The correlation of the fading gains between these 2 locations is given by R[ d/v W ], where 1/W is the sampling interval of the discrete time model. The fading gains are zero mean, circularly symmetric jointly complex Gaussian, so their joint distribution is completely determined by their correlation matrix. Letting h = [h1 h2 ]T be the gains at the locations of the 2 antennas at a given time, we have: h CN (0, Kh ) (3.15) where Kh = R[0] R[ d/v W ] R[ d/v W ] R[0] (3.16)
2. The received signal at the 2 antennas is y[n] = hx[n] + w[n] where w[n] CN (0, N0 I2 ). Conditioned on h the error probability for BPSK is Q( 2 h 2 SNR) and the average error probability is: Pe = E Q( 2 h 2 SNR) (3.17)
where the expectation is taken with respect to the distribution of h which was found in part (1). 3. The problem of directly doing a high SNR approximation in the computation of (3.17) is that the 2 components of h are correlated. However, noting that any nonsingular transformation of y[n] is a sucient statistic, we can do a transformation that decorrelates the entries of h. Let Kh = UU where U is unitary and is diagonal with non-negative entries. We can always nd this decomposition because Kh is positive semidenite. Then
Tse and Viswanath: Fundamentals of Wireless Communication
46
dene y[n] = U y = U h[n]x[n] + U w[n] = h[n]x[n] + w[n]. It follows that [n] CN (0, ) and w[n] CN (0, N0 I2 ). Now we can use the error probability h expression found in b) but with uncorrelated fading gains, which in the case of circularly symmetric complex Gaussian random variables implies independence, and use a high SNR approximation for the Q function: Pe = E Q = E e|h1 |
2 h 2 SNR
2 SNR
E e
2 SNR
h
2 SNR
E e|h2 |
=
1 1 (1 SNR + 1) (2 SNR + 1)
(3.18)
where 1 and 2 are the diagonal elements of (the eigenvalues of Kh ). These eigenvalues can be computed explicitly. Assuming R[0] = 1 and letting = R[ d/v W ] we obtain 1 = 1 + || and 2 = 1 ||. If || > 0 then i > 0 (i = 1, 2), and we can further approximate (3.18) for high SNR: Pe 1 1 2= 1 2 SNR (1 ||2 )SNR2 (3.19)
In this case we get a diversity gain of 2 and the correlation between antennas increases the error probability by the factor 1/(1 ||2 ) as compared to the uncorrelated antenna case. If on the other hand || = 1 (perfect correlation 1 between antennas) then 2 = 0 and Pe 2SNR . In this case the diversity gain reduces to 1. As we increase the antenna separation d the correlation || decreases since the correlation function R[m] is monotonically decreasing, and the probability of error decreases. Exercise 3.25. We have yt = h1 h2 hL x[1] x[2] x[3] 0 x[1] x[2] x[3] ... ... + w 0 0 0 0 0 x[1]
Now, we want to argue that the typical decoding error is when all the channel gains hi s are small (i.e. |hi |2 < 1/SNR). Consider the following sub-optimal decoder: Suppose |h1 |2 > 1/SNR1 : then using the rst received symbol alone, we can decode x[1] and then using the second symbol, after subtracting of x[1], decode x[2] and so on.
Tse and Viswanath: Fundamentals of Wireless Communication
47
Now suppose |h1 |2 < 1/SNR but |h2 |2 > 1/SNR1 , then we use the second received symbol to decode x[1]. Note that since the rst channel tap is small the interference caused in the second symbol because of x[2] is also small, and hence we can decode x[1] using the second received symbol. Similarly, we can use the third symbol to decode x[2]. And so on for |h1 |2 , |h2 |2 , , |hm |2 < 1/SNR and |hm+1 |2 > 1/SNR1 : we use the m + 1th symbol to decode x[1]. For this sub-optimal decoder because of the tail behavior of the Q-function, even if one of the channels gains is larger that 1/SNR1 , then the probability of error decays exponentially in SNR. Thus, typically error happens only when all the channel gains are small which gives us a best possible diversity order of L. Exercise 3.26. 1. With x[0] = 1, det(XA XB )(XA XB ) 4L and P {XA XB } SNRL for any pair of codewords that dier in the rst component. Hence by the union bound, the probability of error p0 on the rst symbol is p0 < 2(L1) SNRL ( 1L ). 2SNR
2. To get the same rate using the naive scheme, one has to use 2L - PAM. The distance between constellation points is of the order of 2L . Hence , the error probability is of the order of L 4 2L SNR Hence, the rst scheme uses a factor of 2(L+1) less energy than the naive scheme for the same error probability. This coding gain is exponential in L (linear in L in dB.) 3. Even if we are trying to calculate the error probability for a middle stream, the determinant is still lower bounded by 4L (consider the rst non-zero stream). Then, the probability of error is upper bounded by: 2(N 1) SNRL . Exercise 3.27. 1. The channel matrix is h0 0 h1 h0
Tse and Viswanath: Fundamentals of Wireless Communication
48
2. The ZF equalizer would only look at y [0] since there is interference on the y [1] component. But then it makes an error whenever h0 is in deep fade. Hence, its diversity is 1. This is less than the diversity gain of 2 for the ML approach.
Exercise 3.28. The DFT of the fading coecients {h }L=01 after extending the sequence adding N L zeros is:
1 hn = N
L1
h ej 2
=0
n/N
(3.20)
N for n = 0, 1, . . . , N 1. Since {h }L=01 are i.i.d. CN (0, 1/L) it follows that {hn }n=01 are circularly symmetric jointly complex Gaussian, so their statistics are completely specied by the correlation function
R[r] = E [hn+r h ] = n 1 = NL
L1
1 N
L1 L1
E [hm h ]ej N [m(n+r)nl]
=0 m=0
2
e
=0
j 2 r N
1 sin = N L sin
rL N r N
ej
r (L1) N
(1 [r]) +
1 [r] N
(3.21)
valid of |r| (N 1)/2. The coherence bandwidth is given by Wc = W/(2L) and the tone spacing is W/N , so the coherence bandwidth expressed in number of tones is N/(2L). We expect the gains of the carriers separated by more than the coherence bandwidth to be approximately independent. To state this more precisely, we want to show that |R[r]| R[0] for N/(2L) |r| (N 1)/2, where we have assumed N to be odd for simplicity, and is some constant, say 10. Note that the carrier gains are periodic in n so the correlation function is also periodic in r with period N , so we need to consider values of r only within one period. To prove this statement we need to upper bound |R[r]| for N/(2L) |r| rL (N 1)/2. We rst note that | sin N | 1 and | sin(x)| 2|x|/ for |x| /2, r so sin N 2|r|/N /L in the range of interest. Therefore for N/(2L) |r| (N 1)/2, 1L 1 1 |R[r]| = = R[0] (3.22) NL N It follows that the correlation decays at least as 1 for tones separated by times the coherence bandwidth. Exercise 3.29. In outdoor environments the dierence in the distances travelled by dierent paths is of the order of a few hundred meters. For example, for a path dierence of 300m the delay spread is 1s, so we can say in general that Td is in the
Tse and Viswanath: Fundamentals of Wireless Communication
49
order of a microsecond. On the other hand Tc 1/(4D) 4.5ms for a mobile speed of 60km/h and a carrier frequency of 1GHz. We see that Tc >> Td in typical scenarios. OFDM: OFDM assumes that the channel remains constant during the transmission of an OFDM block, which requires N Tc W . The overhead (both in time and power) incurred by the use of the cyclic prex is L/(L + N ), where L = Td W is the number of taps. To have small overhead we need L N , and this implies Td Tc . Thus the underspread condition is required for a small overhead. DSSS: The Rake receiver requires the channel to remain constant during the transmission of the spreading sequence of length n, so n Tc W . Also we require n L for the ISI to be negligible. These two conditions together require that Td Tc . Channel Estimation: In both systems we need to estimate the channel. We dene Lcrit = K E /N0 as a threshold for determining the channel estimation performance. The channel estimation error is small when L Lcrit , or equivalently when Td K E /(W N0 ). Also K Tc W for the channel to remain constant during the measurement interval. Thus we need Td Tc E /N0 . A large ratio Tc /Td makes the channel estimation easier by requiring a smaller SNR. Note that in the case of OFDM we have control over the number of carriers over which we spread our energy, so it is possible to have small estimation error even when the condition L Lcrit is not met. Exercise 3.30. 1. We know that the taps h [i]s are circularly symmetric. For xed time i and dierent values of the taps are independent random variables. This follows because the dierent taps correspond to dierent signal paths which experience independent reections (attenuations and phase shifts). The complex gain of the nth carrier at the ith OFDM symbol, for an OFDM block length of N and L taps is given by: 1 hn [i] = N
L1
h [i]ej 2
=0
n/N
(3.23)
The multiplication by the complex exponential does not modify the distribution of h [i]. Also, since the L terms in the sum are independent, the multiplication by the complex exponentials does not modify the joint distribution of the dierent terms, and hence does not modify the distribution of the sum. Since the dependence on n only appears through the complex exponential, and we can remove the complex exponentials without modifying the distribution of hn [i], it n [i]}N are identically distributed for xed i. follows that the {h n=1 2. From the text, we know that the eect of movement of paths from tap to tap is negligible compared to the variation of the taps due to Doppler spread whenever fc W . Making this assumption the movement of paths between taps occurs in a time scale much larger than the variation due to Doppler shifts, and therefore
Tse and Viswanath: Fundamentals of Wireless Communication
50
we have that the processes {h [m]]}m are independent across . Now we can use an argument similar to that of part (1) but using vectors. Let hn = [hn [1] . . . hn [r]]T , n = 1, 2, . . . , N , and h = [h [1]h [2] . . . h [r]]T , = 0, 1, . . . , L 1. We can write: 1 hn = N
L1
h ej 2n /N
n=0
(3.24)
We can use the result of Exercise 2.16 to conclude that the vectors h , = 0, 1, . . . , L 1 are circularly symmetric. Hence the distribution of each of the terms in the above sum is not modied by the product with the complex exponential. Also the initial observation implies that the vectors h , = 0, 1, . . . , L 1 are independent, and therefore the sum (3.24) is not modied by the product with the complex exponentials. Since the dependence on n is only through the complex exponentials, and these can be removed without modifying the distrin bution of the sum, it follows that {hn }N=1 are identically distributed. Since r is arbitrary the result follows. Exercise 3.31. Let r = (rT rT )T , where rA and rB are as dened in Exercise 3.6. AB Dene C=
Eb L
+ N0 I L , D = N0 I L , A =
C0 0D
, B =
D0 0C
(3.25)
Then (r | xA ) CN (0, A ) and (r | xB ) CN (0, B ), so noting that det A = det B the log-likelihood ratio is given by: L(r) = log p(r | xA ) = r (1 1 )r = ( rA B A p(r | xB
2
rB 2 )
Eb (3.26) N0 (Eb + LN0 )
By comparing the log-likelihood ratio to 0 we obtain the decision rule (3.63) of the notes. We now compute the error probability: Pe = P r wA
2
rA
2
> rB
2
2
| xB = P r
wA
2
>
Eb h + wB
2
(3.27)
and
Eb h+wB
have 2 distributions with densities f1 (x) =
1 and f2 (x) = (N0 +Eb /L)L (L1)! xL1 ex/(N0 +Eb /L) for x 0. Letting T = wB 2 with density:
1 xL1 ex/N0 L N0 (L1)! wA 2 Eb h +
fT (t) =
f1 (l)f2 (l t)dl
(3.28)
Tse and Viswanath: Fundamentals of Wireless Communication we can express the error probability by: Pe = P r [T > 0] =
0
l 1 lL1 (l t)L1 l/N0 N +Et /L e e 0 b dldt 2 [N (N + E /L)]L (L 1)! 0 0 b 0 t L1 L1 s 1 (s + t) s e(s+t)/N0 e N0 +Eb /L dsdt (L 1)!2 [N0 (N0 + Eb /L)]L
51
=
0
1 = (L 1)!2 = = = 1 (L 1)!2 1 (L 1)!2 N0
L1
k=0 L1
L1 k L1 k L1 k L+k1 k
0 0
sk tL1k sL1 t/N0 s e e L
L1k
1 + N +1 /L N0 0 Eb
(3.29) dsdt (3.30) (3.31)
k=0 L1
(L + k 1)! L+k
t
0
L
et/N0 dt
k=0 L L 1 k=0
(L + k 1)! (L 1 k )! Lk N0 L+k L 1 N0
k
1 where we have dened = N0 + N0 +1 b /L and = N0 (N0 + Eb /L). Here (3.29) is E obtained by the binomial expansion formula, and (3.30) and (3.31) follow by integrating the 2 densities over their domains. This last formula is just what is required.
Exercise 3.32. Dene rA and rB by: rA = h[ ]x1 + wA ,
() rB () ()
()
()
= 1, 2, . . . , L = 1, 2, . . . , L (3.32)
= h[ ]x2 +
() wB ,
where x1 = xA and x2 = 0 if xA is transmitted, and x1 = 0 and x2 = xB if xB is transmitted. () Dene rk as in (3.152) in the text. Then the channel estimates are: h[ ] = E /N0 1 K E h[ ] + w ( ) K E /N0 + L E = ah[ ] + w( )
K E /N0 K E /N0 +L
(3.33) (3.34) and w( )
for
= 0, 1, . . . , L 1, where w( ) CN (0, KN0 ), i.i.d., a = The coherent receiver projects rA and rB onto h obtaining: yA = h rA = a h 2 x1 + w hx1 + (ah + w) wA A
KE CN 0, (K E /N/N0L)2 , i.i.d.. 0+
Tse and Viswanath: Fundamentals of Wireless Communication yB = h rB = a h 2 x2 + w hx2 + (ah + w) wB B
52
(3.35) where x1 = xA and x2 = 0 if xA is transmitted, and x1 = 0 and x2 = xB if xB is transmitted. We assume xA 2 = xB 2 = Eb = E . Also wA and wB are independent CN (0, N0 IL ) random vectors, independent of everything else. The probability of error is given by: Pe = P r ( {yA } > {yB } | xB ) = Pr = Pr {(ah + w) wA } > {a h (ah + w) wA wB E
2
E+
E w h + (ah + w) wB }
2
> {a h
+ w h}
(3.36)
Since it is hard to compute the above probability explicitly, we opt for computing it approximately by simulation.
Probability of error as a function of the number of paths. Eb/N0=10 dB, K=5 0.08
0.07
0.06
0.05
Error Probability
0.04
0.03
0.02
0.01
0
0
20
40
60
80 100 120 Number of paths L
140
160
180
200
Figure 3.2: Probability of error as a function of the number of paths for E /N0 = 10dB and K = 5. We see in Figure 3.2 that the probability of error starts degrading for L 15. For this particular choice of the parameters Lcr = K E /N0 = 50 so for L Lcr the performance of the detector is very poor. Exercise 3.33.
Chapter 4 Solutions to Exercises
Exercise 4.1. 1. For this example C=7 and M=10. The maximal allowable subsets are enumerated as below, S1 S2 S3 S4 S5 S6 S7 S8 S9 S10 = = = = = = = = = = {7} {1, 3} {1, 4} {1, 5} {2, 4} {2, 5} {2, 6} {3, 5} {3, 6} {4, 6}
2. The matrix A can be represented as, 011 0 0 0 0 1 0 A = 0 0 1 0 0 0 0 0 0 100
1 0 0 0 1 0 0
0 1 0 1 0 0 0
0 1 0 0 1 0 0
0 1 0 0 0 1 0
0 0 1 0 1 0 0
0 0 1 0 0 1 0
0 0 0 1 0 0 0
Exercise 4.2. 1. Let Ti (r) be the average trac per channel supported in the cell i. Then since the average trac supported cannot exceed the actual trac present 53
Tse and Viswanath: Fundamentals of Wireless Communication in cell i we have, Ti (r) Average trac per channel in cell i = pi r. Therefore adding over all the cells we have,
C
54
T (r) =
i=1 C
Ti (r), pi r,
i=1 C i=1
= = =
pi B
N
C i=1
, I {Channel l used in cell i} , N
N l=1
where I { Channel l used in cell i} is the indicator function which is equal to 1 if channel l is used in cell i and zero otherwise. Exchanging the order of summation, we have, 1 T (r) N
N C
l=1 i=1
I{ Channel l used in cell i}
Now each channel l is assigned to a subset of one of the maximal sets (note that it does not have to be a strict subset). Lets assume that the channel l is assigned to a subset of the maximal set Sl which corresponds to the column j in the adjacency matrix. Therefore, 1 T (r) N = 1 N
N
1,
l=1 iSl N C
aij ,
l=1 i=1 N j =1,...,M C
1 N
max
aij ,
i=1
l=1
1 N max N j =1,...,M
C
aij ,
i=1
Tse and Viswanath: Fundamentals of Wireless Communication
C
55
=
j =1,...,M
max
aij .
i=1
2. As noted in part (1) above: Let Ti (r) be the average trac per channel supported in the cell i. Then since the average trac supported cannot exceed the actual trac present in cell i we have, Ti (r) Average trac per channel in cell i = pi r. Then, summing over all the cells, we get
C
Ti (r)
i=1
pi r = r.
3. To combine the two upper bounds, simply choose a set of numbers yi [0, 1] for i = 1, . . . , C and observe that Ti (r) pi r = yi pi r + (1 yi )pi r. Then, summing over all i = 1, . . . , C and applying the bound from part (1) to the second term and bound from part (2) to the rst term, we obtain
C C
Ti (r)
i=1
yi pi r + max
j =1,...,M
(1 yi )aij .
i=1
Exercise 4.3. Exercise 4.4. 1.
s(t) = R
n=0
x[n]sinc sinc
n=0
t nT T
exp(j 2fc t)
=R
t nT T
exp(j 2fc t + n )
where x[n] = exp(jn ) and n is uniformly distributed on [0, 2 ] and independent across time samples n. Now, the average power in s(t) over a symbol period is given by Pav 1 =E T
T 0
|s(t)|2 dt
Tse and Viswanath: Fundamentals of Wireless Communication
56
|s(t)|2 = =
sinc
n=0
t nT T t nT T
2
cos(2fc T + n ) cos(2fc T + n )sinc t mT T cos(2fc T + m )
sinc
n=0 m=0
By independence of n and m for n = m we have that
E[|s(t)| ] =
n=0
2
sinc2 1 sinc2 2
t nT T t nT T
E[cos2 (2fc t + n )]
=
n=0
Observe that the series above is uniformly (and absolutely) convergent on 0 t T . To see this note that
sinc2
n=0
t nT T
=
n=0
sin2
t T
t T
n
2 2
n 1
n=0
t T
n
n=1
1 n2
Hence, we can interchange the summation and integration in computing the average power: Pav 1 = T = 1 2
T 0
n=0 n+1 u=n
1 sinc2 2
t nT T 1 2
1 0
1 dt = 2T
0
T
sinc2
0
n=0
t nT T
dt
sinc2 (u)du =
sinc2 (u)du +
n=0
1 2
sinc2 (u)du
nT where u = tT . Using Parsevals identity, the energy in sinc(u) is the same as the energy in its Fourier transform rect(f ) so
Pav =
11 + 42
1 0
sinc2 (u)du
Using MATLAB, the above quantity can be calculated to be Pav 0.476. To estimate the peak-power as a function of T , suppose that the maximum of |s(t)|2 over [0, T ] occurs at some t0 [0, T ]. Then we can write:
0tT
max |s(t)|2 =
n=0
sin t0 nT T t0 nT T
2
cos(n )
,
Tse and Viswanath: Fundamentals of Wireless Communication
57
where we have used the fact that 2fc T is an integer. The time t0 is a random variable since n is random. However, we can obtain the following simple upper bound by expanding the above expression: max |s(t)|
2
0tT
n=2 m=2 sin t0 n T + t0 T n n=0,1 m=0,1
1 2
t0 T
cos n cos m n t0 m T sin t0 m T t0 T m cos n cos m , (4.1) (4.2)
1 2
cos n cos m (1 n) (1 m) n=2 m=2
+
n=0,1 m=0,1
cos n cos m ,
t In the rst inequality above, the rst term was obtained by upper bounding sin T n t by 1. In the second inequality, we use the observation that sinc T n is maximized at t = nT (and this is feasible since n = 0, 1) where it evaluates to 1, to obtain the t second term. The rst term is upper bounded by noting that the minimum of | T n|, for n = 0, 1, occurs when t = T . Now, taking the expectation of the above upper bound over the distribution of the data symbols and observing that n and m are independent for n = m, we get that:
E max |s(t)|2
0tT
1 2 2
n=2
1 + 1. (n 1)2
Hence P P := E max |s(t)|2
0tT
1 2 + 1 1.083. 2 2 6
PP The Peak-to-average-power ratio Pav is then approximately 2.77. Also, from (4.1), we see that as T , the peak-power approaches this value.
Exercise 4.5. From equation (4.31), for each k = 1, . . . , K , we get, GPk gk,ck k
n=k
Pn gn,ck + N0 wk
Pk
1 G
Pn
n=k
k gn,ck gk,ck
N0 wk gk,ck
Tse and Viswanath: Fundamentals of Wireless Communication
58
Using the notation p to denote the vector (P1 , . . . , PK ), we can rewrite the above K inequalities as, (Ik F) p = b where, b = N0 W and F is a K K matrix given by, fij = 0
gj,ci i Ggi,ci
1 K ,... g1,c1 gK,cK
if i = j if i = j
Exercise 4.6. 1. To see that F is irreducible, simply take m = 2 and observe that, for K 3, we have that Fm is a matrix with strictly positive entries. Hence, by denition, F is irreducible. 2. The reference [106] gives an in-depth treatment of non-negative matrices. 3. The following proof is taken from [106]. First we show that (a) implies (b). Suppose that, for a given strictly positive constraint b > 0, there exists a strictly positive solution p > 0 to (4.32), i.e., (I F)p b. Then, b + Fp p, and hence Fp < p. Now, since F is irreducible and nonnegative it has a unique left (and right) strictly positive eigenvector xT > 0 associated with the Perron-Frobenius eigenvalue r. Pre-multiplying both sides of the inequality p > Fp by this eigenvector, we get that xT p > xT Fp = r xT p, and hence we conclude that r < 1. Now, we show that (b) implies (c). Suppose that the Perron-Frobineus eigenvalue r < 1. Hence, all other eigenvalues i satisfy |i | < 1 for i = 1, . . . , K 1. Here, for simplicity, we assume that all the eigenvalues are distinct. The arguments are only slightly more technical if this condition is violated. We can express F = B1 B1 , for some matrix B, where = diag(r, 1 , . . . , K ). Hence, we can write Fn = B1 diag(rn , n , . . . , n )B, 1 K
Tse and Viswanath: Fundamentals of Wireless Communication
59
for n 1 integer. Thus, we see that rn 0, and n 0, for all i, as n . i Hence, Fn 0 elementwise. Recall the following matrix lemma: If Fn 0 elementwise, then (I F)1 exists and
(I F)
1
=
n=0
Fn ,
(4.3)
where the convergence takes place elementwise. Also observe that, by irreducibility, for any pair (i, j ), (Fn )ij > 0 for some n as a function of (i, j ). Hence, the right-hand side of (4.3) is a strictly positive matrix. Hence we conclude that (I F)1 > 0. Hence we have established (c). To prove that (c) implies (a), note that if (I F)1 > 0, then (I F)1 b > 0, for any b > 0. Let p = (I F)1 b and hence p > 0. Thus, we have proved the equivalence of the three statements (a), (b) and (c). Exercise 4.7. 1. Given that c, t(1) is feasible, we know that there exists a vector (1) (1) (2) p(1) of powers such that user k s b /I0 meets the target level of k . If k k then the same vector of powers, p(1) will satisfy the threshold requirements for each mobile as the cell allocation remains unchanged. Hence c, t(2) is also feasible. 2. The feasibility of c(3) , t(3) is evident from the denition of k . Now it remains (3) to be shown that k k . Let us consider a particular user r. Without loss of (1) (3) (2) (1) generality, assume that pr pr . Then the new assignment will be cr = cr . Therefore,
(3) r (3)
=
gr,c(3) pr r N0 W +
k =r
(1) (3)
gk,c(3) pk
k
=
gr,c(3) pr r N0 W +
k =r (1) (2)
(1)
and
gk,c(3) pk
kI1
k
(3)
+
k=r
and
(1)
gk,c(3) pk
kI2
k
(3)
where I1 = {k |pk pk } and I2 = {k |pk corresponding cell assignment we have,
(3) r
(2)
pk }. Therefore with the
=
gr,c(3) pr r N0 W +
k =r
(1)
and
gk,c(1) pk
kI1
k
(1)
+
k=r
and
gk,c(2) pk
kI2
k
(2)
Tse and Viswanath: Fundamentals of Wireless Communication
(3)
60
On comparing the above expression for r with r we see that the numerator (3) remains the same in both the cases but the denominator in r is reduced due to (3) the new cell assignment. Hence it follows that r r for all mobiles 1 r K 3. Given that uplink communication is feasible we know that there exists at least one vector of powers, p(1) such that the threshold requirements are made. If this is the only solution possible then its trivially the unique coordinate-wise minimum vector of power that allows successful communication. On the other hand if there is another vector of powers which is feasible, say p(2) , then using the cell and power allotment in part (2) we get a new power vector p(3) each coordinate of which is less than or equal to the coordinates of p(2) and p(2) . Now applying the algorithm in part (2) to p(3) and the other feasible vectors and proceeding in a similar fashion as above, we can conclude that there is a coordinate-wise minimum vector of powers. Exercise 4.8. 1. (a) From the denitions, gnl 0 for all n and l Sn , and (m) (m) Inl 0 for all n,l Sn if pn 0 for all n. Hence n Inl = min 0 lSn Ggnl
(m)
p(m+1) n
Thus I (p) 0 for every p 0. (b) Suppose p p (component-wise dominance). Then we have that Inl =
k=n (m) (m) (m)
gkl pk
+ N0 W
k=n
gkl pk
(m) + N0 W := Inl
Hence
(m) (m) n Inl n Inl I (p) = min min = I (p). lSn Ggnl lSn Ggnl
(c) Assume > 1 n Inl (p) I (p) = min lSn Ggnl Inl (p) =
k=n (m) (m)
gkl pk
(m)
+ N0 W <
k=n
gkl pk
(m)
+ N0 W = Inl (p)
(m)
Hence I (p) < I (p).
Tse and Viswanath: Fundamentals of Wireless Communication
61
2. Suppose I has a xed point p1 , i.e., I (p1 ) = p1 . Assume that there exists another xed point p2 0 such that p1 = p2 . Then, without loss of generality, we can assume that there exists at least one j {1, . . . , K } such that p1 (j ) < p2 (j ). p2 (j ) p2 (j ) Now, observe that := maxj {1,...,K } p1 (j ) > 1. Let i = arg maxj {1,...,K } p1 (j ) and so p2 (i) = p1 (i). Consequently p1 p2 . Hence, we have that p1 = I (p1 ) > I (p1 ) I (p2 ) = p2 . But, by denition, is such that p1 (i) = p2 (i). Hence we have contradiction and it must be that p1 = p2 . 3. Suppose p is the unique xed point of I . Since p (j ) > 0 for all j , given any initial p, we can nd 1 such that p p. Now, from question part (1), (c), I (p ) < I (p ) = p . Now take z to be the all-zero vector. It is clear that z p p . By part (b) of question 1, we know that I (n) (z) I (n) (p) I (n) (p ). Claim 1: I (n) (p ) p as n . Proof: From above we have I (p ) < p . Assume that I (n) (p ) < I (n1) (p ). Then by part (b) of question 1, we have that I (I (n) (p )) I (I (n1) (p )), i.e., I (n+1) (p ) I (n) (p ). Hence I (n) (p ) is a decreasing sequence (componentwise). It is also bounded away from zero so it must converge to the unique xed point p . Claim 2: I (n) (z) p as n . Proof: Clearly z < p and I (z) z. Suppose that z I (z) . . . I (n) (z) p . Then by part (b) of question 1, we have p = I (p ) I (I (n) (z)) I (I (n1) (z)) = I (n) (z). In other words we have shown that p I (n+1) (z) z. Therefore, the sequence I (n) (z) is nondecreasing and bounded above by p hence it must converge to p . Combining Claim 1 and Claim 2 together, we get that I (n) (p) p for any initial vector p. 4. Using the identical argument as in the proof of Claim 1, we observe that I (n) (p) is a decreasing sequence that converges to the xed point p for any feasible vector p. Hence it follows that p p for any feasible vector p. In other words, the xed point p is the solution to p I (p) corresponding to the minimum total transmit power. Exercise 4.9.
Tse and Viswanath: Fundamentals of Wireless Communication
62
Exercise 4.10. 1. We assume that there is only 1 base station. The set of feasible power vectors is given by: A= (P1 , P2 , . . . , PK ) : GPk gk,1 0 Pk P , k = 1, 2, . . . , K Pj gj,1 + N0 W j =k
2. For K=2 we need: 0 P1 P , 0 P2 P , G g1,1 GP1 g1,1 P1 P2 g2,1 + N0 W g2,1 GP2 g2,1 g1,1 P1 P1 g1,1 + N0 W G g2,1 For example let N0,W = 1, g1,1 = g2,1 , P = 5, for g2 1 power vectors is shown in gure 4.1. N0 W P2 G g2,1 N0 W P2 g2,1
G
= 2 the set A of feasible
For this example the set of feasible power vectors is non-empty, with componen twise minimum solution (P1 , P2 ) = (1, 1). Clearly if P < 1, A = . For a more G interesting example of an outage situation, let = 1 with the above choices for the other parameters. As shown in Figure 4.2 the feasible set of power vectors empty. is 3. For the parameters g1,1 = g2,1 = 1, G = 2, N0 W = 1 and = 1 we have that the component-wise minimum feasible power vector is (P1 , P2 ) = (1, 1). We plot in Figure 4.3 the evolution of P1 and P2 over time when the power control algorithm is run with a probability of error of 103 in the power control bit for each user. 4. We rst redo part (1). Let K be the number of users, M the number of base stations, and gk,m the gain from user k to base station m. Then the set A of feasible power vectors to support a given Eb /N0 = requirement under soft hando in the uplink can be expressed as: A = {(P1 , P2 , . . . , PK ) : GPk gk,m
N
j =1 j =k
Pj gj,m + N0 W
, for some m, 1 m M, (4.4) 0 Pk P , k = 1, 2, . . . , K }
which can be specialized for the case of 2 base stations by choosing M = 2. We now redo part (2). Let A = (P1 , P2 ) : P2 C = (P1 , P2 ) : P2
1 g2,1 GP1 g1,1
N0 W
, B = (P1 , P2 ) : P2
Gg2,2
1 g2,2
GP1 g1,2
N0 W
,
Gg2,1
(P1 g1,1 + N0 W ) , D = (P1 , P2 ) : P2
(P1 g1,2 + N0 W ) ,
Tse and Viswanath: Fundamentals of Wireless Communication
63
mnP2 = 2P1 1 mnP2
5
1
mnA 11111111111111111 00000000000000000 11111111111111111 00000000000000000 11111111111111111 00000000000000000 11111111111111111 00000000000000000 11111111111111111 00000000000000000 11111111111111111 00000000000000000 11111111111111111 00000000000000000 mn(P , P ) = (1, 1) 11111111111111111 00000000000000000 11111111111111111 00000000000000000
1 2
1 mnP2 = 2 P1 +
1 2
1 1
5
mnP1
Figure 4.1: Set A of feasible power vectors for M=1 and K=2. The gure corresponds to N0,W = 1, g1,1 = g2,1 , P = 5, and G = 2. Here (P1 , P2 ) is the componentwise g2 1 minimum power vector in the feasible set.
Tse and Viswanath: Fundamentals of Wireless Communication
64
mnP 111111111111111 000000000000000= P + 1 mnP 111111111111111 000000000000000 1111111111111111 0000000000000000 111111111111111 000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 mnP = P 1 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 mnA = 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 mnP 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000 1111111111111111 0000000000000000
2 1 2
5
2
1
1
1
5
1
1
Figure 4.2: Empty set A of feasible power vectors for M=1 and K=2. The gure corresponds to N0,W = 1, g1,1 = g2,1 , P = 5, and G = 1. g2 1
Tse and Viswanath: Fundamentals of Wireless Communication
5 P1 P2 Minimum feasible power
65
4.5
4
3.5
Power
3
2.5
2
1.5
1
0.5
0
5
10
15 time
20
25
30
Figure 4.3: Trajectory of power updates over time for a power control bit error probability of 103 . E = (P1 , P2 ) : 0 P1 P , 0 P2 P . Then for K = M = 2 we can express A as A = ((AB )(C D))E . For the case of G/ = 2, g1,1 = g1,2 = g2,1 = 1/4, g2,2 = 1/2, P = 5, and N0 W = 1, the corresponding set of feasible power vectors is shown in Figure 4.4. We see that there is a componentwise minimum power vector (P1 , P2 ) = (20/7, 12/7) in the feasible set. Also if G/ < min
g1,1 g2,2 , g1,2 g2,1 g1,2 g2,1 g1,1 g2,2
then A = .
5. With the parameter choices of part (4) we plot in Figure 4.5 the evolution of P1 and P2 over time when the power control algorithm is run with a probability of error of 103 in the power control bit for each user. 6. This is not true. What determines the choice of base station is the signal to interference plus noise ratio (SINR), and not the channel gain. If the channel gain from a given user to a particular base station is large but that base station is supporting many users, then the SINR may be small, even smaller than , in which case the given base station cannot be chosen. On the other hand there can be another base station for which the channel gain is not so large, but if the interference seen by this user in the base station is small, then the SINR can be large, possibly exceeding , in which case the base station can be chosen to serve the user. Exercise 4.11. 4. We analyze the CDMA system rst. The cell model for this
Tse and Viswanath: Fundamentals of Wireless Communication
aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa aaaaaaaaaaaaaaaaaaaaaaaaaaaaa
66
Figure 4.4: Set A of feasible power vectors for M = K = 2. The gure corresponds to G/ = 2,
g1,1 = g1,2 = g2,1 = 1/4, g2,2 = 1/2, and N0 W = 1. Here (P1 , P2 ) is the componentwise minimum power vector in the feasible set.
problem is shown in gure 4.6. Here it is assume that users are perfectly power controlled in their respective cells. Therefore the received power at each base station from the users in the cell is equal to some constant Q, independent of the user. Neglecting noise we have GQ (k 1)Q +
K k=1
Q( r2,k ) 1,k
r
(4.5)
for any user of cell 1 that it is not in the outage. Then
K
G k1+
k=1 K
(
r2,k ) r1,k
k=1
(
r2,k G ) (K 1) r1,k
If this condition is not met there is an outage. Then
K
Pout = P r{
k=1
(
r2,k G ) > (K 1)} r1,k
Tse and Viswanath: Fundamentals of Wireless Communication
3.5
67
3
2.5
Power
2
1.5 P1 P2 P1* * P2
1
0.5
0
5
10
15 time
20
25
30
Figure 4.5: Trajectory of power updates over time for 2 base stations, for a power control bit error
probability of 103 .
Let X =
r
r2,k K k=1 ( r1,k ) .
For large K we approximate X by a Gaussian random variable v ar(X ). We rst note that
r r
with mean = E (X ) and standard deviation = { r2,k }k are iid. Then = KE ( r2,k ) and = 1,k 1,k
K var ( r2,k ) . It is not possible 1,k
to obtain a closed form expression for and for general . We will approximate r1,1 by E (r1,1 ) = d then 2K r2,k )= KE ( r1,k d = and
2
d 2
0
1 r dr2.1 d 2,1
2K 1 d d
d 2
+1
1 K1 = +1 2 +1
K 2 2 d K 1 = 2 2 d d =
d 2
0
1 2 r dr2,1 d 2,1 d 2
2+1
d 2
2
1 ( + 1)2
2
1 2 + 1
d 2
1 ( + 1)2
1 1 K 2 2 + 1 2 ( + 1)2
Tse and Viswanath: Fundamentals of Wireless Communication mn 1111 0000 11 0r 0 mnr 11 00 11111111111111 00000000000000 1 0 11 00 1 0 11 00 1 0 11 00 1 0 11 00 1 0 1 0 11111111111111111111 00000000000000000000 1 0 1 0 1 0 1 0 11111111111 00000000000 1 0 1 0 1 0 1 0
2,k 1,k
68
cell #1
cell #2
user k of cell #2
d/2
d/2
d
d
Figure 4.6: Cell model for Exercise 4.11 Then G 1 (K 1) G 1 (K 1) Q1 (Pout ) Pout Q Replacing G =
W R
, and we have
1 2
W 2 1 2 1 (K 1) K R +1 K 2 + 1 ( + 1)2 or 1 W 1 1 = Q1 (Ppout ) RK K 2 2 + 1 ( + 1)2 Therefore the spectral eciency f =
RK W
1 2
= Q1 (Pout )
+
2 1 +1 +1 K
is given by
1 2
1 RK 1 1 1 f= = Q1 (Ppout ) 2 + 1 W ( + 1)2 K2 Now as K and W go to we obtain 1 1 lim = 1+ K,W 2 ( + 1)
1 Also as increases, f increases to .
1 2 + +1 +1 K
1
(4.6)
1
Tse and Viswanath: Fundamentals of Wireless Communication
0.9
69
0.8
0.7
Spectral efficiency RK/W
0.6
0.5
0.4
0.3
0.2
0.1
0
10
20
30
40 50 60 Number of users in each cell K
70
80
90
100
Figure 4.7: Spectral eciency as a function of the number of users in each cell for Q1 (Pout ) = 2,
= 2 and = 7dB.
For the case that = 2 and = 7dB the resulting spectral eciency is plotted in the gure 4.7 and we have
K,W
lim f ( = 2, = 7dB ) = 0.1833
1. Now consider the orthogonal case. Since the users are orthogonal within the cell there is no intra-cell interference. Also we will assume that the users are power controlled at each base station so that the received power from all the users is the same. The out-of-cell interference is averaged over many OFDM symbols, so that each base station observes an interference that is the average over all the users of the neighboring cell. We will reuse most of the derivation of part (4). We have GQ
K k=1
Q
r2,k r1,k
for any user of cell 1 that is not in outage, where G is the processing gain, Q is the received power of a user to its base station, and ri,k is the distance of user k to base station i. Note that this is exactly the same expression found in part (4) without the term (K-1)Q in the denominator. Dening and as before, we can reuse all the expressions found in part (4) eliminating the term (K 1). We obtain Pout G 1
Tse and Viswanath: Fundamentals of Wireless Communication and RK 1 f= = W Q1 (Pout ) 2 K 1 1 1 + 2 2 + 1 (1 + ) 2 ( + 1)
70
1
as the spectral eciency. 2. As K and W go to we obtain: 2 ( + 1) lim f = K,W 3. We see in Figure 4.8 that the spectral eciency increases as the bandwidth and the number of users grow.
2.4
2.2
2
Spectral efficiency RK/W
1.8
1.6
1.4
1.2
1
0.8
0
10
20
30
40 50 60 Number of users in each cell K
70
80
90
100
Figure 4.8: Spectral eciency as a function of the number of users in each cell for Q1 (Pout ) = 2,
= 2 and = 7dB.
4. From the previous point we observe that removing the intra-cell interference made the spectral eciency grow with the bandwidth and the number of users. The intra-cell interference contributes with a term (K 1)Q to the total interference. When this interference is normalized by dividing it by the number of users we get (1 1/K )Q which increases with K . We see that the intra-cell interference per user increases with K . For example for K = 1 there is no other user in the cell, and hence there is no intra-cell interference in the CDMA case. As K increases, the normalized intra-cell interference also increases reducing the spectral eciency. This eect turns out to be more important than the interference averaging, dominating the dependence of the spectral eciency with K and W as we observed in the gure in part (4).
Tse and Viswanath: Fundamentals of Wireless Communication Exercise 4.12. Pout = P r 1. The outage probability for user 1 is given by: G 1P N j =2 j P + < = Pr
1 N j =2 j
71
N0 W
+ N0 W/P
< /G
(4.7)
Since we dont have any power constraint we can let P . Also for large N we can use the CLT to approximate N j N ((N 1), (N 1) 2 ), where j =2 2 = E ( j ) and = V ar( j ). The pdf of can be obtained by the transformation = eX where X N (, 2 ). It is given by: 1 log 2 1 f( )= e 2 ( ) (4.8) 2 for > 0. Using this density we can compute = e+ Therefore we can write: Pout = E
2 /2
and 2 = e2+
2
e 1 .
2
P Q
1 N j =2 j
< /G
1
1 log 2 1 e 2 ( ) d 2
=
0
G/ (N 1) N 1
(4.9)
We would like to compute the spectral eciency = N R/W = N/G as a function of the number of users N for a given outage probability Pout . For this we need to solve numerically the implicit function G(N ) dened by equation (4.9). 2. We show in Figure 4.9 a plot of the spectral eciency as a function of the number of users, for the parameter choices = 7dB , = 0, 2 = 0.053019 = 1/(10 log10 e)2 (which corresponds to a standard deviation of 1dB in ). As N increases the spectral eciency always decreases. There is an averaging eect but it is masked by the fact that (N 1)/N increases with N . 3. In the other examples considered in the text the only randomness in the SINR was due to the interference, which was averaged out as N increased, converging to a constant. However in this problem the power control error for the given user remains random even for large N . There is an averaging of the interference, but this is not enough to make the SINR converge to a constant. This randomness in the SINR results in a degraded spectral eciency. As a basis for comparison we have plotted in Figure 4.9 the spectral eciency that results when the user of interest has perfect power control, but the interference has the same power control error considered before. In this case we see how the interference averaging eect
Tse and Viswanath: Fundamentals of Wireless Communication
Spectral efficiency as a function of N. =5, =0, 2=0.053019, Pout=103 0.28 Imperfect power control Perfect power control for user of interest, but imperfect for interference 0.26
72
0.24
Spectral efficiency [bits/s/Hz]
0.22
0.2
0.18
0.16
0.14
0.12
0.1
0.08
0
10
20
30
40 50 60 Number of users N
70
80
90
100
Figure 4.9: Spectral eciency as a function of the number of users. eventually takes over resulting in a larger spectral eciency for large N . By comparing the two curves we see that imperfect power control produces a large performance degradation. Exercise 4.13. 1. Let hi [l] be the channels lth tap from base station i (i = A, B ) to the user of interest. Assume for simplicity that both channels have L taps, and assume that the received signals from the two base station are chip and symbol synchronous. Assuming no ISI and using the notation of section 3.4.2 of the notes we can write the received signal vector as:
L1 M L1 (l) hA [l]x1,A l=0 L 1 (l) hA [l]xm,A m=2 l=0 N (l ) hB [l]x1,B l=0 L1
y=
+
+
+
m=2 l=0
hB [l]xm,B + w
(l )
(4.10) where M and N are the number of users in cells A and B respectively. Since the user is in soft hando we can assume that the signals received from the two base stations have comparable power, and we can use a Gaussian approximation c for the interference plus noise term. Letting w CN (0, ( M=2 hA 2 EA + m N 2c c m=2 hB EB + N0 )In+L1 ), where Ei is the chip energy of the transmitted signal from base station i, and n is the symbol length. Then we can write:
L 1 L1
y=
l=0
hA [l]x1,A +
l=0
(l)
hB [l]x1,B + w
(l)
(4.11)
Tse and Viswanath: Fundamentals of Wireless Communication
uA uA
(l)
73
uB uB
(l)
L1
The received signal without the noise lies in the span of the vectors
,
.
l=1
Assuming that the spreading sequences are orthogonal, and that their shifts are also orthogonal, we have that the previous set of vectors is an orthonormal set, and we can project onto these vectors to obtain 2L sucient statistics: ri = hi [l] ui x + wi
(l ) (l)
(4.12)
where i = A, B and l = 0, 1, . . . , L 1. We can further project onto the direction of [( uA hA [0])( uA hA [1]) ( uA hA [L1])( uB hB [0])( uB hB [1]) ( uB hB [L 1])]T to obtain the sucient statistic: r= ( hA
2
uA
2
+ hB
2
uB 2 )x + w
(4.13)
c c where w CN (0, ( M=2 hA 2 EA + N =2 hB 2 EB + N0 )). Letting ui 2 = m m c GEi , i = A, B and assuming x {1, 1} we can write the error probability as: c c 2G( hA 2 EA + hB 2 EB ) M m=2 c hA 2 E A + N m=2 c h B 2 E B + N0
Pe = Q
(4.14)
2. Let N be the total number of base stations, K the total number of users, S the set of all base stations, Sk the active set of user k (i.e. the set of base stations with which user k is in soft hando), Ai the set of users that have base station i in their active sets, gk,i the channel gain from base station i to user k , and Pk,i the power of the signal transmitted to user k from base station i. Then the SINR seen by user k is given by: SINRk =
iSk
gk,i
G iSk gk,i Pk,i j Ai ,j =k Pj,i + iS \Sk gk,i
j Ai
Pj,i + N0 W
(4.15)
Assuming that there is a minimum SINR requirement for reliable communication, the set of feasible power vectors is given by:
A = {(P1,1 P1,2 P1,N P2,1 P2,N PK,1 PK,N ) : SINRk , Pk,i = 0 if i Sk , k = 1, 2, . . . K / (4.16) The power control problem consists of nding a power vector in A.
a Exercise 4.14. 1. The 2 latin squares have entries Ri,j = (ai + j ) mod N and b Ri,j = (bi + j ) mod N , where N is prime and a = b.
Consider the pair of ordered pairs: (ka , kb )i,j = ((ai + j ) mod N, (bi + j ) mod N )
Tse and Viswanath: Fundamentals of Wireless Communication (ka , kb )l,m = ((al + m) mod N, (bl + m) mod N )
74
We want to show (i, j ) = (l, m) (ka , kb )i,j = (ka , kb )l,m , that is, any 2 ordered pairs must be dierent. Let (d1 , d2 ) = (ka , kb )i,j (ka , kb )l,m . We have to show that (d1 , d2 ) = (0, 0). (d1 , d2 ) = ([a(i l) + (j m)] mod N, [b(i l) + (j m)] mod N ) (4.17)
If i = l, then we must have j = m and (d1 , d2 ) = ((j m) mod N, (j m) mod N ). Since (N 1) (j m) (N 1) and (j m) = 0 it follows that (j m) mod N = 0, therefore (d1 , d2 ) = (0, 0). If on the other hand i = l then d1 d2 = (a b)(i l) mod N . d1 d2 = 0 requires that N divides (a b) or (i l) (note that here we use the fact that N is prime). But both are in [(N 1), 1] [1, (N 1)] so they are not divisible by N . It follows that d1 d2 = 0 and hence (d1 , d2 ) = (0, 0). 2. Adapted from J. van Lint, R. Wilson, A course in Combinatorics, Second Ed., Cambridge University Press, 2001. Consider a set of M mutually orthogonal latin squares. The entries in each latin square correspond to virtual channel numbers. We are free to rename the channels so that the rst row of each latin square is (1, 2, . . . , N ). Then the pairs (kl , km )1,j for any pair of matrices (l, m) (l = m, l, m [1, . . . , M ]) are (j, j ). Now consider the (2, 1) entry of each latin square. It cant be 1 because 1 already appears in the position (1, 1). Also these elements must be dierent in all the matrices, because the pairs (kl , km ) with repeated entries have already appeared. Thus we have M (N 1). Note that N need not be prime for this result to hold. Exercise 4.15. 1. Let M = N/n.
2
1 P= T 1 = 2N
T
T 0
1 s(t) dt = T
2
T 0 T
1 2N
n1
D[i]e
i=0
j 2 (fc +iM/T )t
+ D[i] e
T 0
j 2 (fc +iM/T )t
dt
n1 n1
i=0 k=0
1 T
D[i]D[k ]ej 2(2fc +(i+k)M/T )t dt +
T 0
D[i]D[k ] ej 2((ik)M/T )t dt (4.18)
0
+
0
D[i] D[k ]ej 2((i+k)M/T )t dt +
D[i] D[k ] ej 2(2fc +(i+k)M/T )t dt
The magnitudes of the rst and last integrals of each term can be shown to be upper bounded by T |D[i]|2 (2 )1 , so when divided by T they are negligible.
Tse and Viswanath: Fundamentals of Wireless Communication The other 2 integrals evaluate to T |D[i]|2 i,k . Therefore we can write: 1 P= 2N
n1
75
|D[i]|2
i=0
(4.19)
When the symbols are chosen from an equal energy constellation with |D[i]| = 1 we get Pav = 2n . N 2. 1 d[i] = N
n1
D[kN/n]ej N
k=0
2 ikN n
(4.20)
The symbols D[i] are chosen uniformly on the unit circle, so their distribution is invariant to rotations in the complex plane. Therefore their distribution is circularly symmetric, and so is the distribution of their sum. Also the rotations induced by the complex exponential in the IDFT do not change the resulting distribution, so d[0], . . . , d[N 1] are identically distributed. As N with the ratio n/N = kept constant, we can apply a version of the CLT for circularly symmetric random variables to conclude that the distribution of d[i] converges to a circularly symmetric complex Gaussian distribution. Since E [|D[k ]|2 ] = 1 we get that d[i] CN (0, ) for large N . c) Since |d[0]|2 / Exp(1) and Pav / = 1/2 we can write Pr |d[0]|2 < ( ) Pav = Pr |d[0]|2 < ( )Pav / = 1 e()/2 = 1 (4.21)
Thus ( ) = 2 loge . For the special case of = 0.05 we obtain (0.05) = 5.99. Exercise 4.16.
Chapter 5 Solutions to Exercises
Exercise 5.1. See handwritten solutions in le s04.h6sol 1.pdf (ex 3, part b)) Exercise 5.2. The received SNR at the base-station of the user at the edge of the cell (at distance d from it) is given by: SNR = P , N0 W d
where it is assumed that the bandwidth allocated to the user is W Hz, > 2 is the path-loss exponent and N0 is noise variance. Using a reuse ratio of 0 < 1, the closest base-station reusing the same frequency as the given base-station is at distance 2d/. Since we are dealing with a linear arrangement, there are only 2 such basestations (the interference due to the others are signicantly smaller and are going to be ignored). Thus, the received interference is given by I = 2P hence the received SINR is: SINR = and so f := 2
. 2 P d
d
,
N0 W + 2
()
2d
P
=
SNR +2
2
SNR
,
76
Tse and Viswanath: Fundamentals of Wireless Communication
77
Solution 5.2: First, we observe the following simple method of evaluating the distance between
mn = 5/6
mnm = 1
mnl = 3
(centers of) any two cells in a hexagonal packing of the plane: r(l, m, ) = 2d 3m2 + l2 + 2 3lmcos, where 2d is the distance between the centers of two adjecent cells and the triple (l, m, ) uniquely species the relative positions of the two cells w.r.t. one another in the following way (also see the diagram above): To get from one cell to another, we go m steps along the oset axis (the one that runs through the vertices of the cells) for a distance of 2 3dm, and l steps along the principal axes (the one that bisects the sides of the cell) for a distance of 2ld. The angle between the two axes is given by . In the diagrams below, we illustrate optimal reuse patterns for = 1 , 1 , 1 , 1 . For 3479 these reuse ratios we can see that identical distances separate the cells using the same frequency band, for all frequency bands. This is not true in general and only holds for 1 specic reuse ratios (we only show up to = 1 , but there are other ones like = 16 , 9 etc. For instance: 1. For = 1 , there are 6 nearest cells at distance 2d 3 interfering. Hence, the 3 received SINR at the base-station due to the user at the edge of the cell (at distance d, i.e., on the side of the cell not on the vertex) is given by SINR = hence, we can write f =
P 2 N0 W 3
+
P 6 (2d3)
=
+
, 6 SNR (2 3)
SNR
6 , (2 3)
Tse and Viswanath: Fundamentals of Wireless Communication
78
3 1 2 3 1 2 3 1 2 3 1 1 2 3 2 3 1 1 2
1 2. For = 4 , there are 6 cells at distance 2d 4 and so f =
6 , (2 4)
1 1 2 4 2 3 1 4 1 2 3 1 4 1 2 3 2 4 1
1 3. For = 7 , there are 6 cells at distance 2d 7 and so f =
6 , (2 7)
4 4 1 5 6 3 4 1 5 6 4 1 5 6 7 7 3 2 5 2 5 6 4 4 1 7 7 3 2 5 6 1
2
7 3 4 1 5 6 3 2 1 7 2
6
Tse and Viswanath: Fundamentals of Wireless Communication 1 4. For = 9 , there are 6 cells at distance 2d 9 and so f =
6 . (2 9)
79
1 5 9 1 5 7 6 3 9 1 5 8 9 4 1 8 2 6 3 2 4 5 7 4 8 9 1 8 2 6 3 2 7 4 8
6 3 9 1 5 7 6 3 9 1 5 7 6 8 2 4 8 2
Thus we see that, for these reuse ratios, the approximation f =
(2
6 q
1 )
is a good one.
In the plot below, we show the high-SNR approximation for the rate: W log2 (1 + 1 ), fp
for = 2, 4, 6 in the hexagonal packing of the plane. Note that the universal reuse ratio = 1 yields the largest rate. Exercise 5.3. Exercise 5.4. Exercise 5.5. In the gure below, we see that the optimal reuse ratio is = 1/2 when studying the high-SNR approximation for the per-user rate in a linear network: W log2 (1 + 1 ), fp
where fp = 2( ) . The plots were generated using = 2, 4, 6 as sample values. 2 Exercise 5.6. 1. This strategy achieves a rate: R = log 1 + P1 N0 + (1 ) log 1 + P2 N0 (5.1)
Tse and Viswanath: Fundamentals of Wireless Communication
HighSNR rate as a function of 1/rho for alpha=2,4,5 4
80
3.5
3
Peruser rate bps/Hz
2.5
2
1.5
1
0.5
0
1
2
3
4
5 6 7 Inverse of reuse ratio: 1/rho
8
9
10
HighSNR rate as a function of 1/rho for alpha=2,4,6 5
4.5
4
3.5 Peruser rate bps/Hz
3
2.5
2
1.5
1
0.5
1
2
3
4
5 6 7 Inverse of reuse ratio: 1/rho
8
9
10
where P1 + (1 )P2 = P . By Jensens inequality and the concavity of log() we have: P1 P2 + (1 ) log 1 + N0 N0 P1 + (1 )P2 P log 1 + = log 1 + N0 N0 log 1 +
= CAW GN
If P1 = P2 we have a strict inequality and it follows that this strategy is suboptimal. 2. As in a) assume we use a strategy of transmitting with power constraint P1 a fraction of the time, and with power constraint P2 the remaining time. Also let
Tse and Viswanath: Fundamentals of Wireless Communication
81
P = P1 + (1 )P2 be the total power constraint. Since this is just a particular strategy that satises the power constraint, the achievable rate cannot exceed the capacity of the channel, which is the supremum of all achievable rates for strategies that satisfy the power constraint. Thus we can write: R = C (P1 ) + (1 )C (P2 ) C (P ) = C (P1 + (1 )P2 ) (5.2)
valid for any P1 and P2 and [0, 1]. Therefore C (P ) is a concave function of P. Exercise 5.7. For QAM with 2k points, the average error probability can be expressed as (for some constant ): Pe = Q 2a2 N0 , (5.3)
where the distance between two consecutive points on each of the axes is 2a and N0 is the background noise. This expression is obtained by applying the union bound to the pairwise errors between the nearest neighbors to some central point. However, from Exercise 3.4, we know that the average SNR per symbol is given by SNR = Eav , where N0 Eav = Hence, we have that Pe = Q 3 SNR 2k 1 exp 3 SNR 2(2k 1) , (5.5) 2a2 k (2 1). 3 (5.4)
where the last inequality holds by the high-SNR approximation of Q(SNR) exp(SNR2 /2). Thus we have that the number of bits k log2 3 SNR 2 ln(/Pe ) = log2 SNR + constant, (5.6)
and the rate of QAM has the optimal order of growth with SNR on the AWGN channel. Exercise 5.8. Exercise 5.9. Exercise 5.10. First, we compute the high-SNR approximation of the mean of log(1+ |h|2 SNR): := E[log(1 + |h|2 SNR)] =
0
log(1 + xSNR)f (x)dx,
Tse and Viswanath: Fundamentals of Wireless Communication
1 SNR
82
=
0
log(1 + xSNR)f (x)dx +
1 SNR
log(1 + xSNR)f (x)dx,
P |h |2
1 SNR
+ log SNR
1 SNR
f (x)dx +
1 SNR
f (x) log xdx.
Where in the last line we have used the high-SNR approximation log(1 + SNR) 1 1 log SNR. Observe that, in the high-SNR regime, P |h|2 SNR SN R and the last two terms are approximately log SNR and E[log |h|2 ], respectively. So we have that 1 + log SNR + E[log |h|2 ] log SNR + E[log |h|2 ]. SNR
Similarly we can dene 2 := E[log2 (1 + |h|2 SNR)] and use the same method to obtain the following high-SNR approximation: 2 log2 SNR + log SNRE[log |h|2 ] + E[log2 |h|2 ]. Finally, the standard deviation is dened as STD := approximation is computed using the above expressions: STD E[log2 |h|2 ] E[log |h|2 ]2 , 2 2 and its high-SNR
which is a constant as a function of SNR. Hence STD goes to zero as SNR increases. On the other hand, in the low-SNR regime we use the approximation log(1+ SNR) SNR log2 e to get SNRE[|h|2 ] log2 e, 2 SNR2 E[|h|4 ] log2 e, 2 Hence, at low-SNR, we have that, STD E[|h|4 ] E[|h|2 ]2 = constant. E[|h|2 ]
This makes sense because at high-SNR, the capacity formula is degree-of-freedom limited and changes in |h|2 have a diminishing marginal eect, whereas in the low-SNR regime, the capacity formula is very sensitive to changes in the overall received SNR and hence even the smallest changes in |h|2 aect the performance. Exercise 5.11. The received SNR is given by hNx . Hence we need to maximize this 0 quantity over all x CL such that x 2 P for xed N0 and some constant P > 0. By the Cauchy-Schwarz inequality, we have that h x
2
2
h
2
x 2,
Tse and Viswanath: Fundamentals of Wireless Communication
83
with equality if and only if x = h for some constant scalar . Hence, the optimal choice is x= Ph , h2
which is exactly the transmit beam-forming strategy. Exercise 5.12. Exercise 5.13. 1. Let 1 = [1, 1, , 1]T RL . Then the channel equation is: y = 1x + z where z CN (0, N0 IL ) and x must satisfy the power constraint E [x] P . We note that we can project the received signal onto the direction of 1 obtaining the sucient statistic: 1 r = y = Lx + z (5.8) L where z CN (0, N0 ). Dening x = Lx we see that we have an AWGN channel with power constraint LP and noise variance N0 . Therefore C = log 1 + LP . N0 We see that there is a power gain of L with respect to the single receive antenna system. 2. Let h = [h1 , h2 , , hL ]T C L . Then the channel equation is: y = hx + z (5.9) (5.7)
where z CN (0, N0 IL ) , h is known at the receiver and x must satisfy the power constraint E [x] P . Since the receiver knows the channel, it can project the received signal onto the direction of h obtaining the sucient statistic: r= h y = h x+z h (5.10)
where z CN (0, N0 ). Then the problem reduces to computing the capacity of a scalar fading channel, with fading coecient given by h . It follows that: h 2P C = E log 1 + N0 LP h 2 = E log 1 + N0 L (5.11)
2
In contrast, the single receive antenna system has a capacity C = E log 1 + |h| 0P N The capacity is increased by having multiple receive antennas for two reasons:
.
Tse and Viswanath: Fundamentals of Wireless Communication rst there is a power gain L, and second V ar [|h| ] = 1. As L ,
h2 L 2 h2 L
84
has the same mean but less varih2 L
ance than |h|2 , and we get a diversity gain. Note that V ar a.s. 1, so it follows that C log 1 +
LP N0
= 1/L whereas
for large L.
3. With full CSI, the transmitter knows the channel, and for a given realization of the fading process {h[n]}N=1 the channel supports a rate: n 1 R= N
N
log 1 +
n=1
h[n] 2 P [n] N0
(5.12)
and the problem becomes that of nding the optimal power allocation strategy. We note that the problem is the same as the one corresponding to the case of a single receive antenna, replacing |h[n]|2 by h[n] 2 . It follows that the optimal solution is also obtained by waterlling: P ( h 2) = 1 N0 h2
+
(5.13)
where is chosen so that the power constraint is satised, i.e. E [P ( h 2 )] = P . The resulting capacity is: h 2P C = E log 1 + N0 (5.14)
At low SNR, when the system is power limited, the benet of having CSI at the transmitter comes from the fact that we can transmit only when the channel is good, saving power (which is the limiting resource) when the channel is bad. The larger the uctuation in the channel gain, the larger the benet. If the channel gain is constant, then the waterlling strategy reduces to transmitting with constant power, and there is no benet in having CSI at the transmitter. When there are multiple receive antennas, there is diversity and h 2 /L does not uctuate much. In the limit as L we have seen that this random variable converges to a constant with probability one. Then, as L increases, the benet of having CSI at the transmitter is reduced. 4. Pout = P r log 1 + h 2P N0 < R = Pr h
2
< (2R 1)
N0 P
(5.15)
Tse and Viswanath: Fundamentals of Wireless Communication We know that we can approximate the pdf of h f (x)
2
85
around 0 by: (5.16)
2
1 xL1 (L 1)!
where Rayleigh fading was assumed, and hence the distribution function of h evaluated at x is approximately given by: F (x) 1L x L!
(5.17)
for x small. Thus, for large SNR we get the following approximation for the outage probability: L 1 N0 (2R 1) Pout (5.18) L! P We see that having multiple antennas reduces the outage probability by a factor of (2R 1)L /L! and also increases the exponent of SNR1 by a factor of L. Exercise 5.14. 1. The Alamouti scheme transmits two independent symbols u1 , u2 over the two antennas in two channel uses as follows: X= u1 u 2 . u2 u 1
To show that the scheme radiates energy in an isotropic manner, we need to show that the energy in the projection of this codeword in any direction d C2 depends only on the magnitude of d and not its direction. Let E[u1 u ] = 0 and 2 E[|u1 |2 ] = E[|u2 |2 ] = P/2. We then have: d E[XX ]d = [d d ] 12 P0 0P d1 = P d 2. d2
2. Suppose that the transmitted vector x = [x1 x2 ]T is such that E[x1 x ] = 0 and 2 E[|x1 |2 ] = E[|x2 |2 ] = P . Then, for any d = [d1 d2 ]T , d E[xx ]d = d E x1 x2 x x d = d P Id = P d 2 , 2 1
hence the scheme radiates energy isotropically. To prove the converse, assume that the scheme x = [x1 x2 ]T is isotropic, i.e., for any two vectors da and db such that da 2 = db 2 = 1, we have that d E[xx ]da = d E[xx ]db . a b (5.19)
Tse and Viswanath: Fundamentals of Wireless Communication
86
Then we must prove that E[x1 x ] = 0 and E[|x1 |2 ] = E[|x2 |2 ]. To see that this 2 must be so, rst choose da = [1 0]T and db = [0 1]T . Substituting this into (5.19) we obtain that E[|x1 |2 ] = E[|x2 |2 ]. Now, choose da =
1 2 1 2
, and db =
1 2 1 2
. Then, (5.19) yields
E[|x1 |2 + x x2 + x1 x + |x2 |2 ] = E[|x1 |2 x x2 x1 x + |x2 |2 ], 2 1 2 1 Hence we get that E[x x2 + x1 x ] = 0 which implies that Real(E[x x2 ]) = 0. 1 2 1 Now, choose da = E |x 1 |2 +
1 2 1 j 2
, and db =
1 2 1 j2
. Then, (5.19) yields
x x2 x1 x x x2 x1 x 1 2 2 + |x 2 |2 = E |x 1 |2 1 + + |x 2 |2 . j j j j
x x x x
Hence we get that E 1j 2 1j 2 = 0, which implies that Imag(E[x x2 ]) = 0. 1 Thus we conclude that E[x1 x2 ] = 0 and we have established the converse. Exercise 5.15. 1. A MISO channel is given by the following input-output relation: y [m] = h x[m] + z [m], with the total power constraint E[ x 2 ] P . The received SNR is given by SNR = E[|h x|2 ] E[h xx h] h Kx h = = . N0 N0 N0
Thus this channel is equivalent to a scalar channel with the same received SNR. Hence, the maximal rate of reliable communication on this channel is given by C = log(1 + SNR) = log 1 + h Kx h N0 .
2. Since the covariance matrix Kx is positive semi-denite, it admits the decomposition Kx = UU , where is a diagonal matrix and U is a unitary matrix. Since the channel is i.i.d. Rayleigh, the vector h is isotropically distributed, i.e., h U has the same distribution as h . Thus the quadratic form h Kx h = h UU h = (h U)(h U) has the same distribution as h h. Therefore, we can restrict Kx to be diagonal without sacricing outage performance.
Tse and Viswanath: Fundamentals of Wireless Communication Exercise 5.16. Exercise 5.17.
87
Exercise 5.18. The outage probability of a parallel channel with L i.i.d. Rayleigh branches is given by the following expression:
L parallel Pout
:= P
l=1
log(1 + |hl |2 SNR) < LR .
Observe that the following inclusion holds:
L L
log(1 + |hl |2 SNR) < LR
l=1
{log(1 + |hl |2 SNR) < R}.
l=1
Hence, since hl s are i.i.d.,
parallel Pout P log(1 + |hl |2 SNR) < R L
.
Since the branches are Rayleigh distributed, we have that, at high-SNR, P log(1 + |hl |2 SNR < R) = P |hl |2 < 2R 1 SNR 2R 1 . SNR
Hence, the outage of the parallel channel at high-SNR satises
parallel Pout
2R 1 SNR
L
.
A simple upper bound that exhibits identical scaling with SNR is obtained by observing that
L L
log(1 + |hl | SNR) < LR
l=1
2
{log(1 + |hl |2 SNR) < LR},
l=1
hence yielding
parallel Pout
2LR 1 SNR
L
.
The SNR scaling of the lower and upper bounds is identical, though the pre-constants are slightly dierent. However, a more precise analysis can be done (see Section 9.1.3,
Tse and Viswanath: Fundamentals of Wireless Communication
88
equation (9.19) and Exercise 9.1)) which shows that the lower bound is actually tight, i.e., the outage probability scales as
parallel Pout
2R 1 SNR
L
.
when the rate is given by R = r log SNR for 0 r 1. We now give a heuristic argument as to why this is true. For the complete proof, see [156], Theorem 4. Let R = r log SNR and let |hl |2 = SNRl , for l R and l = 1, . . . , L (observe that we can always do this since |hl |2 is a non-negative random variable). The |hl |2 are independent and exponentially distributed with mean 1, i.e., the joint density ph is given by ph (|h1 |2 , . . . , |hL |2 ) = e
PL
l=1
|hl |2
.
Applying the change of variable to the above density, we obtain the joint density of l s, p : p (1 , . . . , L ) = (log SNR)L e
PL
l=1
SNRl
SNR
PL
l=1
l
.
Now, we can express the outage probability as
L parrallel Pout
=P
l=1
(1 + SNR|hl |2 ) < 2LR
PL
+ l=1 (1l )
,
P SNR
L
< 2LR ,
=P
l=1
(1 l )+ < Lr ,
+
where, weve used the high-SNR approximation (1 + SNR|hl |2 ) SNR(1l ) (the function (x)+ , denotes max{0, x}). Hence, the outage probability is given by the following integral:
parrallel Pout =
(log SNR)L e
A
PL
l=1
SNRl
SNR
PL
l=1
l
d1 . . . L ,
where A = 1 , . . . , L R : L (1 l )+ < Lr . Since we are considering the highl=1 SNR regime, the term (log SNR)n has no eect on the SNR exponent. Furthermore, the PL term e l=1 SNR l decays exponentially with SNR for l < 0, so we can concentrate only on l > 0. Moreover, the exponential terms approach 1 for l > 0 and e for
Tse and Viswanath: Fundamentals of Wireless Communication
89
l = 0. Hence, the exponential terms have no eect on the SNR exponent. Thus, we can approximate the outage probability as
parrallel Pout
SNR
A+
PL
l=1
l
d1 . . . L ,
where A+ = 1 , . . . , L > 0 : L (1 l )+ < Lr . By Laplaces principle of large l=1 deviations, we have that the integral above is dominated by the term with the largest SNR exponent. Thus,
parrallel Pout SNR inf A+
PL
l=1
l
.
It can be veried that inf A+
L l=1
l = (1 r)L. Hence, we have that 2R SNR
L
parallel Pout
.
when the rate is given by R = r log SNR for 0 r 1. Exercise 5.19. 1. If we transmit the same signal x[m] on each of the parallel channels, the received signal can be written as: y[m] = hx[m] + z[m]. The optimal receiver performs maximal ratio combining and hence this channel becomes equivalent to a scalar AWGN channel with received signal-to-noise ratio given by h 2 SNR, where SNR is the per-channel signal-to-noise ratio on the original parallel channel. Now, suppose that the rate requirement is R bits/sec/Hz per channel. Then, this scheme has outage probability given by
repetition Pout := P(log(1 + h 2 SNR) < LR) = P
h
2
<
2LR 1 SNR
L
,
1 L!
2LR 1 SNR
,
where the last line comes from the high-SNR approximation of the distribution of h (chi-square with 2L degrees of freedom). 2. Using the result of Exercise 5.18, we see that in order to guarantee the same outage probability, we require a larger SNR in the repetition scheme then the minimal required SNR dictated by the outage performance of the channel. In particular, let SNRparallel and SNRrepetition be the minimum required SNR and the
Tse and Viswanath: Fundamentals of Wireless Communication
90
SNR required under repetition coding, respectively. Then, in order to have the same outage probabilities, we need that (at high-SNR), 2R SNRparallel
L
1 = L!
2LR SNRrepetition
L
,
and consequently we obtain that SNRrepetition 2R(L1) = . SNRparallel (L!)1/L For instance, with R = 1 bps/Hz, and L = 5, the repetition scheme requires roughly 18 dB more power over the minimal power requirement to achieve the required outage performance. 3. For small x > 0, log(1+ x) x log2 e. We use this for the low-SNR approximation for both the outage probability of the repetition scheme as well as that of the parallel channel itself. Hence we have that
L parallel Pout := P l=1 repetition Pout := P(log(1 + h 2 SNR) < LR) P
log(1 + |hl |2 SNR) < LR
P h
2
h <
2
<
LR SNR log2 e ,
,
LR SNR log2 e
hence the outage performance of the repetition scheme is approximately optimal in the low-SNR regime. To conclude, in the high-SNR regime, the AWGN parallel channel is degree-offreedom limited and the repetition scheme performs poorly in this regime since it is wasteful of the available degrees of freedom (independent channels) by virtue of sending the same information on all of them at any given time-slot. However, at low SNR, the channel is SNR limited and this shortcoming of the repetition scheme is not evident since the scheme does reap the receive beamforming (coherent combining) benet and hence match the power gain achievable by any other scheme. Exercise 5.20. 1. The low-SNR -outage probability approximation of the parallel channel is given by (see Exercise 5.19, part 3): P h
2
<
LC SNR log2 e
=,
where C denotes the per-channel -outage capacity, i.e., the largest rate achievable while maintaining outage probability below . Let F (x) = P( h 2 > x) be the complementary CDF of h. Then we have that 1 C = F 1 (1 )SNR log2 e. L
Tse and Viswanath: Fundamentals of Wireless Communication 2. For Rayleigh i.i.d. fading branches, F 1 (1 ) (L!) L C=
1 1 (L!) L L 1 L 1 1 L
91 , and so
SNR log2 e,
is the per-channel outage capacity. 3. The delay-spread of the channel is 1s. Hence, from equation (2.47), page 33, 1 we know that the coherence bandwidth is 2106 = 0.5 MHz. But the available bandwidth is 1.25 MHz. Hence, if we exploit the frequency coherence, we can have two independent, parallel channels in frequency. Also, since our time constraint is 100ms and the coherence time is 50ms, we have two parallel channels in time. This makes a total of four parallel channels that we can exploit. Consequently, we let L = 4 in our calculations. Since the SINR per chip is 17 dB and the processing gain is G = W/R = 1.25MHz 100ms = 125000, the SINR per bit per user is roughly 34 dB. Plugging in these values into the formula given in part (2) of this question, we get that C0.01 is roughly 631 bps/Hz/user. On the other hand, the capacity of the unfaded AWGN channel with the same SNR is roughly 3607 bps/Hz. Thus the 1%-outage capacity of this parallel channel is roughly 17.5% of the unfaded AWGN channel with the same received SNR. Solution 5.21: 1. Using only one antenna at a time, we convert the MISO channel into a parallel channel. The maximal rate achievable with this strategy is given by: C parallel = 1 L
L
log(1 + |hl |2 SNR),
l=1
compared by the capacity of this MISO channel (observe that the channel gain is constant and known to both the transmitter and receiver): C MISO = log(1 + h 2 SNR). At high-SNR, we can approximate the two rates as follows: C
parallel
1 log SNR + L
MISO
L
log |hl |2 ,
l=1
C
log SNR + log h 2 .
Hence, at high-SNR, the ratio of the two rates goes to 1.
Tse and Viswanath: Fundamentals of Wireless Communication 2. At low-SNR, we can make the following approximations: C
parallel
92
1 L
L
|hl |2 SNR log2 e =
l=1
1 h 2 SNR log2 e, L
C MISO
h 2 SNR log2 e.
1 L
Thus, the loss from capacity goes to
as SNR 0.
The parallelization scheme is degree-of-freedom ecient so at high-SNR its performance is close to the optimal performance on the MISO channel due to the fact that the AWGN MISO channel is degree-of-freedom limited at high-SNR. However, the optimal strategy is for the transmitter to do beamforming (having knowledge of the channel) and hence harness the power gain aorded in this way. The parallelization scheme does not perform beamforming and hence suers a loss from capacity in the SNR-limited low-SNR regime. 3. The outage probability expressions of the MISO channel and the scheme which turns it into a parallel channel are given by:
L parallel Pout MISO Pout
:= P
l=1
log(1 + |hl |2 SNR) < LR ,
:= P log(1 + h 2 SNR) < R .
Assuming i.i.d. Rayleigh fading, we can use the result of Exercise 5.18 to obtain the high-SNR approximations:
parallel Pout MISO Pout
2R 1 SNR
L
,
L
1 L!
2R 1 SNR
,
hence, the outage probability of the scheme which converts the MISO channel to a parallel channel is L! times larger than the actual outage probability of the MISO channel at high-SNR. At low-SNR, we have C MISO F 1 (1 )SNR log2 e, 1 1 C parallel F (1 )SNR log2 e, L hence, the outage capacity of the scheme is L times smaller than the outage capacity of the MISO channel at low-SNR.
Tse and Viswanath: Fundamentals of Wireless Communication Exercise 5.21. Exercise 5.22. Exercise 5.23. by:
93
1. For the AWGN channel the maximum achievable rate is given R = W log 1 + P N0 W = W log 1 + Eb R N0 W (5.20)
where we used P /R = Eb . Then, the minimum required Eb /N0 for reliable communication is: Eb N0 For the IS-95 system we get
Eb N0
=
req
W R/W 2 1 R
(5.21)
req
= 0.695 = 1.58dB.
At low SNR R/W is small, and we can approximate 2R/W = exp[(R/W ) ln 2] 1 + (R/W ) ln 2, to get Eb N0 ln 2 = 1.59dB
req
(5.22)
and we see that as the SNR goes to zero, the minimum Eb /N0 requirement is -1.59dB. 2. Since we are forced to repeat each transmitted symbol 4 times, we consider the received signal in a block of length 4: y = 1x + z (5.23)
and use 3.a) to conclude that I (x; y) log(1 + 4P/N0 ) where the upper bound can be achieved by choosing the input distribution to be CN (0, P ) i.i.d.. Then the maximum achievable rate (in bits/s/Hz) of this strategy is: Rmax = 4P 1 log 1 + 4 N0 (5.24)
which is strictly smaller than log(1 + P/N0 ), the capacity of the AWGN channel. The loss is due to the concavity of the log() function. For small x, log(x) is approximately linear and the loss due to concavity is small for low SNR. On the other hand, repetition coding has a large loss for high SNR. 3. Loss is greater at high SNR where the loss of d.o.f. is felt more.
Tse and Viswanath: Fundamentals of Wireless Communication
94
4. For repetition coding the minimum Eb /N0 required for reliable communication is given by: W 24R/W 1 Eb = (5.25) N0 req R 4 The increase in Eb /N0 requirement is:
Eb N0 Eb N0 req (rep) req (AW GN )
=
24R/W 1 4(2R/W 1)
(5.26)
For the IS-95 system this loss is only 0.035dB. Exercise 5.24. 1. The channel model is y [m] = h[m]x[m] + w[m], and the rate it can support when channel state is h[m] is R = log 1 + |h[m]|2 P (h[m]) N0 .
Using channel inversion to keep a constant rate R, we need P (h[m]) = Thus the average power needed is E[P ] = (2R 1)N0 E 1 |h[m]|2 1 x = (2R 1)N0 e dx x 0 M 1 x > (2R 1)N0 e dx x 0 M 1 dx = > (2R 1)N0 eM x 0 (2R 1)N0 . |h[m]|2
2. The Channel model is yl [m] = hl [m]x[m] + wl [m], l = 1, . . . , L
Tse and Viswanath: Fundamentals of Wireless Communication
95
and the rate it can support when channel state is h[m] = (h1 [m], . . . , hL [m]) is R= 1 |h[m]|2 P (h[m]) log 1 + 2 N0 .
Using channel inversion to keep a constant rate R, we need P (h[m]) = (2R 1)N0 . |h[m]|2
Since |h[m]|2 is a 2 distribution with pdf f (x) = the average power needed is E[P ] = (2R 1)N0 E = (2R 1)N0 (2R 1)N0 = . L1 3. Assume the noise w CN (0, 1), for dierent target rate and L, the average power is plotted in the following gure. We can see that the power needed is decreasing with increasing number of receiver antennas (actually inversely proportional to L 1). Exercise 5.25. 1. Using optimal scheme, the capacity is C = W log(1 + SINR), where W = 1.25MHz. Hence the SINR threshold for using capacity achieving codes is C SINR = 2 W 1 The following table compares the SINR threshold of using capacity achieving codes to that of IS-856. The dierences are always larger than 3dB. So the codes in IS-856 are not close to optimal. 1 |h[m]|2 1 xL1 x e dx x (L 1)! 0 xL1 x e, (L 1)!
Tse and Viswanath: Fundamentals of Wireless Communication
7 R=1b/s/Hz R=2b/s/Hz R=3b/s/Hz 6
96
5
Average Power
4
3
2
1
0
2
4
6
8 10 Number of Receiver Antennas
12
14
16
Requared rate(kb/s) 38.4 76.8 153.6 307.2 614.4 921.6 1228.8 1843.2 2457.6
Optimal SINR threshold(dB) -16.7 -13.6 -10.5 -7.3 -3.9 -1.8 -0.1 2.5 4.6 W log(1 + L SINR). L
LC
SINR threshold using IS-856(dB) -11.5 -9.2 -6.5 -3.5 -0.5 2.2 3.9 8.0 10.3
2. When repeated L times, the capacity is C= So the threshold SINR is
2W 1 . SINR = L When C = 38.4kb/s, W = 1.25M Hz , and L = 16, the SINR threshold is -16.0 dB. Compared to -16.7 dB computed in part (1), there is not much performance loss from the repetition. Exercise 5.26. 1. Given h[m + 1] =
1 h[m] +
w[m + 1]
Tse and Viswanath: Fundamentals of Wireless Communication the auto-correlation function of the channel process is R[n] = E [h [m]h[m + n]] = E h [m]( 1 h[m + n 1] + w[m + n 1]) = 1 E [h [m]h[m + n 1]] = 1 R[n 1]. Thus R[n] = ( 1 )n R[0] = ( 1 )n .
97
2. If coherence time is Tc and sampling rate is W = 2 1.25M Hz , then ( 1 )W Tc = 0.05 leads to = 1 (0.05) W Tc So for Tc = 25ms (walking), = 0.0000958; for Tc = 2.5ms (driving), = 0.000958. 3. Since h[0] and h[n] are jointly Gaussian, the optimal estimator is MMSE. h[n] = E[h[n]|h[0]] E[h [0]h[n]] = h[0] E[|h[0]|2 ] = ( 1 )n h[0]. 4. From the property of MMSE for jointly Gaussian random variables, we can write h[n] = h[n] + he [n], where the estimation error he [n] is independent of h[n], with variance
2 e = E[|he [n]|2 ]
2
= E[|h[n]|2 ] = 1 (1 )n
E[h [0]h[n]]E[h[0]h [n]] E[|h[0]|2 ]
For IS-856 with 2-slot delay in the feed back,
2 e = 1 (1 )n = 1 (0.05) W Tc , 2 where n 4000. For Tc = 25ms (walking), e = 0.318; for Tc = 2.5ms (driving), 2 e = 0.978.
2n
Tse and Viswanath: Fundamentals of Wireless Communication Exercise 5.27. Exercise 5.28. Exercise 5.29. Exercise 5.30.
98
Chapter 6 Solutions to Exercises
Exercise 6.1. Channel model is y [m] = x1 [m] + x2 [m] + w[m]. The signal power at receiver is P = E[(x1 [m] + x2 [m])2 ] = P1 + P2 + 2E[x1 [m]x2 [m]]. When the two users can cooperate, they can choose the correlation of x1 [m] and x2 [m] to be one and thus get the largest total power P = P1 + P2 + 2 P1 P2 ,
and hence the maximum sum rate they can achieve is P1 + P2 + 2 P1 P2 Ccoop = log 1 + N0 In the case of P1 = P2 = P , Ccoop = log 1 + 4P N0 1 + 2P N0 ,
.
whereas the sum rate without cooperation is Cnoc oop = log At high SNR, Ccoop Cnoc oop .
log(4P/N0 ) 1 as P . log(2P/N0 )
At low SNR,
Ccoop 4P/N0 = 2 as P 0. Cnoc oop 2P/N0 Thus in lower SNR region the cooperative gain is more eective. 99
Tse and Viswanath: Fundamentals of Wireless Communication
100
Exercise 6.2. For orthogonal multiple access channel the rates of the two users satisfy R1 < log 1 + R2 < (1 ) log 1 + P1 N0
P2 (1 )N0
When the degrees of freedom are split proportional to the powers of the users, we have = Thus the sum rate satisfy R1 +R2 < P1 log 1 + P1 + P2 P1 P1 N P1 +P2 0 + P2 log 1 + P1 + P2 P2 P2 N P1 +P2 0 = log 1 + P1 + P2 N0 , P1 . P1 + P2
which is the optimal sum rate. For arbitrary split of degrees of freedom, from the strictly concavity property of log (1 + x), we have R1 + R2 < log 1 + P1 P2 + (1 ) log 1 + N0 (1 )N0 P1 P2 log 1 + + (1 ) N0 (1 )N0 P1 + P2 = log 1 + , N0 P1 P2 = , (1 )
and equality holds only when
that is when the degrees of freedom are split proportional to the powers of the users. Any other split of degrees of freedom are strictly sub-optimal. Exercise 6.3. The symmetric capacity is Csym = 1 P1 + P2 log 1 + 2 N0 .
There are three scenarios of capacity region shown in the following Figure. In scenario I and II, the point A is superior to the symmetric rate point, in scenario III we do not have a superior point. Exercise 6.4.
Tse and Viswanath: Fundamentals of Wireless Communication
R2 A Csym A R2
101
Csym
Scenerio I
R1
Scenerio II
R1
R2
Csym
Scenerio III
R1
Exercise 6.5. Exercise 6.6. Exercise 6.7. Exercise 6.8. Exercise 6.9. Exercise 6.10. Exercise 6.11. Exercise 6.12. Let ei means the event of decoding incorrectly at stage i, and ec means i the event of decoding correctly at stage i, then the probability of error for the k th user under SIC satises pe = P e1
k
(e2 |ec ) 1
(e3 |ec , ec ) 12
(ek |ec . . . ec 1 ) 1 k
P(e1 ) + P(e2 |ec ) + + P(ek |ec . . . ec 1 ) k 1 1 =
i=1
p(i) , e
Tse and Viswanath: Fundamentals of Wireless Communication
(i)
102
where pe = P(ei |ec . . . ec1 ) is the probability of decoding the ith user incorrectly 1 i assuming that all the previously users are decoded correctly. Exercise 6.13. In the following we denote SNR1 =
P1 Tc N0
2 and SNR2 = |h2 |2 PNTc . 0
1. By neglecting user 2 and using training signal xt [m], we have y [m] = h1 [m]xt [m] + [m]. Let h1 [m] be the estimation of channel state h1 [m], then the MMSE of h1 [m] from y [m] can be calculated using (A.85) in Appendix A, and it is E (h1 [m] h1 [m])2 = E [|h1 |2 ] N0 E [|h1 |2 ] xt 2 + N0 N0 = 0.2P1 Tc + N0 1 = . 0.2SNR1 + 1
2. The channel can be written as y [m] = h1 [m]x1 [m] + h2 [m]x2 [m] + (h1 [m] h1 [m])x1 [m] + [m], where the SIC decoder can subtract h1 [m]x1 [m] from channel estimation and user 1 [m])x1 [m]+ [m] is the noise plus the interference 1s signal. The term (h1 [m] h from inaccurate estimation of the channel. Thus E ((h1 [m] h1 [m])x1 [m] + [m])2 = and the SINR of user 2 is SINR2 = = = |h2 |2 P2 Tc
P1 Tc 0.2SNR1 +1
P 1 Tc + N0 , 0.2SNR1 + 1
+ N0 +1
SNR2
SNR1 0.2SNR1 +1
SNR2 (0.2SNR1 + 1) . 1.2SNR1 + 1
The numerical calculation is shown in the following gure. We can see that the degradation is worse if the power of user 1 increases. This is because user 1s signal is the interference to user 2 due to inaccurate estimation of the channel, and with the increase of the power of user 1, the interference also increases, hence the SINR for user 2 decreases.
Tse and Viswanath: Fundamentals of Wireless Communication
103
3. If user 1 is decoded correctly, then we can estimate the channel state from both the training symbol and user 1s signal. That is, we estimate h1 [m] from y1 and y2 where y 1 = h 1 xt + 1 , y2 = h1 x1 + 2 . Using (A.85) in Appendix A, we have E (h1 h1 )2 = E [|h1 |2 ] N0 E [|h1 |2 ] ( xt 2 + x1 2 ) + N0 N0 = P 1 T c + N0 1 = . SNR1 + 1 SNR2
SNR1 SNR1 +1
Using the above estimation error to redo part(2), we get SINR2 = = +1
SNR2 (SNR1 + 1) . 2SNR1 + 1
1
0.9
0.8 channel estimation from training only improved channel estimation 0.7 SINR2/SNR2
0.6
0.5
0.4
0.3
0.2
0.1
0
1000
2000
3000
4000
5000 P1/N0
6000
7000
8000
9000
10000
From the gure we can see that the SINR for user 2 improves, especially at high P1 /N0 .
Tse and Viswanath: Fundamentals of Wireless Communication Exercise 6.14. 1. At very high SNR, we have log 1 + for k = 1, 2, . . . , K . Thus for any S {1, . . . , K }, we have that |S |R > log 1 + is approximately equivalent to R> 1 P log |S | N0
iS k i=1
104
P |h i |2 N0
log
P N0
P |h i |2 N0
at very high SNR, and hence the dominating event for pul is when |S | = K , that out is, the one on sum rate. 2. At high SNR,
K
pout
P K R > log(1 + SNR
k=1 K
|h k |2 )
=P
k=1
|hk |2 < 2KR 1 SNR
sym
2
KR
1 SNR .
K
1 K! Let 1 K! we get C sym = 3. At very high SNR, C sym = C
K
2KC 1 SNR
=,
1 log 1 + SNR(K ! )1/K . K log 1 + SNR(K ! )1/K log(1 + SNR) 1 log(SNR) + log(K ! )1/K K log(SNR) + log 1 . K
1 K
Tse and Viswanath: Fundamentals of Wireless Communication Exercise 6.15. Exercise 6.16. Exercise 6.17. Exercise 6.18. Exercise 6.19. Exercise 6.20. Exercise 6.21.
105
Exercise 6.22. The probability that no one user sends a request rate is the probability that all users channel is less than , that is
K
P{no one sends a request rate} =
k=1 K
P{user ks channel is less than } P{|hk |2 SNR < }
k=1 K
=
= 1 e/SNR = (1 e )K ,
where we used SNR = 0dB = 1. We need this probability to be , thus (1 e )K = , and the solution for is = ln 1
1/K
.
Now, the probability that any user sends a request is p = e , and the number of users that sends in a requested rate is a Binomial random variable with parameter p, hence E[number of users that sends in a requested rate] = Kp = Ke = K 1
1/K
.
The expected number of users that sends in a requested rate for dierent K and is shown in the following gure. Exercise 6.23.
Tse and Viswanath: Fundamentals of Wireless Communication
4.5
106
4
= 0.1 = 0.01
Expected number of feed back users
3.5
3
2.5
2
1.5
1
0
10
20
30 40 Number of users
50
60
70
Exercise 6.24. Exercise 6.25. Exercise 6.26. Exercise 6.27. Exercise 6.28. Exercise 6.29. Exercise 6.30. Exercise 6.31. 1. Under Alamouti scheme, the eective SNR is u1 = (|h1 |2 + |h2 |2 ) SNR , 2 for the single antenna case, the eective SNR is u2 = |h1 |2 SNR, where h1 and h2 are i.i.d CN (0, 1). Hence the distributions are 2 2u1 SNR SNR 1 u2 f2 (u2 ) = e SNR . SNR f1 (u1 ) = e SNR ,
2u1
Tse and Viswanath: Fundamentals of Wireless Communication
1
107
0.9
0.8 Alamouti scheme Single antenna
0.7
0.6 f(x)
0.5
0.4
0.3
0.2
0.1
0
0
0.5
1
1.5
2
2.5 x
3
3.5
4
4.5
5
Figure 6.1: Probability distribution functions
2. From the plot in part (1), we can see the probability that eective SNR is low is smaller for the Alamouti scheme. So we expect that dual antenna can provide some gain. Rigorous proof is as follows. Let CA be the capacity for Alamouti scheme, and CS be the capacity for the single antenna case, we have CS = E log(1 + |h1 |2 SNR) 1 1 = E log(1 + |h1 |2 SNR) + E log(1 + |h2 |2 SNR) 2 2 2 2 |h1 | + |h2 | ) E log 1 + SNR 2 = CA , where in the rst step we used the fact that |h1 |2 and |h2 |2 are i.i.d, and in the second step we used the Jensens inequality. 3. When there are K users, assume SNR = 1, we have For the Alamouti scheme, the eective SNR is u1 = P 1 max (|h1k |2 + |h2k |2 ) < x 2 k=1,...,K
1 max (|h1k |2 2 1,...,K
+ |h2k |2 ), and
K
= 1 (1 + 2x)e2x
,
Tse and Viswanath: Fundamentals of Wireless Communication
1
108
0.9
0.8
0.7
0.6 F(x)
0.5 Alamouti scheme Single antenna
0.4
0.3
0.2
0.1
0
0
0.5
1
1.5
2
2.5 x
3
3.5
4
4.5
5
Figure 6.2: Cumulative distribution functions
so f1 (u1 ) = 4Ku1 e2u1 1 (1 + 2u1 )e2u1 Thus the capacity under Alamouti scheme is
K 1 K 1
.
CA =
0
4Ku1 e2u1 1 (1 + 2u1 )e2u1
log(1 + u1 )du1 .
For the single antenna case, the eective SNR is u2 = max |hk |2 , and
k=1,...,K K
P so
k=1,...,K
max |hk |2 < x
= 1 ex
,
f2 (u2 ) = Keu2 1 eu2 Thus the capacity for the single antenna case is
K 1
K 1
.
CS =
0
Keu2 1 eu2
log(1 + u2 )du2 .
The achievable throughput under both schemes at SNR=0dB for dierent number of users is shown in Figure 6.3.
Tse and Viswanath: Fundamentals of Wireless Communication
2
109
1.8
1.6 Capacity(b/s)
1.4
1.2
Alamouti scheme Single antenna
1
0.8
1
2
3
4
5 6 Number of users
7
8
9
10
Figure 6.3: Throught vs. number of users
4. We can see from the plot in part (3) that single antenna scheme performs better than the Alamouti scheme when K 2. So in this case we do not need use dual transmit antenna. This is because the probability of getting one good channel is larger than getting two good channels.
Chapter 7 Solutions to Exercises
Exercise 7.1. 1. Refer to Figure 7.3(a) in the textbook, we have di = d2 + ((i 1)r c )2 + 2(i 1)dr c cos =d 1+ ((i 1)r c )2 2(i 1)r c cos + d2 d
1/2 1/2
.
For large d (large compare to the size of receiver antenna array), we can expand the above equation to the rst order term and get di d 1+ 1 2(i 1)r c cos 2 d = d + (i 1)r c cos,
which is equation (7.19). 2. From Figure 7.1 we have d2 = [d (k 1)t c cos t + (i 1)r c cos r ]2 ik + [(k 1)t c sin t (i 1)r c sin r ]2 d2 2d(k 1)t c cos t + 2d(i 1)r c cos r for large d. Hence dik 1 1 d 1 (k 1)t c cos t + (i 1)r c cos r d d = d (k 1)t c cos t + (i 1)r c cos r ,
1/2
which is equation (7.27). 110
Tse and Viswanath: Fundamentals of Wireless Communication
111
k
dik
i
(k 1)tc (i 1)r c
t
d
r
Figure 7.1: MIMO case
Exercise 7.2. 1 er () = nr
1 exp(j 2 r ) exp(j 2 2r ) . . . exp(j 2 (nr 1)r )
.
We have er 1 + r 1 = nr 1 exp(j 2 r ) exp(j 2 ) exp(j 2 2r ) exp(j 4 ) . . . exp(j 2 (nr 1)r ) exp(j 2(nr 1) ) Hence er () is periodic with period 1 er ( + ) = nr
1 . r
= er ().
1 Next suppose there exists a 0, r
such that = er (),
1 exp(j 2 r ) exp(j 2 r ) exp(j 2 2r ) exp(j 4 r ) . . . exp(j 2 (nr 1)r ) exp(j 2(nr 1) r )
Tse and Viswanath: Fundamentals of Wireless Communication and we must have exp(j 2 r ) = 1. Clearly there is no exp(j 2 r ) = 1. So the smallest period is Exercise 7.3.
1 . r 1 0, r
112 such that
fr (r ) = fr (r2 r1 ) = er (r1 ) er (r2 ) 1 = [1 exp(j 2 r r1 ) exp(j 2 2r r1 ) . . . exp(j 2 (nr 1)r r1 )] nr 1 exp(j 2 r r2 ) exp(j 2 2r r2 ) . . . exp(j 2 (nr 1)r r2 ) 1 = [1 + exp(j 2 r r ) + exp(j 2 2r r ) + + exp(j 2 (nr 1)r r )] nr 1 1 exp(j 2nr r r ) = nr 1 exp(j 2 r r ) exp(jnr r r ) exp(jnr r r ) exp(jnr r r ) = nr exp(j r r ) exp(j r r ) exp(j r r ) exp(j (nr 1)r r ) sin(nr r r ) = nr sin( r r ) exp(j (nr 1)r r ) sin(Lr r ) = , nr sin(Lr r /nr ) which is equation (7.35). Exercise 7.4. The degree of freedom of MIMO channel depends on the angular spread of the scatters/reectors, and depends on the antenna array length. There are two problems of (7.82). 1. The degree of freedom depends on the antenna array length, which determines the resolution of the channel. Simply increasing number of antenna does not necessarily increase channel resolution. 2. The number of multipath does not have direct impact on the degree of freedom, it is the angular spread that inuences the degree of freedom. Exercise 7.5. Exercise 7.6. Exercise 7.7.
Tse and Viswanath: Fundamentals of Wireless Communication Exercise 7.8. Exercise 7.9. Exercise 7.10. Exercise 7.11.
113
Chapter 8 Solutions to Exercises
Exercise 8.1. Consider the singular value of decomposition: H = UV. Then, the channel model y = Hx + w, can be rewritten as y = x + w , where y = U y and x = Vx. Moreover, x has the same total power constraint as x since V is a unitary matrix. Thus, the channel capacity depends only on the singular values of H for a total power constraint. But, H has the same non-zero singular values as H and hence the capacity of the reciprocal channel is same as the original channel. Exercise 8.2. Exercise 8.3. Exercise 8.4. 1. Let A = HU. Then, A is zero mean. Let ai and hi denote the ith columns of A and H respectively. Then, E [a ai ] = U E [h hi ]U, j j = U ij IU, = ij I. Also, E [at ai ] = Ut E [ht hi ]U, j j = 0. Thus, A has i.i.d. CN (0, 1) entries. 114
Tse and Viswanath: Fundamentals of Wireless Communication 2. Consider the eigenvalue decomposition of Kx : Kx = UDU ,
115
where D is a positive diagonal matrix and U is a unitary matrix. Then, the mutual information can be written as: EH log det I + 1 HKx H N0 = EH log det I + 1 (HU)D(HU) , N0 1 (HU)D(HU) , = EHU log det I + N0 1 = EH log det I + HDH , N0
where the last step follows from the rst part. Moreover, we also have Trace(Kx ) = Trace(D). Thus, the input covariance matrix can be restricted to be a diagonal matrix. 3. Note that log det(X) is a concave function of X for positive denite X (see Page 74 of Convex Optimization by Stephen Boyd and L. Vandenberghe). Also 1 I + HDH is linear in D. Thus, the function log det I + N0 HDH is concave in D. Moreover, the function is symmetric in the diagonal entries of D: this follows from the fact that reordering rows of H does not change the distribution of H. Thus, for a trace constraint on D the mutual information is maximized when D is a multiple of the identity matrix. Exercise 8.5. 1. Multiple receive antennas provide a power gain which is very crucial for a wideband CDMA system which works at low SNR. However, the gain is not very crucial for a narrowband GSM which works at high SNR. The gain can be crucial for a wideband OFDM system if the mobile device is on the boundary of a cell and hence is working at low SNR. 2. Multiple transmit antennas can provide a degree of freedom gain for the high SNR narrowband GSM system. However for a wideband CDMA system, the transmit antennas will not be useful as the optimal coding strategy will be to use a single transmit antenna at low SNR. Similar arguments work for the wideband OFDM system depending which SNR regime it is working in. Exercise 8.6. Exercise 8.7.
Tse and Viswanath: Fundamentals of Wireless Communication
116
Exercise 8.8. 1. The pilot signal with transmit power SNR will estimate the chan1 nel H with an mean squared estimation error of order of SNR . This estimation error in the channel multiplied by the transmitted signal will act as an additive noise during rest of the communication. Thus, while communicating, the eective noise seen by the receiver is the additive Gaussian noise and the noise 1 due to estimation error of H which is of the order SNR SNR = 1. Thus, for the eective coherent channel, the total additive noise has a bounded variance for any SNR. Thus, for k nr channel, assuming that eective noise is Gaussian (worst-case assumption) the channel capacity for the eective coherent channel is lower bounded by min(k, nr ) log SNR. But since k time slots were used for the pilot scheme, the eective rate of communication is given by at least Tc k min(k, nr ) log SNR bits/s/Hz Tc . 2. With nt transmit antennas, we only need at most nt time slots for training. Additional training time can only hurt the overall rate at high SNR. Thus, we have k nt . Now, if nr nt , then we only use nr transmit antennas. As other antennas will not provide a degree of freedom gain. Thus we also have k min(nr , nt ). Now, if k min(nr , nt ), Tc k k is increasing in k for k < Tc /2 Tc and decreasing for k > Tc /2. Thus, the optimal value of k is given by: k = min(nr , nt , Tc /2). Exercise 8.9. Consider the channel from a particular transmit antenna i to a particular receive antenna j . This channel can be modeled as a simple scalar ISI channel with tap coecients Hl (i, j ). For this channel, the usual scalar OFDM scheme will yield Nc tones dene by:
L1
Hn (i, j ) =
l=0
Hl (i, j )e
j 2nl Nc
.
Now, for the original MIMO channel, we can look at each receive antenna separately. For each receive antenna, signals transmitted from dierent transmit antennas add linearly at the receiver. Since the OFDM scheme is a linear operation, the overall eective OFDM channel for one particular receive antenna can be written as the sum of the individual OFDM channels. Now, for the original MIMO channel, since for each receive antenna, the OFDM scheme is the same, we get that the overall OFDM scheme can be written as:
L1
Hn =
l=0
Hl e
j 2nl Nc
.
Tse and Viswanath: Fundamentals of Wireless Communication
117
Exercise 8.10. For a xed physical environment (i.e., xed H), the capacity for a total power constraint P is given by max W log det I + 1 HKx H No W bits/sec.
Trace(Kx )P
This can be rewritten as:
Trace( Kx )1 P
max
W log det I +
P Kx H H No W P
bits/sec.
When, both P and W are both doubled, the corresponding capacity can be written as: 2 max
Trace( Kx )1 2P
W log det I +
P Kx H H No W 2P
bits/sec.
Since the two optimization problems are essentially the same, the optimal solution for the second case is given by K = 2K . Therefore the capacity is exactly doubled. x x Exercise 8.11. Exercise 8.12. Exercise 8.13. Exercise 8.14. Exercise 8.15. Exercise 8.16. The general capacity expression is given by: C = E log det Inr + SNR HH nt .
SNR HH nt
The apparent paradox is because of the behavior of log det Inr + values of SNR it behaves like log Trace behaves like log det
SNR HH nt SNR HH nt
. At low
, whereas at medium SNR it
. Thus, we have the following consistent behavior: C1n nSNR log SNR + log n Cnn nSNR n log SNR
low SNR medium SNR Exercise 8.17. Exercise 8.18.
Tse and Viswanath: Fundamentals of Wireless Communication Exercise 8.19.
118
Exercise 8.20. For matched lters, the interference seen by the k th stream is given by: h hi k xi + h w . k ||hk ||
i=k
Thus, the variance of the interference seen by the k th stream is approximately given by: h hi k P i + N0 . ||hk ||
i=k
Therefore, the rate for the k th stream is given by (assuming the worst-case assumption that the interference is Gaussian): Rk = E log 1 + = E log 1 + E log 1 + E PK ||hk ||2
h hi k i=k ||hk || Pi PK ||hk ||2 N0 h hi Pi k i=k ||hk || N0
+ N0 +1 ,
,
PK ||hk ||2 N0
,
PK ||hk ||2 , N0
where the last two steps follow from the low SNR assumption. Thus, the total sum-rate is given by: Rk =
k
SNR nt
E ||hk ||2 ,
k
= nr SNR, which at low SNR is the capacity of a 1 nr channel (see Soln 8.16). Exercise 8.21. Exercise 8.22. We use the following matrix identity which follows from the matrix inversion lemma: log |A + xx | log |A| = log(1 + x A1 x). (8.1)
Tse and Viswanath: Fundamentals of Wireless Communication Taking x =
P1 h N0 1 nt
119
implies: Pi hi h i N0
Pk h N0 k nt
log det I +
i=1
log(1 + SINR1 ) = log det I +
i=2
Pi hi h i N0
.
Similarly, taking x =
nt
we get: log(1 + SINRk ) = log det I + Pi hi h . i N0 i=k+1
nt
log det I +
i=k
Pi hi h i N0
Adding all such equations, we get:
nt
log det I +
i=1
Pi hi h i N0
nt
=
i=1
log(1 + SINRi ).
Thus, taking Kx to be a diagonal matrix with entries Pi we get: log det I + 1 HKx H i N0
nt
=
i=1
log(1 + SINRi ).
Exercise 8.23. We have the following sequence of steps: pout (R) P {log det (Inr + SNRHH ) < R} , P {SNRTr[HH ] < R} , nr nt (c) R P SNR|h11 |2 < , nr nt
(d) (b) (a)
=
1e
n
R r nt SNR
nr nt
,
(e)
R nt nR . (nr nt SNR)nt nr
Each of these steps can be justied as follows: (a): follows from letting each antenna power be SNR rather than SNR/nt . (b): follows from the equation: SNRTr[HH ] < det (Inr + SNRHH ) and hence a simple set theoretic containment relationship. (c): again follows from a simple set theoretic containment relationship: SNR|hij |2 < R i, j nr nt {SNRTr[HH ] < R i, j } ,
and the facts that hij s are i.i.d.
Tse and Viswanath: Fundamentals of Wireless Communication (d): follows from the fact |h11 |2 is exponential. (e): follows from a simple Taylor series expansion.
120
Exercise 8.24. At high SNR, MMSE-SIC receiver is same as the decorrelator followed by SIC. At high SNR, for the rst stream, a decorrelator projects nr dimensional receive vector along a sub-space orthogonal to nt 1 other directions. Thus, the diversity seen by the rst stream will be nr (nt 1) = nr nt + 1. Decoding of this stream in fact will be bottleneck for all other streams. Thus, for each stream the diversity is given by nr nt + 1. However, for the k th stream, if all the previous streams have been decoded correctly, then the diversity seen by the k th stream is given by nt nt + k (projection of nr dimensional vector onto a sub-space orthogonal to a nt k dimensional sub-space). Exercise 8.25. 1. From MMSE estimation of streams, we have SNR|g1 |2 = h (I/SNR + h2 h )1 h1 . 1 2 Using the matrix inversion lemma we get: |g 1 |2 = h I 1 SNRh2 h 2 h1 , 1 + SNR||h2 ||2 SNR||h h2 ||2 1 = ||h1 ||2 . 1 + SNR||h2 ||2
Now, consider: ||h1||2 ||2 ||h1||2 ||2 2 2 ||h12 || + = ||h1 || ||h1||2 || + , 1 + SNR||h2 ||2 1 + SNR||h2 ||2 SNR||h2 ||2 ||h1||2 ||2 = ||h1 ||2 , 1 + SNR||h2 ||2 SNR||h h2 ||2 1 = ||h1 ||2 , 1 + SNR||h2 ||2
2
which matches with the expression above for |g1 |2 . Thus, we have |g1 |2 = ||h12 ||2 + ||h1||2 ||2 1 + SNR||h2 ||2
The fact that |g2 |2 = ||h2 ||2 follows directly since the rst symbol doesnt see any interference. 2. At high SNR, the second term is small with high probability, thus the marginal distribution of |g1 |2 is same as ||h12 ||2 . Now taking h2 as a basis vector, we see
Tse and Viswanath: Fundamentals of Wireless Communication
121
that h12 is always orthogonal to a basis vector. Thus it is a projection of h1 onto one dimension. Thus, statistically it should be similar to simple complex Gaussian. Note, that by circular symmetry of h1 , the fact that h2 is along a random direction does not change the statistics. Since, ||h12 ||2 is exponential, |g1 |2 is marginally exponential at high SNR. Moreover, we have: |g1 |2 = ||h12 ||2 + ||h1||2 ||2 , 1 + SNR|g2 |2
where h12 and h1||2 are independent of |g2 |. Thus, we see that |g1 | and |g2 | are negatively correlated. 3. The maximum diversity given by the parallel channel is same as the original MIMO channel since the D-BLAST structure preserves mutual information and hence the outage behavior. Thus, the total diversity is given by 4. 4. If |g1 |2 and |g2 |2 were independent with the same marginals, then the diversity oered by |g1 |2 is 1 and that oered by |g2 |2 is 2. Thus, the total diversity is given by 3. Exercise 8.26. The coding scheme can be written as (1) (2) (T n +1) 0 0 p1 p1 p1 t . . . . . .. . .. . .. . .. . . . . .. 0 p(1)1 p(2)1 . . 0 nt nt (2) (1) 0 0 pnt pnt where P (k) = p1 , . . . , pnt
(k ) (k )
,
are the independent data streams. The decoding can be
done using successive interference cancelation: estimate stream P (k) one by one, then jointly decode it and then estimate P (k+1) after canceling out P (k) . Exercise 8.27. For an nt transmit antenna channel a D-BLAST scheme with blocklength T has a rate loss of: nt 1 , T because a zero is sent on the rst transmit antenna for the rst nt 1 time slots. So instead of sending T streams, we send only T (nt 1) streams.
Chapter 9 Solutions to Exercises
Exercise 9.1. Exercise 9.2. Exercise 9.3. Exercise 9.4. Exercise 9.5. Exercise 9.6. Exercise 9.7. Exercise 9.8. Exercise 9.9. Exercise 9.10. Exercise 9.11. Exercise 9.12. Exercise 9.13. Exercise 9.14. Exercise 9.15. Exercise 9.16. Exercise 9.17. Exercise 9.18. 122
Tse and Viswanath: Fundamentals of Wireless Communication Exercise 9.19. Exercise 9.20. Exercise 9.21. Exercise 9.22. Exercise 9.23. Exercise 9.24. Exercise 9.25.
123
Chapter 10 Solutions to Exercises
Exercise 10.1. Exercise 10.2. Exercise 10.3. Exercise 10.4. Exercise 10.5. Exercise 10.6. Exercise 10.7. Exercise 10.8. Exercise 10.9. Exercise 10.10. Exercise 10.11. Exercise 10.12. Exercise 10.13. Exercise 10.14. Exercise 10.15. Exercise 10.16. Exercise 10.17. Exercise 10.18. 124
Tse and Viswanath: Fundamentals of Wireless Communication Exercise 10.19. Exercise 10.20. Exercise 10.21. Exercise 10.22. Exercise 10.23. Exercise 10.24. Exercise 10.25.
125
Appendix A Solutions to Exercises
Exercise A.1. 1. n = 1. Let A = w2 , then using the formula for the density of a function of a random variable, we get: fw ( a) fw ( a) + f1 (a) = 2a 2a 1 = = exp (a/2) 2a 2. Let n ( ) denote the characteristic function of ||w||2 . Then since convolution corresponds to multiplication of characteristic functions, we get n ( ) = 1 ( )n = 2 ( )n/2 = 1 1 2j
n/2
,
where the last step follows from the fact for n = 2, ||w||2 is an exponential random variable. Then, from this we see that dn ( ) 1 =n d(j ) 1 2j = nn+2 ( ).
n/2+1
Since dierentiation corresponds to multiplication in time domain, we get a fn+2 (a) = fn (a). n 3. Using simple recursion: 1 an/21 exp(a/2) for n odd fn (a) = 2 1 3 (n 2) 126
Tse and Viswanath: Fundamentals of Wireless Communication 1 an/21 exp(a/2) for n even 2 2 4 (n 2)
127
fn (a) =
Exercise A.2. (zi )M is a linear transformation of a Gaussian process, so it has a i=1 jointly Gaussian distribution, which is completely specied by the rst and second moments.
E [zi ] = E
w(t)si (t)dt =
E [w(t)]si (t)dt = 0
E [zi zj ] = E = =
N0
w(t)w( )si (t)sj ( )dtd 2 (t )si (t)sj ( )dtd
N0 N0 si ( )sj ( )d = i,j 2 2
Therefore E [zzT ] = (N0 /2)IM and z N(0, (N0 /2)IM ). Exercise A.3. For simplicity, let us assume that x is zero mean. 1. The covariance matrix, K, of x is given by K = E [xx ] = E [(R[x] + j C [x])(R[x]t j C [x]t )] = E [R[x]R[x]t ] + E [C [x]C [x]t ] jE [R[x]C [x]t ] + jE [C [x]R[x]t ]. Similarly, the pseudo-covariance matrix of x is given by J = E [R[x]R[x]t ] E [C [x]C [x]t ] + jE [R[x]C [x]t ] + jE [C [x]R[x]t ]. The covariance of matrix of [R[x], C [x]]t is given by E [R[x]R[x]t ] E [R[x]C [x]t ] E [C [x]R[x]t ] E [C [x]C [x]t ] = 1 R(K + J ) C (J K ) 2 C (K + J ) R(K J ) (A.1)
2. For a circularly symmetric x, J = 0 and the covariance of matrix of [R[x], C [x]]t is given by E [R[x]R[x]t ] E [R[x]C [x]t ] E [C [x]R[x]t ] E [C [x]C [x]t ] = 1 R(K ) C (K ) 2 C (K ) R(K ) (A.2)
Tse and Viswanath: Fundamentals of Wireless Communication
128
Exercise A.4. 1. Necessity of the two conditions is proved in appendix A. For proving suciency, let y = ej x, then E [y] = ej E [x] = 0. Then the pseudo-covariance of y is given by E [yyt ] = e2j E [xxt ] = 0. The covariance of y is given by E [yyx ] = ej E [xx ]ej = E [xx ]. Thus, y and x have the same second order statistic and hence have identical distribution. 2. Since x is not given to be zero mean, the answer is no. But in addition if we assume that x is zero mean, then from (A.1) and (A.2) we see that J must be zero and hence x will be circularly symmetric. Exercise A.5. Let x = xr + jxi . Then xr and xi are zero mean and are jointly Gaussian. Since the pseudocovariance for a circularly symmetric Gaussian is zero, we get 0 = E [x2 ] = E [x2 ] E [x2 ] + 2jE [xr xi ]. r i Thus, E [x2 ] = E [x2 ] and E [xr xi ] = 0. For jointly Gaussian random variables, uncorr i related implies independent. Also, since they are zero mean and have the same second moment, xr and xi are i.i.d. random variables. Exercise A.6. Let x be i.i.d. complex Gaussian with real and imaginary part distributed as N (0, Kx ). Then the covariance and pseudo-covariance of x is given by: K= = J= = E [xx ], (Kx (1, 1) + Kx (2, 2))I, E [xxt ], (Kx (1, 1) Kx (2, 2) + 2j Kx (1, 2))I.
Now, let y = Ux, then the covariance of y is given by E [yy ] = UE [xx ]U
Tse and Viswanath: Fundamentals of Wireless Communication = (Kx (1, 1) + Kx (2, 2))I. The pseudo-covariance of y is given by E [yyt ] = UE [xxt ]Ut = (Kx (1, 1) Kx (2, 2) + 2j Kx (1, 2))UUt .
129
Since for general U, UUt cannot be identity, we get Kx (1, 1) = Kx (2, 2) and Kx (1, 2) = 0. That is, x should be circularly symmetric. Exercise A.7. The ML decision rule is given through the likelihood ratio which is P(y|x = 1) Pz (y h) = , P(y|x = 1) Pz (y + h)
n
=
i=1
Pzi (yi hi ) , Pzi (yi hi )
where zi , yi and hi are two dimensional vectors with entries as the real and complex parts of the ith entry of z, y and h respectively. Now, Pzi (yi hi ) exp ((yi hi )t K1 (yi hi )/2) x = t K1 (y + h )/2) Pzi (yi hi ) exp ((yi + hi ) x i i t 1 t 1 , = exp (hi Kx yi + yi Kx hi )/2 , = exp ht K1 yi . ix Thus, the likelihood ratio can be written as
n
exp
i=1
ht K1 yi . ix
Note that i ht yi = h y. Thus for the likelihood ratio to be a function of only h y, i we need that every hi should be a right-eigenvector of K1 with the same eigenvalue. x Note that this condition is trivially satised if Kx is a scalar multiple of the identity matrix. Exercise A.8. 1. z N(0, 2 In ). Let H0 = {x = 1} and H1 = {x = 1}. Then,
p(y | H0 ) = K exp( y h 2 /(2 2 )) p(y | H1 ) = K exp( y + h 2 /(2 2 )) p(y | H0 ) 1 LLR(y) = log = 2[ y p(y | H1 ) 2
2
+h
2
2yT h y
2
h
2
2yT h]
Tse and Viswanath: Fundamentals of Wireless Communication
H =0
130
=
2T yh < 2
H =1
0
This last expression is the ML detection rule. Therefore yT h is a sucient statistic.
Pe = P (yT h > 0 | x = 1) = P [(h + z)T h > 0] = P (hT z > h 2 ) hT z h =P >h =Q h where we have used the fact that E
hT z zT h h h
= 2.
We see that Pe is minimized by choosing any h such that h = 1. In this case Pe = Q(1/ ). 2. z N(0, Kz ), with non-singular Kz . Let H0 = {x = 1} and H1 = {x = 1}. Then, p(y | H0 ) = K exp (y h)T K1 (y h)/2 z p(y | H1 ) = K exp (y + h)T K1 (y + h)/2 z p(y | H0 ) LLR(y) = log = y T K 1 h + hT K 1 y z z p(y | H1 ) H=0
H =0
= 2 y T K1 h z
<
H =1
0
H=1 This last expression is the ML detection rule. Therefore yT K1 h is a sucient z statistic.
Pe = P (yT K1 h > 0 | x = 1) = P [(h + z)T K1 h > 0] = P (zT K1 h > hT K1 h) z z z z =P zT K1 h z hT K1 h z > hT K1 h z =Q hT K1 h z
Tse and Viswanath: Fundamentals of Wireless Communication
131
We see that Pe is minimized by choosing h to be the norm one eigenvector of K1 z associated with its maximum eigenvalue, which is the norm one eigenvector of 1 . Kz associated with its minimum eigenvalue min . In this case Pe = Q min This choice of h makes the signal of interest hx lie in the direction where the noise is smallest, maximizing the signal to noise ratio in the received signal y. 3. If Kz is singular then the noise vector z lies in a proper subspace of Rn which we call S . Then letting S be the orthogonal subspace associated to S , we can choose h S , project onto this direction, and obtain a noise-free sucient statistic of x. The corresponding probability of error is 0.
Appendix B Solutions to Exercises
Exercise B.1. Exercise B.2. 1. f (x) is concave if for any [0, 1]: f (x1 ) + (1 )f (x2 ) f [x1 + (1 )x2 ] for all x1 , x2 Dom(f ). 2. Jensens inequality: for a random variable X and a concave function f (): E [f (X )] f [E (X )] Proof by picure:
f(x) f(E(X)) f(x3) f(x1)
(B.1)
(B.2)
E(f(X))
x1
x2=E[X]
x3
Figure B.1: Example of Jensens inequality for a discrete random variable that takes only 3 values
with equal probability and a concave function f ().
132
Tse and Viswanath: Fundamentals of Wireless Communication 3. H (X ) H (X |Y ) = I (X ; Y ) =
x,y
133
p(x, y ) log p(x)p(y ) p(x, y )
p(x, y ) p(x)p(y ) (B.3) (B.4)
=
x,y
p(x, y ) log
log
x,y
p(x, y )
p(x)p(y ) p(x, y )
= log 1 = 0 where the inequality follows from Jensens inequality and the convexity of log(). We have equality i p(x)p(y ) = p(x, y ) for all x, y , i.e. X and Y are independent. For the required example consider X |y = 0 Bernoulli(1/2), X |y = 1 Bernoulli(0) and Y Bernoulli(1/2). It is easy to check that X Bernoulli(1/4) and we can compute the dierent entropies. H (X ) = H (1/4) = 0.811, H (X |y = 0) = H (1/2) = 1, H (X |y = 1) = 0, and H (X |Y ) = (1/2)H (X |y = 0) + (1/2)H (X |y = 1) = 1/2, where we used H (p) to denote the entropy of a Bernoulli(p) random variable. We see that H (X |y = 0) > H (X ) but H (X |Y ) < H (X ) in agreement with the inequality that we just proved. Exercise B.3. Exercise B.4. Exercise B.5. Exercise B.6. Exercise B.7. Exercise B.8. Exercise B.9. Exercise B.10. Exercise B.11. Exercise B.12.

**Find millions of documents on Course Hero - Study Guides, Lecture Notes, Reference Materials, Practice Exams and more. Course Hero has millions of course specific materials providing students with the best way to expand their education.**

Below is a small sample set of documents:

BC - HIST - HS012

March 10, 2012: origins of WWINot on midterm.-_* Triple entente* Triple alliance* Deadnaught* Schlieffen Plan* Serbia* Bosnia* Black Hnad* Archduke Ferdinand* Gavrilo Princip* WWI casualties: Russia the mist, USA the least* Russian and German e

BC - HIST - HS012

1Andrew ProcaccinoAtlantic Worlds II Essay Assignment: the hydrosphereHuman developments during industrialization and urbanization in the 19th and 20thcenturies greatly changed society for the better. Economic successes emerged from variousinventions

Oregon - GEOG - 141

s.l\rALAte .~o" _~ _ - ~'h~ ,oteo.UQf).(Neo-t~er()._Y~iO\1-~-J!.-'-!':'!'~'-~1-1- s"'dc~-ch~f1i~Cll\It ~Y~'-i~ 1.~___ (AVt~(7S.iS_( on3_:.Jet'tM._w(?(;\.tk.er -r~tf"rYt >_ oVr.? r tl-~L "~~-=-_~-,-~nJ_$-p~L~ Ctk1o\' i1~col'1 trol~_-_Th.Ot

Oregon - GEOG - 141

PY:e cCfi-f:o.-t;I') 5 ,~J-~IAi:eJT-:-~_ft"-~t'6h<;US.0I1t\.[~fcfw_",-tr6V1WtCl'lI.(YY\l.Wt~rcfw_hJ-,-I d.lL!h.il-L,-, _~e.(A r- (h;- r. , ~ IA VI ,<;'eQ. f"l) ,_~P iil:kJ/.Y0 ill_,-'-I-'-~-~_ve c;- M("':cfw_lo~IJII1~~ -+-"-'-'-~t~-L(-

Oregon - GEOG - 141

I .e.e_e et~-t\ovt,~So\-=-_ '.' II!,_,_'__O\V\d ~hey-' '-'j~,~t. ,_l~o(:th).t~H(;\1.,.~t.,.ye._e_ ~ _ ~_.-,_,-,_.PJj!~\~.~:~-~'"-'-'!'-:':'-'-'I.J.:.,~eV~Me-nt-_~.!~-.-~-i<-ty-p-~ ~_ S"~k-~\2on.sib~LC>\ 1-;.;- j)~2

Oregon - GEOG - 141

1\k,Q~bd w~tQ_y-fu~i;I1-,-_150ilwo-i~r~.J.~_~t~- - ,4,-.'_.~_ -,~_-_~:~,I!\,-,-.~-~-._ '~:LO~1~.rre~tyL~,_L.:_.t,oVt1~.-.-~- -.I- . - -_._--~-- - -_-,-~_:.:.-.~--._-_._-d -e l7 \ g n ~i 0 n .l_~_I-_L-,-. f=-_- _-t 0 r WI_t

Oregon - GEOG - 141

i I"'\'" tiS*-eC0'1s-fJ<,"> _ Yejj-"AlllV\~\A.t bf-ostVis",-=-cfw_ Ot-=r-I-~-,~I~-_=-v;"-h'~-, _-_&;c-f:t ccOfl1~i~,-'-.I_ '-:-_".._.__~.-~.-,-.-.-,_.-._-- _.p ,_ ,-C I,.-.,-.-.- .J.-~-.~~\_._---,Q11 I V\1.Cl~_._.-

Oregon - GEOG - 141

:~:(~t~;I;~1:J~tI-~-i:-;~:-:~'--,--:-. jQ.oJ.~-~-~- _ __~_",:r: /. ~o[0 V 0,_,.- - -'-.-'-,-~-'-.=~-~-~t_~-_,._-M.-.- -'-"-t~:t-.-'-'-,C~_;-;-~-,-,-,-,,_'o~-w~i~-~-"F-Ids ~('"-~-~rJ'_~I~-~_._ _ .~_.~hi~\:L~(~1t_L?_ _~t~_!_

Oregon - GEOG - 141

-~-.br~x1'1..~:=-~.~=-~._-._~=-=-_.=~=:=~_=-_Ct~.L~j~.Ti~"~=-.-..-=e~fui~Js~-_=t_-.b.-k-~1itz.tL-Jl:_~c- L-_._~<2LcL. ~-. -~[-on s, f;,J J5_~.J,-.-1--.--.__-_ce- -~v.f4-bt\G~;-s-eo.y~- ;TI.~-~>-t- eX.i~L_I_.a.r\'~_-.- ~- DC~-S=t ; cl~

Oregon - GEOG - 141

-~-,- M\Y\'<.~'II~II'_C\~ty~-ttb~Dtc.~r >-,_-+_n-=-(_k~s_ ose tY~nSYlYleJh~'" ~~iV\~rlAcJ~J~,\COv\l1~D-tht. E~ytk/~or ~.oJ\-+_M.:.!.OY-"-e-"=-s.~.vJ,htwl~l~(@ n~ov.>Wli~trt\,l~ tkOl,;t- titYe.-S\A~G_-.I,-,-l ~IS-'ck<"c",\V1I\

Oregon - GEOG - 141

~o~neJ_-.k~ _ we~jf1~_l7LI\~"isetOy- 17M. Scfw_~ \ e :- t_iv~l1sr-cfw_yt~ilOVlC:;Y0~-+( \.kY\~cfw_l\blt:"it ,17 _.-:-c;cfw_VeYlJb_.e Y- ce_:-LJe _-"o-.-=-\t 1~-t~Ylo.l~y~\Ncfw_lA.lqy-t~Olh_ _~-4-_-L:\Y\.:.-.-vJhiCh_-tk~t_-rlp'(e

Oregon - GEOG - 141

l=i.I'1:-~iV'ne S-cfw_o/ll.-+-WI11-LDl1ta ,'VIJ.y'5f ",tvZ~k(;(~-",-O~3_-PD_,-,"-,-Y~_\~f-~tt ree-1,.$-ff=v,yS iJ<'ve('Ep,OU"_~7'.f.Jt~>ollA.t(OVl~-. _1_[;La~A(;.(tI.-P-('hHstnh""J_hi $-1J1~-'-L_'_!IIYlP.;,.;y'-'-h1-'-'

Oregon - GEOG - 141

--C041he___~-(Ilc~ sJ~_Gft~-1~CL C~ .__._-O\_Gt~Gl~r_I-_:E~j: <;i veRrt\-~ut SD_ b7-_~lcQ. t=\e\r=---~S~\Qr_-.-~--SD-J~(LQ_' ~g_Ic~\~Y1J.- --'-,- JnYlf-t_rO;m_~1Y1~II_-~-_-_-~_-_-_-_-_G.E(.!'J-.~Jlj~-'--+-ICD

CSU San Marcos - BUS - 444

Chapter OneStrategic Leadership: Managing the g g StrategyMaking Process for Competitive AdvantageChapter Outline:Strategic Leadership, Competitive Advantage, Superior Performance Strategic Managers gy g Strategy-Making Process Strategy as an Emergent

CSU San Marcos - BUS - 444

Alaska Gold Mine*You have taken a three-month option on a possible gold mine in Alaska. It took you two months of dangerous journey to get there. After two weeks of exploration (and recuperation), you have got your health back except for your left hand,

CSU San Marcos - BUS - 444

Industry AnalysisBUS444 / Meilich / CSUSMStrategy DefinitionA strategy is a pattern or plan that integrates an organization's major goals, policies, and action sequences into a cohesive whole. A wellformulated strategy helps to marshal and allocate an

CSU San Marcos - BUS - 444

Chapter TwoExternal y Analysis: The Identification of Opportunities and ThreatsExternal AnalysisThe purpose of external analysis is to identify the strategic opportunities and threats in the organization's operating environment.External Analysis requi

CSU San Marcos - BUS - 444

Chapter ThreeInternal Analysis: Distinctive Competencies, Competitive Advantage, and Profitability"In preparing for battle I have always found that plans are useless, but planning is indispensable."- Dwight D. EisenhowerCopyright Houghton Mifflin Comp

CSU San Marcos - BUS - 444

Chapter FiveBuilding p Competitive Advantage Through BusinessLevel StrategyBusiness-Level StrategyA successful business model results from business-level strategies that create a competitive advantage over its rivals.Firms must decide/evaluate: 1. Cus

CSU San Marcos - BUS - 444

Chapter SixBusinessLevel Strategy and the Industry EnvironmentThe Industry EnvironmentThere is the need to continually formulate and implement business-level strategies to sustain competitive advantage over time in different industry environments.Diff

University of Phoenix - CJS - 250

In todays society when you talk about the Pinkerton clan, people think of the old west and thecivil war. Back then people were afraid of the Pinkertons. Americas first undercover agent wasa man who was the guardian for President Abraham Lincoln and who

Northwestern - PSYCH - 215-CN

Chapter 15We Use Mental Strategies to Maintain our Views of SelfMost people think of themselves in favorable termso Belongings, letters in their own name, etcBetter-than-average effect most people describe themselves as above average in everypossible

Northwestern - PSYCH - 215-CN

Chapter 16:What Determines the Quality of Relationships?Situational and Personal Factors Influence Friendshipso Festinger, Schacter, Back college dorm study Students more likely to become friends with people who they interact withmore often, many tim

Northwestern - PSYCH - 215-CN

Chapter 14Are Personality Disorders Truly Mental Disorders?Personality disorder a class of mental disorders marked by inflexible and maladaptiveways of interacting with the worldo Symptoms: indecisive, self-absorbed, emotionally unstableo Must be at

Northwestern - PSYCH - 215-CN

Chapter 12:Personality Reflects Learning and Cognitive ProcessesB.F. Skinner viewed personality as learned responses to patterns of reinforcement, asopposed to a result of internal processesPersonal constructs George Kelly emphasized importance of peo

Northwestern - PSYCH - 215-CN

Chapter 11:Brain Development Promotes LearningNewborns born with ability to process sensory stimuli (see, smell, hear taste, touch)o Infants prefer sweet tasteso Turn toward pad containing mothers breast milk (as opposed to anothermothers)o Startled

Northwestern - HISTORY - 377

14.) According to Duesberg, what was the cause of AIDS?immune overload theory; gay lifestylerepeated, constant infections may eventually overload the immune system, causing itsfailure (originally proposed Sonnabend) 134heavy drug use causing immunosup

Northwestern - HISTORY - 377

16) What assumptions about biochemistry and psychology are built into the PET process?Biochemistry: that the radioactivity is still there (half-life wise) and the tracer is on the rightmolecule and exists throughout the entire experiment. The high numbe

Northwestern - HISTORY - 377

What was at stake in the dispute between Dr. Gallo and the French researchers (J. L.Montagnier)? Physicians enlisted in Montagniers help to find the retrovirus with Gallos virus in mind(HTLV) However, Montagnier ended up discovering another virus, LAV,

Northwestern - HISTORY - 377

What is the immune overload theory? What was the evidence for this theory? What wereit social implications? Did gay activists like it?Immune overloado Represented initial medical frame for understanding the epidemico Syndrome essentially linked to gay

Northwestern - HISTORY - 377

What do you think are the main strengths and weaknesses of Joseph Dumits "PicturingPersonhood"? Make sure to discuss and evaluate specific claims made in the assignedchapters.- Strengths:o Goes through the entire process of PET and talks about each de

Northwestern - HISTORY - 377

Discuss at least four medical, social, ethical, and religious objections against the cosmeticuse of Prozac. Make sure to discuss the potential consequences that the introduction of newpsychoactive drugs has on social norms, individual psychological deve

Northwestern - HISTORY - 377

How and when was regular medicine professionalized (i.e. transformed into an occupationaccessible only to people who passed licensing examinations, belonged to professionalassociations, and attended acceptable medical schools?) Which factors made thepr

Northwestern - HISTORY - 377

3) What are the main strengths of Steve Epsteins Impure Science? How does Epsteinssociological approach compare to Bruno Latours? And how does it compare withRosenbergs social history approach in Cholera?IS+PF+++++++CY+++++Written By So

Northwestern - HISTORY - 377

2) How were syphilis and other venereal diseases perceived in turn-of-the-twentiethcentury America? How did cultural perceptions and social response to STDs affect publichealth policies? How did the management of venereal diseases in the army differ in t

Northwestern - HISTORY - 377

1) What caused the crisis of psychiatry in the last decades of the 19th century and in thefirst decades of the 20th? How did the practice of psychiatry and its assumptions changefrom the mid 19th century to the mid 20th century? Pay special attention to

Arkansas - CHEG - 4813

disconsider10^8*f*ilrn cfw_hrcfw_-'i'; l'f'tin^. t:il"Ytrcfw_thl i'l$jrv))- is)lriCile,Gqtl] '1ii' "'iIC *'s;t*,-6) ,*=l -,b)ls Cs-'s;ds' G ' * Sf l.'J Yqq* nh= aJ *tryr klsg, t1lil kohl' l*-l 1f\tu"ryil ilcfw_oM1!o,.rrn 0*,pAf".Tm-

Arkansas - CHEG - 4813

Bruno Paixao Martins HW 10 CHEG4183PropaneStatus RuleComm ; a)Comm ; Lp=0Comm ; Equation (9) from sheetSati (1/G^2)=1/(2*Cd^2*(Po-Pa)*)Comm ; b)Comm ; Lp=Le=0.1mComm ; Equation (8) from sheetSati G1=(2*Co^2*(Po-*Po)*+(H^2)/(To*Cp*V^2)^(1/2)Comm

Arkansas - CHEG - 4813

Arkansas - CHEG - 4813

Arkansas - CHEG - 4813

Arkansas - CHEG - 4813

Arkansas - CHEG - 4813

Arkansas - CHEG - 3143

HEAT TRANSPORTSPRING 2012HOMEWORK 1Due: January 25, 2012MARTINS1/20/12PROBLEM 1-1GIVEN:1. House; area is 2,000 ft with avg. height of 9ft.2. Sea level pressure, initial temperature inside 20 F, heated temperature is 70 F3. Heating condition A: c

Arkansas - CHEG - 3143

HEAT TRANSPORTHOMEWORK 2MARTINSSPRING 12Due: January 30, 20121/27/12PROBLEM 2-1GIVEN:1. A square cross section wire; it is 2 m long and 0.3 cm on a cross-section side.2. The surface temperature of the wire is 162 C and the ambient air temperature

Arkansas - CHEG - 3143

HEAT TRANSPORTHOMEWORK 3MARTINSSPRING 2012Due: February 5, 20122/03/12PROBLEM 3-1GIVEN: 1. A 0.54m x 0.54m thin square plate in a room at 27 C.2. One side of the plate is at 99 C while the other is insulated.REQD:1. The rate of heat transfer fro

Arkansas - CHEG - 3143

HEAT TRANSPORTHOMEWORK 4MARTINSSPRING 2012Due: February 13, 20122/07/12PROBLEM 4-1GIVEN:1. A 5-cm-diameter cylindrical uranium rods in a nuclear reactor. Length of the rods is 1m. (2-16)2. Heat is generated uniformly at a rate of 7 x 107 W/m3. (2

Arkansas - CHEG - 3143

taLtb298.980.278349.44xtfnxnt0349.442.54334.445.08321.677.61313.3310.15307.7812.69304.4415.22303.3317.76302.781 0.702735 0.449663 0.284384 0.174396 0.108205 0.086207 0.0753070.01270.02540.02540.02540.02540.02540.02540.01

Arkansas - CHEG - 3143

HEAT TRANSPORTHOMEWORK 5MARTINSSPRING 2012Due: February 20, 20122/17/12PROBLEM 5-1GIVEN:1. A two layer 0.6-in-thick wall made of sheetrock with thermal conductivity of k 1 = 0.10 Btu/hr*ft*F.2. The two layers are separated by a layer of gypsum, p

Arkansas - CHEG - 3143

HEAT TRANSPORTHOMEWORK 6MARTINSSPRING 2012Due: March 7, 20123/05/12PROBLEM 6-1GIVEN:1. A hollow, cylindrical copper fin of length L = 27.8 cm with outer and inside diameter of 0.635 cm and0.475 cm, respectively.2. In a forced convection experime

Arkansas - CHEG - 3143

HEAT TRANSPORTHOMEWORK 2SPRING 2012Due: March 16, 2012PROBLEM 7-1GIVEN:1. Chickens with average mass of 2.2 kg and average specific heat of 3.54 kJ/kg*C2. The chickens initially at 15C are cooled in a chiller at 0.5C at a rate of 500 chickens/hour.

Arkansas - CHEG - 3143

HEAT TRANSPORTHOMEWORK 2SPRING 2012Due: March 16, 2012PROBLEM 8-1GIVEN:1. 15 C gas is flowing over a 145 C, 1.5 m x 6 m flat plate.2. Measured Q = 10,000 W.3. = 0.87 kg/m3; Cp = 1,007 J/kg K; k = 0.023 W/m K; Pr = 0.72; = 1.83 x 10-5 kg/m s; = 2.

Arkansas - CHEG - 3143

HEAT TRANSPORTSPRING 2012PROBLEM 9-1GIVEN:REQD:1HOMEWORK 9Due: April 2, 2012MARTINSMarch 26, 2012(20 points)1. T he copper, 4 tube pass, Show and Tell S&T heat exchanger shown in Figures 1 & 2. The tubelayout pattern is close to square (Do = 1

Arkansas - CHEG - 3143

HEAT TRANSPORTHOMEWORK 9MARTINSSPRING 12Due: April 11, 20124/10/12Problem 10-1(30 points)GIVEN:1. An atmospheric distillation column, separating water and ethanol.2. The condenser duty is 250,000 Btu/hr (73,000 W).3. Ahe distillate is close to

Arkansas - CHEG - 3143

HEAT TRANSPORTMARTINSSPRING 2012MBASEMESTER PROJECTDue: April 30, 2012N NAApril 29, 2012INTRODUCTIONIt is well known that raw turkey and other poultry are highly contaminated with pathogens such asCampylobacter spp. (80-90%), Salmonellaspp. (15%

Arkansas - CHEG - 3143

DATE:TO:FROM:SUBJECT:January 18, 2012HEAT TRANSFER STUDENTSW. Roy PenneySpring 2012 Syllabus CHEG 3143 Heat TransportCOURSE CONDUCTThe course will be conducted in a manner to encourage you to attend class. The primary encouragementmeans will be:

Arkansas - CHEG - 4163

Exercise 2.1Subject: Minimum work for separating a hydrocarbon stream. Given: Component flow rates, ni , of feed and product 1, in kmol/h. Phase condition; temperature in K; enthalpy, h, in kJ/kmol; and entropy, s, in kJ/kmol-K for feed, product 1, and p

Arkansas - CHEG - 4163

Exercise 3.1Subject: Evaporation of a mixture of ethanol (AL) and ethyl acetate (AC) from a beaker into still air within the beaker. Given: Initial equimolar mixture of AL and AC, evaporating into still air at 0oC and 1 atm. Vapor pressures and diffusivi

Arkansas - CHEG - 4163

Exercise 9.1Subject: Selection of type condenser and operating pressure for distillation of propionic and n-butyric acids. Given: Distillate of 95 mol% propionic acid. Bottoms of 98 mol% n-butyric acid. Assumptions: Ideal solutions so that Raoult's K-val

Arkansas - CHEG - 4163

Exercise 15.1Subject: Estimation of adsorbent characteristics.Given: Porous particles of activated alumina. BET area = 310 m2/g = Sg. Particle porosity = 0.48 = p. Particle density = 1.30 g/cm3 = p. Assumptions: Straight pores of circular cross-section

Arkansas - CHEG - 4163

Exercise 16.1Subject: Mass balance check on test data around a leaching unit. Given: Feed rate to extractor of 6.375 lb/h with 10.67 wt% moisture and 0.2675 g oil per g dry, oil-free flakes. Solvent rate to extractor of 10.844 lb/h. Extract of 7.313 lb/h