Problem Set 7: Y 2 X Y X ∞ −∞ X dω 2π
Problem Set 7: Y 2 X Y X ∞ −∞ X dω 2π
Problem Set 7: Y 2 X Y X ∞ −∞ X dω 2π
Spring 2016
Problem Set 7
Due Date: 5/4
1. Problem 1
(a) True. Since SY () = |H()|2 SX (), it follows that SY () SX () for
R all . Integrating over all frequencies yields the result, because the power of X is SX () d
2 and
similarly for Y .
(b) True. The process Y is WSS because X is WSS and the system is time-invariant.
Since
is periodic with some period T . Thus, RY ( + T ) =
R X is WSS and periodic, RXR
e
e
RX ( + T t)(h h)(t)dt = RX ( t)(h h)(t)dt = RY ( ), so that RY is also
periodic with period T . Therefore, Y is periodic.
(c) False. The output power is zero if (and only if) SX ()|H()|2 0. This happens, for
example, if H() = I{||2} and SX () = I{4||6} .
2. Problem 2
(Note: Since cos(80t + ) = cos() for any integer time t, parts (a) and (b) of the problem as stated are degenerate. It would make more sense either in continuous time, or if the
frequency of the cosine term were non-integer, such as Yt = Xt cos(2t/40 + ). The form
of the answer given here would be correct for non-integer frequencies, thinking of SX as a
period 2 function.)
(a) Since and (( + c) mod 2) have the same distribution, the process cos(80t + ) is
stationary. Two independent stationary processes are jointly stationary, and the product
of two jointly stationary processes is also stationary.
(b) Using the fact cos(a) cos(b) = (cos(a b) + cos(a + b))/2, we find
RY (s, t) = E[Xs Xt ]E[cos(80(s t)) + cos(80(s + t) + 2)]/2 = RX (s, t) cos(80(s
)
t))/2. So RY ( ) = cos(80
RX ( ). In the transform domain this is SY () = 41 [SX (
2
80) + SX ( + 80)].
(c) The impulse
is given by h( ) =
P response function
a
j
k
H() = k=0 a[(1 a)e
] = 1(1a)e
j .
k=0 a(1
(e) Examining H further shows that |H()| = 1 if and only if is a multiple of 2, and in
that case H() = 1. Now, if X is a periodic process with power one and period one (or
a smaller period that divides one), then Y will have power 0.5 and also have period one
(or a smaller period that divides one) because the frequencies in Y are the frequencies
in X plus or minus 40 Hz. Then, Y will pass through the filter H unchanged, and the
power of Z will therefore be 0.5.
3. Problem 3
(a) RY X ( ) = h RX ( ) =
h and of RX yields
h(
Z
RY X (0) = h RX (0) =
Z a/2
2
et dt
a 0
2
2
a 1 a 2
1 a 3
a/2
=
(1 e
)=
11+
+
+
a
a
2 2 2
3! 2
a
a a2
+ = 1 + o(a)
= 1 +
4 24
4
R
h)( t)RX (t)dt. First we find that h e
h is the
(b) RY ( ) = h e
h RX ( ) = (h e
triangle function over the interval [a, a] with height 1/a. Thus, taking = 0 and using
the symmetry of h e
h and of RX yields
Z
RY (0) =
(h e
h)(0 t)RX (t)dt
Z a
2
=
(a t)et dt
a2 0
2
2 a2 a3 a4
a
=
(a 1 + e ) = 2
a2
a
2
3!
4!
a a2
a
= 1 +
+ = 1 + o(a)
3 12
3
=
(c) E[|Xt Yt |2 ] = E[(Xt Yt )(Xt Yt ) ] = RX (0) RXY (0) RY X (0) + RY (0). Since
RX (0) = 1 and RXY (0) = RY X (0) = RY X (0), it follows that
E[|Xt Yt |2 ] = 1 2RY X (0) + RY (0) = 1 2(1 a4 + o(a)) + 1 a3 + o(a) = a6 + o(a)
4. Problem 4
(a) Since
X()
+
we have
K() =
Y () = Y (),
1 + j
Y ()
1 + j
=
=1+
1 + j
1 + j
X()
(b) The impulse response is the inverse Fourier transform of K(), which is
e(1)t u(t),
if 1 0
k(t) =
e(1)t u(t), if 1 < 0
(c)
Z
1
1
, |1 | < 1
(1)2
SY ()d
2
2
1+
Z
sup
= sup|K()|2 = sup
= , |1 | = 1
2
2
1
(1 ) +
SX
SX ()d
2
1, |1 | > 1
When |1 | 1, the supremum can be achieved by choosing SX () in the form of
SX () = c(), c > 0. When |1 | > 1, the supremum cannot be achieved but
approached by choosing SX () = c(( 0 ) + ( + 0 )), c > 0 and let 0 approach
infinity.
5. Problem 5
(a) By the orthogonality principle, the estimator is optimal if and only if
bn+1
Xn+1 X
1
bn+1 Xi
Xn+1 X
for 1 i n
b
Xn+1 Xn+1 Xi Xl for 1 i l n
or equivalently,
Pn Pj
(a) h0 + j=1 k=1 h2 (j, k)E[Xj Xk ] = 0
(b) for 1 i n,
Pn
Pn Pj
E[Xn+1 Xi ] h0 E[Xi ] k=1 h1 (k)E[Xk Xi ] j=1 k=1 h2 (j, k)E[Xj Xk Xi ] = 0
(c) for 1 i l n,
Pn
Pn Pj
E[Xn+1 Xi Xl ] h0 E[Xi Xl ] k=1 h1 (k)E[Xk Xi Xl ] j=1 k=1 h2 (j, k)E[Xj Xk Xi Xl ] = 0
6. Problem 6
(a) Since X is a Gaussian process, X[0,T ] are jointly Gaussian. Then, the optimal estimate
on XT is a linear function of X[0,T ] . Hence, without loss of optimality we can assume that
R
T = T g(t)Xt dt. By orthogonality principle,
the optimal estimator has the form of X
0
RT
g(t)X
dt
is
the
linear
MMSE
estimator
if
and
only if
t
0
E
Z
XT
0
g(t)Xt dt Xu = 0, u [0, T ],
RX (T u) =
0
e|T u|
Z
=
g(t)e
(1)
|tu|
dt, u [0, T ]
(b) It is easy to verify that g(t) = (tT ) satisfies (1) for all u [0, T ]. Hence,
XT is the linear MMSE estimator and also the MMSE estimator.
RT
0
g(t)Xt dt =
7. Problem 7
(a) Since RX is a sinc function scaled by fo , its Fourier transform is a rectangle function
of width fo centered at the origin, and RX (0) is the integral of the Fourier transform.
2
SX (2f )
So SX (2f ) = Afo I{|f |fo /2} . Also, SN 2 . Thus, H(2f ) = SX (2f
)+SN (2f ) =
A2 I{|f |fo /2}
,
A2 + 2 fo
A2 fo sinc(fo t)
.
A2 + 2 fo
(b) Since X and N are jointly Gaussian, and Y and D are obtained from them by linear
operations, the processes X, N , Y , and D are jointly Gaussian. All processes have mean
R
2 2 f
SN
o
zero. The mean square error is given by the formula MSE= SSXX+S
df = AA2 +
2f .
o
N
Thus, for t fixed, Dt is a N (0, M SE) random variable.
(c) The conditional distribution is also N (0, M SE), because by the orthogonality principle,
Dt is orthogonal, and hence uncorrelated with, the process Y , and because uncorrelated
jointly Gaussian random variables are independent. Therefore, Dt is independent of the
random process Y .
(d) First, RDY 0 by the orthogonality principle. It follows also that RDXb 0. Since X =
b +D and R b 0, it follows that RD = RX R b . But S b (2f ) = |H(2f )|2 SY (f ) =
X
DX
X
X
A4 I{|f ||fo |/2}
.
fo (A2 + 2 fo )
A4 sinc(fo t)
,
A2 + 2 fo
so that RD (t) =
A2 2 fo sinc(fo t)
.
A2 + 2 fo