EE6110 Adaptive Signal Processing Problem Set 3 Solutions
1.
Xˆ =K0Y, whereK0 is any solution to K0RY =RXY
We want to minimize E[( ˜X′)HWX˜′] for some W ≥0 ( ˜X′ =X−Xˆ′, Xˆ′ =KY)
E[( ˜X′)HWX˜′] =E[(X−KY)HW(X−KY)]
=E[(X−K0Y+K0Y−KY)HW(X−K0Y+K0Y−KY)]
=E[(X−K0Y)HW(X−K0Y)] +E[(X−K0Y)HW(K0Y−KY)]
+E[(K0Y−KY)HW(X−K0Y)] +E[(K0Y−KY)HW(K0Y−KY)]→eqn
We know that K0Y is the linear MMSE estimate ofX from Y
Therefore, we know (X−K0Y) is perpendicular to any linear function ofY
=⇒ second term in the eqn
E[(X−K0Y)HW(K0Y−KY)] =E[(X−K0Y)H(W(K0−K))Y] = 0 Also,E[(K0Y−KY)HW(X−K0Y)] = 0
=⇒ E[( ˜X′)HWX˜′] =E[(X−K0Y)HW(X−K0Y)] +E[(K0Y−KY)HW(K0Y−KY)]
Both the terms in the right hand side of the above equation are ≥0, (since W ≥0 =⇒ aHWa≥0,∀a)
=⇒ E[( ˜X′)HWX˜′]≥E[(X−K0Y)HW(X−K0Y)]
This lower bound is achieved by K =K0 i.e the linear MMSE estimate K0Y also minimizes E[( ˜X′)HWX˜′] for anyW ≥0
2.
J(x) = (x−c)HA(x−c)
Since A is Hermitian nonnegative-definite, J(x)≥0 1
=⇒ minimum possible value of J(x) is 0.
Forx=c+d where Ad=0,
J(x) = (c+d−c)HA(c+d−c) = dHAd=dH0= 0
ThusJ(x) = 0 is acheived at x=c+d for any d satisfying Ad=0
3.
Y1 =X1+N1, X1 is zero mean, variance 1
Y2 =X2+N2 , N1, N2 i.i.d zero mean ,variance σ2, independent of X1
X2 =ρX1+p
1−ρ2Z,
Z is zero mean, variance 1, independent of X1, X2, N1, N2 E[X22] =ρ2 + 1−ρ2 = 1
(a) EstimateX1 fromY1
Xˆ1 =k0Y1, k0(1 +σ2) = 1
=⇒ Xˆ1 = (1+σY12)
(b) EstimateX2 fromY1
Xˆ21 =ρXˆ1 = (1+σρY12)
(c) EstimateY2 fromY1
Yˆ2 =kY1, k(1 +σ2) =RY1Y2 =E[X1X2] =ρ Yˆ2 = ˆX2+ ˆN2 =ρXˆ1 = (1+σρY12)
=⇒ Yˆ2 = (1+σρY12)
(d) LetE2 =Y2−Yˆ2 =Y2− ρY1
(1+σ2)
EstimateX2 fromE2
Xˆ2e=keE2, keRE2 =RE2X2
2
RE2X2 =E[(Y2− (1+σρY12))X2] = 1− (1+σρ22)
RE2 =E[(Y2−Yˆ2)(Y2−Yˆ2)]
=E[(Y2)(Y2−Yˆ2)] →(1)
=E[Y22]−E[Y2Yˆ2]
= (1 +σ2)− ρ
(1 +σ2)E[Y2Y1]
= 1 +σ2− ρ
(1 +σ2)E[X2X1]
= 1 +σ2− ρ2 (1 +σ2)
(1) comes from the fact that (Y2−Yˆ2)⊥AY1 =⇒ (Y2−Yˆ2)⊥Yˆ2
So, ke = 1−
ρ2 (1+σ2)
1+σ2− ρ
2 (1+σ2)
=⇒ Xˆ2e= 1−
ρ2 (1+σ2)
1+σ2− ρ
2 (1+σ2)
(Y2− ρY1
(1+σ2)) (e)
Xˆ2 = ˆX21+ ˆX2e = ρY1
(1 +σ2) + 1−(1+σρ22) 1 +σ2− ρ2
(1+σ2)
(Y2− ρY1 (1 +σ2))
= σ2
1 +σ2− ρ2
(1+σ2)
! ρY1
(1 +σ2)
+ 1− ρ2
(1+σ2)
1 +σ2− ρ2
(1+σ2)
! Y2
=
ρσ2 (1 +σ2)2−ρ2
Y1+
1 +σ2−ρ2 (1 +σ2)2−ρ2
Y2
This can also be done directly by using ˆX2 =h k0i
"
Y1 Y2
#
h k0
i
=RX2YR−1Y
=h ρ 1i
"
(1 +σ2) ρ ρ (1 +σ2)
#−1
=h ρ 1i
"
(1 +σ2) −ρ
−ρ (1 +σ2)
#
1 (1 +σ2)2−ρ2
=⇒ Xˆ2 =
ρσ2 (1+σ2)2−ρ2
Y1+
1+σ2−ρ2 (1+σ2)2−ρ2
Y2
3