Assume
x∼N(a,A)
x
∼
N
(
a
,
A
)
, prove that
Px∼N(Hx,PAPT)(266)
(266)
P
x
∼
N
(
H
x
,
P
A
P
T
)
where x∈Rn x ∈ R n , P∈Rm×n P ∈ R m × n . Notation N(a,A) N ( a , A ) denotes a Gaussian distribution with a a mean and A A covariance martix. In addition, (⋅)T ( ⋅ ) T refers to transport operation such as (P)T ( P ) T .
Proof : Let
y=Px
y
=
P
x
, then its CHF ( characteristic function ) is
ϕy(w)=E[exp(jwTy)]=E[exp(j(PTw)Tx)]=exp(j(PTw)Ta−12(PTw)TAPTw)=exp(jwT(Pa)−12wT(PAPT)w)(267)(268)(269)(270)
(267)
ϕ
y
(
w
)
=
E
[
exp
(
j
w
T
y
)
]
(268)
=
E
[
exp
(
j
(
P
T
w
)
T
x
)
]
(269)
=
exp
(
j
(
P
T
w
)
T
a
−
1
2
(
P
T
w
)
T
A
P
T
w
)
(270)
=
exp
(
j
w
T
(
P
a
)
−
1
2
w
T
(
P
A
P
T
)
w
)
Note that the CHF of x x is exp(jwTa−12wTAw) exp ( j w T a − 1 2 w T A w ) . Hence, the mean and variance of y y are Pa P a and PAPT P A P T , respectively.