EE385 Class Notes 7/6/2015 John Stensby

Chapter 4 - Function of Random Variables

Let X denote a random variable with known density fX(x) and distribution FX(x). Let y =

g(x) denote a real-valued function of the real variable x. Consider the transformation

Y = g(X). (4-1)

This is a transformation of the random variable X into the random variable Y. Random variable

X() is a mapping from the sample space into the real line. But so is g(X()). We are interested

in methods for finding the density fY(y) and the distribution FY(y).
When dealing with Y = g(X()), there are a few technicalities that should be considered.

1. The domain of g should include the range of X.

2. For every y, the set {Y = g(X)  y} must be an event. That is, the set {  S : Y() = g(X())

 y } must be in F (i.e., it must be an event).

3.The events {Y = g(X) = ± } must be assigned a probability of zero.

In practice, these technicalities are assumed to hold, and they do not cause any problems.

Define the indexed set

Iy = { x : g(x)  y }, (4-2)

the composition of which changes with y. The distribution of Y can be expressed as

FY(y) = P[ Y  y ] = P[g(X)  y] = P[ X  Iy ]. (4-3)

This provides a practical method for computing the distribution function.

Example 4-1: Consider the function y = g(x) = ax + b, a  0, and b are constants.

I y = {x: g(x) = ax + b  y} = {x : x  (y - b)/a}

Updates at http://www.ece.uah.edu/courses/ee385/ 4-1

EE385 Class Notes 7/6/2015 John Stensby

y = g(x)=x2
y

x-axis
 y y

Figure 4-1: Quadratic transformation used in Example 4-2.

so that

yb yb
Fy (y)  P[X  I y ]  P[X  ]  Fx ( ).
a a

Example 4-2: Given random variable X and function y = g(x) = x2 as shown by Figure 4-1.

Define Y = g(X) = X2, and find FY(y). If y < 0, then there are no values of x such that x2 < y.

Hence

FY  y   0, y  0 .

If y  0, then x2  y for  y  x  y . Hence, Iy = {x : g(x)  y } = { y  x  y} , and

FY (y)  P   y  X  y   FX  y   FX  ( 
y ) , y  0.

Special Considerations
Special consideration is due for functions g(x) that have “flat spots” and/or jump
discontinuities. These cases are considered next.
Watch for places where g(x) is constant (“flat spots”). Suppose g(x) is constant on the
interval (x0, x1]. That is, g(x) = y1, x0 < x  x1, where y1 is a constant, and g(x)  y1 off x0 < x 
x1. Hence, all of the probability that X has in the interval x0 < x  x1 is assigned to the single
value Y = y1 so that

Updates at http://www.ece.uah.edu/courses/ee385/ 4-2

EE385 Class Notes 7/6/2015 John Stensby

P  Y  y1   P[ x 0  X  x1 ]  FX  x1   FX  x 0  . (4-4)

That is, FY(y) has a jump discontinuity at y = y1. The amount of jump is FX(x1) - FX(x0). As an
example, consider the case of a saturating amplifier/limiter transformation.
Example 4-3 (Saturating Amplifier/Limiter): In terms of FX(x), find the distribution FY(y) for
Y = g(X) where

 b, x>b


g(x) =  x, -b<x  b,

b, x  -b.


Both Fx and y = g(x) are illustrated by Figure 4-2.

Case: y  b
For this case we have g(x)  y for all x. Therefore FY(y) =1 for y  b.

Case: -b  y < b
For -b  y < b, we have g(x)  y for x  y. Hence, FY(y) = P(Y = g(X)  y) = FX(y), -b  y < b

Case: y < -b
For y < -b, we have g(x) < y for NO x. Hence, FY(y) = 0, y < -b
The result of these cases is shown by Figure 4-3.

FX(x)
y=g(x) 1
b

-b b x-axis
-b b

-b

Figure 4-2: Transformation y = g(x) and distribution Fx(x) used in Ex. 4-3.

Updates at http://www.ece.uah.edu/courses/ee385/ 4-3

EE385 Class Notes 7/6/2015 John Stensby

Watch for points where g(x) has a jump discontinuity Fy(y)

1
As shown by the following two examples, special
1-FX(b-)
care may be required when dealing with functions that
FX(-b)
have jump discontinuities.
-b b y-axis
Example 4-4: In terms of FX(x), find the distribution FY(y)
Figure 4-3: Result for Ex 4-3.
for Y = g(X), where

 x  c, x0

g(x)   y = g(x)

y-axis
 x-c, x<0

as depicted by Figure 4-4 to the right.

c
Case y  c: If y  c then g(x)  y for x  y-c. Hence,
FY(y) = FX(y-c) for y  c. x-axis

Case -c  y  c: If -c  y  c then g(x)  y for x  0. Hence,

-c
-
FY(y) = P[ X < 0 ] = FX(0 ) for -c  y  c.
Case y  -c: If y  -c then g(x)  y for x  y+c. Hence,
FY(y) = FX(y+c) for y  -c.
Figure 4-4: Transformation for
Example 4-4.
Example 4-5: In terms of FX(x), find the distribution FY(y)
for Y = g(X) where

 x  c, x0

g(x)  
 x-c, x  0

as depicted by Figure 4-5.

Case y  c: If y  c then g(x)  y for x  y-c. Hence,

Updates at http://www.ece.uah.edu/courses/ee385/ 4-4

EE385 Class Notes 7/6/2015 John Stensby

y axis
y = g(x)
FY(y) = FX(y-c) for y  c.

Case -c  y  c: If -c  y  c then g(x)  y for x  0. Hence,

c

FY(y) = P[ X  0 ] = FX(0) for -c  y  c. x-axis

-c
Case y  -c: If y  -c then g(x)  y for x  y+c. Hence,

FY(y) = FX(y+c) for y  -c.

Figure 4-5: Transformation for

Notice that there is only a subtle difference between the Example 4-5.
previous two examples. In fact, if FX(x) is continuous at x = 0, then FY(y) is the same for the
previous two examples.
Determination of fy in terms of fx
Determine the density fY(y) of Y = g(X) in terms of the density fX(x) of X. To
accomplish this, we solve the equation y = g(x) for x in terms of y . If g has an inverse, then we
can solve for a unique x in terms of y (x = g-1(y)). Otherwise, we will have to do it in segments.
That is, x1(y), x2(y), ... , xn(y) can be found (as solutions, or roots, of y = g(x) ) such that

y = g(x1(y)) = g(x2(y)) = g(x3(y)) = ... = g(xn(y)). (4-5)

Note that x1 through xn are functions of y. The range of each xi(y) covers part of the domain of
g(x). The union of the ranges of xi(y), 1  i  n, covers all, or part of, the domain of g(x). The
desired fY(y) is

f (x ) f (x ) f (x )
f Y (y)  X 1  X 2 +  + X n , (4-6)
g(x1 ) g(x 2 ) g(x n )

Updates at http://www.ece.uah.edu/courses/ee385/ 4-5

EE385 Class Notes 7/6/2015 John Stensby

y = g(x) = x 2
y

y
x1 x2

x1 x2 x-axis

Figure 4-6: Transformation y = g(x).

where g(x) denotes the derivative of g(x).

We establish this result for the function y = g(x) that is depicted by Figure 4-6, a simple
example where n = 2. The extension to the general case is obvious.

y y
P(y  Y  y  y)   f Y ()d  f Y (y)y
y

for small y (increments x1, x2 and y are defined to be positive). Similarly,

P(x1 -x1  X  x1 )  f X (x1 )x1

(4-7)
P(x 2 < X  x 2 +x 2 )  f X (x 2 )x 2

This leads to the requirement

P(y < Y  y + y)  P (x1 -x1  X  x1 ) + P (x 2 < X  x 2 +x 2 )

f Y (y)y  f X (x1 )x1  f X (x 2 )x 2 (4-8)

f (x ) f (x )
f Y (y)  X 1  X 2
 y   y 
 x   x 
 1  2

Now, let the increments approach zero. The positive quantities y/x1 and y/x2 approach

Updates at http://www.ece.uah.edu/courses/ee385/ 4-6

EE385 Class Notes 7/6/2015 John Stensby

x1  0 x1  0
x 2  0 x 2  0
y dg(x1 )
y  0 y dg(x 2 )
y  0
. 
 and 
 .. (4-9)
x1 dx x 2 dx

This leads to the desired result

f X (x1 ) f X (x 2 )
f Y (y)   . (4-10)
dg(x1 ) dg(x 2 )
dx dx

Example 4-6: Consider Y = aX2 where a > 0. If y < 0, then y = ax2 has no real solutions and

fY(y) = 0, y < 0. (4-11)

If y > 0, then y = ax2 has solutions x1  y / a and x 2   y / a . Also, note that g(x) = 2ax.
Hence,

f X (x1 ) f X (x 2 )
f Y (y)  
dg(x1 ) dg(x 2 )
dx dx

f ( y / a) f ( y / a )
 X  X , y >0 . (4-12)
2a y / a 2a y / a

=0 y<0

To see a specific example, assume that X is Rayleigh distributed with parameter . The density
for X is given by (2-24); substitute this density into (4-12) to obtain

Updates at http://www.ece.uah.edu/courses/ee385/ 4-7

EE385 Class Notes 7/6/2015 John Stensby

1 y  1 ya 
a exp  2 2 
2   
f Y (y)  U(y)
2a y a (4-13)

1  y 
 exp    U(y)
2
2 a  2 2 a 

which is the density for an exponential random variable with parameter  = 1/(22a), as can be
seen from inspection of (2-27). Hence the square of a Rayleigh random variable produces an
exponential random variable.
Expected Value of Transformed Random Variable
Given random variable X, with density fX(x), and a function g(x), we form the random
variable Y = g(X). We know that


Y  E[Y]   y f Y (y) dy (4-14)


This requires knowledge of fY(y). We can express Y directly in terms of g(x) and fX(x).
Theorem 4-1: Let X be a random variable and y = g(x) a function. The expected value of Y =
g(X) can be expressed as


Y  E[Y]  E[g(X)]   g(x) f X (x) dx (4-15)


y = g(x) = x 2
y

y
x1 x2

x1 x2 x-axis

Figure 4-7: Transformation used in discussion of Theorem 4-1.

Updates at http://www.ece.uah.edu/courses/ee385/ 4-8

EE385 Class Notes 7/6/2015 John Stensby

To see this, consider the following example that is illustrated by Figure 4-7. Recall that
f Y (y)y  f X (x1 )x1  f X (x 2 )x 2 . Multiply this expression by y = g(x1) = g(x2) to obtain

y f Y (y)y  g(x1 )f X (x1 )x1  g(x 2 )f X (x 2 )x 2 . (4-16)

Now, partition the y-axis as 0 = y0 < y1 < y2 < ..... , where y = yk+1 - yk, k = 0, 1, 2, ... . By the
mappings x1   y and x 2  y , this leads to a partition x1k, k = 0, 1, 2, ... ,of the negative
axis and a partition x2k, k = 0, 1, 2, ... ,of the positive x-axes. Sum both sides over their
partitions and obtain

  
 y k f Y (y k )y   g(x1k )fX (x1k )x1k   g(x 2k )f X (x 2k )x 2k . (4-17)
k 0 k 0 k 0

Let y  0, xk1 0 and xk2 0 to obtain

 0 
0 y f Y (y) dy  

g(x)f x (x)dx   g(x)f x (x)dx
0
, (4-18)

 g(x)f x (x)dx


the desired result. Observe that this argument can be applied to practically any function y = g(x).
Example 4-7: Let X be N(0,) and let Y = Xn. Find E[Y]. For n even (i.e., n = 2k) we
know, from Example 2-10, that E[Xn] = E[Xn] = 1·3·5  (n - 1)n. For odd n (i.e., n = 2k +
1) write

 2  2k 1
E[ X 2k 1 ]   x
2k 1 f (x)dx 
 x exp[ x 2 / 22 ] dx . (4-19)
 2  0

Updates at http://www.ece.uah.edu/courses/ee385/ 4-9

EE385 Class Notes 7/6/2015 John Stensby

Change variables: let y = x2/22, dy = (x/2)dx and obtain

 2k
1 2 k 1 x x dx
E[ X 2k 1 ]   (2  ) exp[ x 2 / 22 ]
2  0 (22 )k 2
(4-20)
2 (22 )k 1  k  y

 2 0 y e dy .

However, from known results on the Gamma function, we have


(k  1)   y k e y dy  k! . (4-21)
0

Now, use (4-21) in (4-20) to obtain

n 1  n 2 2
E[ X ] 
2 
- 
x exp[ x / 2 ]dx

 1  3  5  (n  1) n , n  2k (n even) (4-22)

 n21 ! n
n 1
2
2 2
, n  2k  1 (n odd)


for a zero-mean Gaussian random variable X.

Approximate Mean of g(X)
Let X be a random variable and y = g(x) a function. The expected value of g(X) can be
expressed as


E[ g(X) ]   g(x) f X (x) dx . (4-23)


To approximate this, expand g(x) in a Taylor's series around the mean  to obtain

Updates at http://www.ece.uah.edu/courses/ee385/ 4-10

EE385 Class Notes 7/6/2015 John Stensby

(x  )n
g(x)  g()  g()(x  )   +g (n) ()   . (4-24)
n!

Use this expansion in the expected-value calculation to obtain


E[ g(X) ]   g(x) f X (x) dx


  (x  ) n 
   g()  g()(x  )   +g (n) ()   f X (x) dx (4-25)
 n!
 

  
 g()  g() 2  g (3) () 3   + g (n) () n   + .
2! 3! n!

An approximation to E[g(X)] can be based on this formula; just compute a finite number of
terms in the expansion.
Characteristic Functions
The characteristic function of a random variable is


()   f X (x)e jx dx  E[e jX ] . (4-26)


Characteristic function  is complex valued with

 jx 
()   f X (x)e dx   f (x)dx  1 .
 X
(4-27)

Note that  is the Fourier transform of fX(x), so we can write

()  F[f (x)] , (4-28)

X 
replace  by -

Updates at http://www.ece.uah.edu/courses/ee385/ 4-11

EE385 Class Notes 7/6/2015 John Stensby

and

1 
f X (x)   ()e  jx dx . (4-29)
2 

Definition (4-26) takes the form of a sum when X is a discrete random variable. Suppose
that X takes on the values xi with probabilities pi = P[X = xi] for index i in some index set I (i 
I). Then the characteristic function of X is

 pi exp[ jxi ] .

()   f X (x)e jx dx  (4-30)

iI

Due to the delta functions in density fX(x), the integral in (4-30) becomes a sum.
Example 4-8: Consider the Gaussian density function

1 2 2
f X (x)  e x /2 . (4-31)
2

The Fourier transform of fX is F [fX(x)] = exp[-22/2], as given in common tables. Hence,

2 2
()  F [f (x)]  e  / 2 . (4-32)
X 
 -

1 2 2
If f X (x)  e (x ) / 2 , then
2

()  e j F[(1/ 2)e x /2 ]

2 2 2 2
 e j e  /2 . (4-33)
  fold 

Example 4-9: Let random variable N be Poisson with parameter . That is,

Updates at http://www.ece.uah.edu/courses/ee385/ 4-12

EE385 Class Notes 7/6/2015 John Stensby

n
P  N = n   e , n=0, 1, 2, ... . (4-34)
n!

From (4-30), we can write

    n
n  
()    exp[  ]
n!


exp[ j  n]  exp[  ]  
 n!
 exp[ jn]

n 0 n 0

  j  n
 e 
 exp[]   exp[]exp[e j ] (4-35)
n 0
n!

 exp[ e j  1]

as the characteristic function for a Poisson process.

Multiple Dimension Case
The joint characteristic function XY(1,2) of random variables X and Y is defined as

XY (1, 2 )  E  exp{j(1X  2 Y)}   e j(1x i 2 yk ) P[X = x i , Y  y k ] (4-36)

i k

for the discrete case and

 
XY (1, 2 )  E  exp{j(1X  2 Y)}    e j(1x 2 y) f XY (x, y) dxdy (4-37)
 

for the continuous case. Equation (4-37) is recognized as the two dimensional Fourier transform
(with the sign of j reversed) of fXY(x,y). Generalizing these definitions, we can define the joint
characteristic function of n random variables X1, X2, ... , Xn as

X1...X n (1, ... ,n )  E  exp{j1X1  ...  jn X n } . (4-38)

Updates at http://www.ece.uah.edu/courses/ee385/ 4-13

EE385 Class Notes 7/6/2015 John Stensby

Equation (4-38) can be simplified using vector notation. Define the two vectors

 1   X1 
   
   
   X2 
  2
   , X   . (4-39)
     
   
   
n   X n 

Then, we can write the n-dimensional characteristic function in the compact form

T 

 X ()  E e j X  . (4-40)
 

Equations (4-38) and (4-40) convey the same information; however, (4-40) is much easier to
write and work with.
Characteristic Function for Multi-dimensional Gaussian Case
   
Let X = [X1 X2 ... Xn]T be a Gaussian random vector with mean  = E[X ]. Let  = [
2 ... n]T be a vector of n algebraic variables. Note that

 1 
 
 
   2  n
T   [1 2  n ]     k k (4-41)
   k 1
 
 
 n 


is a scalar. The characteristic function of X is given as

  T    exp[ j
()  E exp[ j
T  1  T 
    ] . (4-42)
X] 2

Updates at http://www.ece.uah.edu/courses/ee385/ 4-14

EE385 Class Notes 7/6/2015 John Stensby

Application: Transformation of Random Variables

Sometimes, the characteristic function can be used to determine the density of random
variable Y = g(X) in terms of the density of X. To see this, consider


Y ()  E[e jY ]  E[e jg(X) ]   e jg(x) f X (x) dx . (4-43)


If a change of variable y = g(x) can be made (usually, this requires g to have an inverse), this last
integral will have the form


Y ()   e jy h(y) dy . (4-44)


The desired result fY(y) = h(y) follows (by uniqueness of the Fourier transform).
Example 4-10: Suppose X is N(0;) and Y = aX2. Then

2  2 2  jax 2  x 2 / 22
Y ()  E[e jY ]  E[e jaX ]   e jax f X (x) dx   e e dx .
 2 0

For 0  x < , note that the transformation y = ax2 is one-to-one. Hence, make the change of
variable y = ax2, dy = (2ax)dx = 2 ay dx to obtain

2  jy  y/ 2a2 dy

 jy e  y/ 2a
2 
  dy .
2  0 2 ay 0
Y ()  e e  e
  2ay 
 

Hence, we have

2
e y/ 2a
f Y (y)  U(y) .
 2ay

Updates at http://www.ece.uah.edu/courses/ee385/ 4-15

EE385 Class Notes 7/6/2015 John Stensby

Other Applications
Sometimes, a characteristic function is used to obtain qualitative results about a random
phenomenon of interest. That is, it is a tool that may be used to obtain qualitative results about a
random quantity of interest. For example, suppose we want to show that some random
phenomenon is Gaussian distributed. We may be able to do this by deriving the characteristic
function that describes the random phenomenon (and showing that the characteristic function has
the form given by (4-33)). In Chapter 9 of these notes, we do this for shot noise. We use
characteristic function theory to show that classical shot noise becomes Gaussian distributed as
its intensity parameter becomes large.
Moment Generating Function
The moment generating function is

 sx
 (s)   f (x)e dx  E[esX ] . (4-45)
 X

The nth derivative of  is

dn 
 (s)   x n f X (x)esx dx  E[X n esX ] , (4-46)
n 
ds

so that

dn
 (s)  E[X n ]  m n . (4-47)
n s  0
ds

Example 4-11: Suppose X has an exponential density f X (x)  e x U(x) . Then the moment
generating function is

Updates at http://www.ece.uah.edu/courses/ee385/ 4-16

EE385 Class Notes 7/6/2015 John Stensby

 
 (s)   ex esx dx  .
0  s

This can be differentiated to obtain

d
 (s)  E[X]  1
ds s  0 

d2
 (s)  E[X 2 ]  2 2 .
2  
ds s 0

From this, we can compute the variance as

 
2 2
2  E[X 2 ]   E[X]  2 2  1  1 2.
  

Theorem 4-2
Let X and Y be independent random variables. Let g(x) and h(y) be arbitrary functions. Define
the transformed random variables

Z  g(X)
(4-48)
W  h(Y).

Random variables Z and W are independent.

Proof: Define

A z  [x : g(x)  z]

Bw  [y : h(y)  w].

Updates at http://www.ece.uah.edu/courses/ee385/ 4-17

EE385 Class Notes 7/6/2015 John Stensby

Then the joint distribution of Z and W is

FZW (z, w)  P[Z  z, W  w]  P[g(X)  z, h(Y)  w]  P[X  A z , Y  Bw ] .

However, due to independence of X and Y,

FZW (z, w)  P[X  A z , Y  Bw ]  P[X  A z ] P[Y  Bw ]

 P[g(X)  z] P[h(Y)  w]  P[Z  z] P[W  w] ,

 Fz (z)FW (w) ,

so that Z and W are independent.

One Function of Two Random Variables
Given random variables X and Y and a function z = g(x,y), we form the new random
variable

Z = g(X,Y). (4-49)

We want to find the density and distribution of Z in terms of like quantities for X and Y. For
real z, denote Dz as

Dz = {(x,y) : g(x,y)  z}. (4-50)

Now, note that Dz satisfies

{Z  z} = {g(X,Y)  z} = {(X,Y)  Dz}, (4-51)

Updates at http://www.ece.uah.edu/courses/ee385/ 4-18

EE385 Class Notes 7/6/2015 John Stensby

so that y-axis

FZ (z)  P[Z  z]  P[(X, Y)  D z ]   f XY (x, y) dxdy . (4-52)

Dz
x x-axis
x+yz +
y
=
z
Thus, to find FZ it suffices to find region DZ for every z
and then evaluate the above integral.
Example 4-12: Consider the function Z = X + Y. The Figure 4-8: Integrate over the
shaded region to obtain FZ.
distribution FZ can be represented as

FZ (z)   f XY (x, y) dxdy .

x  yz

In this integral, the region of integration is depicted by the shaded area shown on Figure 4-8.
Now, we can write

 z y
FZ (z)    f XY (x, y) dxdy .
 

By using Leibnitz’s rule (see below) for differentiating an integral, we get the density

d d  zy
f Z (z)  FZ (z)    f XY (x, y) dxdy
dz dz  


 f XY (z  y, y) dy .


Leibnitz’s Rule: Consider the function of t defined by

b(t)
F(t)   (x, t)dx .
a(t)

Updates at http://www.ece.uah.edu/courses/ee385/ 4-19

EE385 Class Notes 7/6/2015 John Stensby

fX(x) = e-xU(x) fy(y)

1

x-axis -1/2 1/2 y-axis

Figure 4-9: Density functions used in Example 4-13.

Note that the t variable appears in the integrand and limits. Leibnitz’s rule states that

d d b(t) b(t) (x, t) db(t) da(t)

F(t)   (x, t)dx   dx + (b(t), t)  (a(t), t) .
dt dt a(t) a(t) t dt dt

Special Case: X and Y Independent.

Assume that X and Y are independent. Then fXY(z-y,y) = fX(z-y)fY(y), and the previous
result becomes


f Z (z)   f X (z  y)f Y (y) dy , (4-53)


the convolution of fX and fY.

fX(z-y) 1
Example 4-13: Consider independent random variables fy(y)

X and Y with densities shown by Figure 4-9. Find

density fZ that describes the random variable Z = X + Y. -1/2 1/2 y-axis

CASE I: z < - 1/2 (see Fig 4-10) y=z

Figure 4-10: Case I: z < -½
There is no overlap, so fZ(z) = 0 for z < - 1/2.
1
CASE II: - 1/2 < z < 1/2 (see Fig 4-11) fy(y)
fX(z-y)
z
f z (z)   e (z  y) dy -1/2 1/2 y-axis
1/ 2

y=z
 1  e(z 1/ 2) ,  1/ 2  z  1/ 2
Figure 4-11: Case II: -½ < z < ½.

Updates at http://www.ece.uah.edu/courses/ee385/ 4-20

EE385 Class Notes 7/6/2015 John Stensby

CASE III: 1/2 < z (see Fig 4-12) 1

fy(y)
fX(z-y)
1/ 2
f z (z)   e(z  y) dy
1/ 2 y-axis
-1/2 1/2
1
 e z [e1/ 2  e1/ 2 ], z y=z
2
Figure 4-12: Case III: ½ < z.

As shown by Figure 4-13, the final result is

0.7

Y-Axis
0.6
f z (z)  0, z  1/ 2
0.5

 1  e(z 1/ 2) , 1/ 2  z  1/ 2 0.4

0.3
 [e1/ 2  e1/ 2 ]e  z , 1/ 2  z
0.2

0.1

Example 4-14: Let X and Y be random variables. 0.0

Consider the transformation -1 0 1 2 3 4 5

X-Axis
Figure 4-13: Final result for Example 4-13.
ZX/Y.

For this transformation, we have

Dz = { (x,y) : x/y  z },

the shaded region on the plot depicted by Figure 4-14. Now, compute the distribution

 yz 0 
Fz (z)    f XY (x, y)dxdy +  yz f XY (x, y)dxdy .
0

The density fZ(z) is found by differentiating FZ to obtain

Updates at http://www.ece.uah.edu/courses/ee385/ 4-21

EE385 Class Notes 7/6/2015 John Stensby

y 
y-axis
Dz is the shaded region

yz
x

x=
Drawn for case z > 0

x-axis

x

y 
Figure 4-14: Integrate over shaded region to obtain FZ for Example 4-14.

d  0 
f z (z)  Fz (z)   y f xy (yz, y)dy    y f xy (yz, y)dy   y f xy (yz, y)dy
dz 0  

Example 4-15: Consider the transformation Z  X 2  Y 2 . For this transformation, the region

DZ is given by

 
D z  (x, y) : x 2  y 2  z  (x, y) : x 2  y 2  z 2  (4-54)

the interior of a circle of radius z > 0. Hence, we can write

Fz (z)  P[Z  z]  P[(x, y)  Dz ]   f (x, y) dxdy . (4-55)

Dz

Now, suppose X and Y are independent, jointly Gaussian random variables with

1  (x 2  y2 ) 
f XY (x, y)  exp    (4-56)
22  22 

Updates at http://www.ece.uah.edu/courses/ee385/ 4-22

EE385 Class Notes 7/6/2015 John Stensby

Substitute (4-56) into (4-55) to obtain y-axis

1  (x 2  y 2 ) 
 exp  22  dxdy

r
Fz (z)  
22 D  
z x-axis

x  r cos 
To integrated this, use Figure 4-15, and
y  r sin 
transform from rectangular to polar     
coordinates Rectangular-to-Polar Transformation

r  x 2  y2 , r0

  tan 1 (y / x),      d

dA
dA  r dr d  .

rd
dr


r
The change to polar coordinates yields


1  r2 
2 z
Fz (z)      22  r dr d
22 0 0
exp
 
Cut-away view detailing differential area
dA = rdrd.
The integrand does not depend on  so the Figure 4-15: Figures that supports Example 4-15.
integral over  is elementary. For the
integral over r, let u = r2/22 and du = (r/2)dr to obtain

z  r 2  dr z 2 / 2 2  u
Fz (z)   exp    r  0 e du
0  2 2  2

2 2
 1  e  z / 2 , z  0 ,

Updates at http://www.ece.uah.edu/courses/ee385/ 4-23

EE385 Class Notes 7/6/2015 John Stensby

so that

d z  z 2 / 2 2
f z (z)  Fz (z)  e , z0
dz 2

a Rayleigh density with parameter  Hence, if X and Y are identically distributed, independent
Gaussian random variables then Z  X 2  Y 2 is Rayleigh distributed.
Two Functions of Two Random Variables
Given random variables X and Y and functions z = g(x,y), w = h(x,y), we form the new random
variables

Z = g(X,Y) (4-57)
W = h(X,Y).

Express the joint statistics of Z, W in terms of functions g, h and fXY. To accomplish this, define

Dzw = {(x,y) : g(x,y)  z, h(x,y)  w }. (4-58)

Then, the joint distribution of Z and W can be expressed as

FZW (z, w)  P[(X, Y)  D zw ]   f XY (x, y)dxdy . (4-59)

Dzw

Example 4-16: Consider independent Gaussian X and Y with the joint density function

1  (x 2  y2 ) 
f XY (x, y)  exp   .
22  2  2


Updates at http://www.ece.uah.edu/courses/ee385/ 4-24

EE385 Class Notes 7/6/2015 John Stensby

Define random variables Z and W in terms of X and Y by y-axis

y = xw
the transformations
tan-1(w)
Dzw
Z  X2  Y2 Dzw x-axis

W Y/X.
Drawn for case
w>0

Figure 4-16: Integrate over

Find FZW, FZ and FW. First, define region Dzw as shaded region to obtain FZW for
Example 4-16.

DZW = {(x,y) : x2 + y2  z2, y/x  w }.

Dzw is the shaded region on Figure 4-16. The figure is drawn for the case w > 0 (the case w < 0
gives results that are identical to those given below). Now, integrate over Dzw to obtain

 Tan 1w z 1  r 2 / 22 

FZW (z, w)   f XY (x, y) dxdy  2    0 22 e r dr d  

D zw  2 

  Tan 1 (w)
z 1  r 2 / 2 2
 2
 
0 2
e r dr ,

which leads to

 
  tan 1 (w)
2 2
FZW (z, w)  2 1  e z / 2 , z  0,   w  
 (4-60)

 0, z  0,   w  

Note that Fzw factors into the product FZFW, where

Updates at http://www.ece.uah.edu/courses/ee385/ 4-25

EE385 Class Notes 7/6/2015 John Stensby

2 2
FZ (z)  {1  e z / 2 }U(z)
(4-61)
FW (w)  1  1 Tan 1(w) ,    w  
2 

Note that Z and W are independent, Z is Rayleigh distributed and W is Cauchy distributed.
Joint Density Transformations: Determine fZW Directly in Terms of fXY.
Let X and Y be random variables with joint density fXY(x,y). Let

z = g(x,y) (4-62)
w = h(x,y)

be (generally nonlinear) functions that relate algebraic variables x, y to the algebraic variables z,
w. Also, we assume that g and h have continuous first-partial derivatives at the point (x,y) used
below. Now, define the new random variables

Z = g(X,Y) (4-63)

W = h(X,Y).

In this section, we provide a method for determining the joint density fZW(z,w) directly in terms
of the known joint density fXY(x,y).
First, consider the relatively simple case where (4-62) can be inverted. That is, it is
possible to solve (4-62) for unique functions

x  (z, w)
(4-64)
y   (z, w)

Updates at http://www.ece.uah.edu/courses/ee385/ 4-26

EE385 Class Notes 7/6/2015 John Stensby

that give x, y in terms of z, w. Note that z = g((z,w),(z,w)) and w = h((z,w),(z,w)) since

(4-64) is the inverse of (4-62) . Later, we will consider the general case where the
transformation cannot be inverted.
The quantity P[z < Z  z + dz, w < W  w + dw] is the probability that random variables
Z and W lie in the infinitesimal rectangle R1 illustrated on Figure 4-17. The area of this
infinitesimal rectangle is AREA(R1) = dzdw. The vertices of the z-w plane rectangle R1 are the
points

P1 = (z, w)

P2 = (z, w+dw)
(4-65)
P3 = (z+dz, w+dw)

P4 = (z+dz, w).

z, w plane x, y plane

dz
w-axis  y-axis P 2
P2 P3
R1 dw R2 P 3
P 1
P1 P4
g,h P 4

z-axis x-axis

P1 = (z, w) P1 = (x, y)

P2 = (z, w+dw)  
P2 = (x+ dw, y+ dw)
w w
P3 = (z+dz, w+dw)
   
P4 = (z+dz, w) P3 = (x+ dz+ dw, y+ dz+ dw)
z w z w
 
P4 = (x+ dz, y+ dz)
z z
Figure 4-17: (z,w) and (x,y) planes used in transformation of two random variables.
Functions  transform from z,w plane to x,y plane. Functions g,h transform from x,y
plane to z,w plane.

Updates at http://www.ece.uah.edu/courses/ee385/ 4-27

EE385 Class Notes 7/6/2015 John Stensby

The z-w plane infinitesimal rectangle R1 gets mapped into the x-y plane, where it shows up as
parallelogram R2. As shown on the x-y plane of Figure 4-17, to first-order in dw and dz,

parallelogram R2 has the vertices

   
P1 = (x, y) P3 = (x+ dz+ dw, y+ dz+ dw)
z w z w
(4-66)
   
P2 = (x+ dw, y+ dw) P4 = (x+ dz, y+ dz).
w w z z

The requirement that (4-64) have continuous first-partial derivatives was used to write (4-66).
Note that P1 maps to P1, P2 maps to P2, etc (it is easy to show that P2 - P1 = P3 - P4 and
P4 - P1 = P3 - P2 so that we have a parallelogram in the x-y plane). Denote the area of
x-y plane parallelogram R2 as AREA(R2)
If random variables Z, W fall in the z-w plane infinitesimal square R1, then the random
variables X, Y must in the x-y plane parallelogram R2, and vice-versa. In fact, we can claim

P  z < Z  z + dz, w < W  w + dw   P  x < X  x + dx, y < Y  y + dy 

 f ZW (z, w) dz dw =  f XY (x, y) dx dy , (4-67)

R1 R2

f ZW (z, w) AREA(R1 )  f XY (x, y) AREA(R2 )

where the approximation becomes exact as dz and dw approach zero. Since AREA(R1) = dzdw,
Equation (4-67) yields the desired fXY once an expression for AREA(R2) is obtained.
Figure 4-18 depicts the x-y plane parallelogram R2 for which area AREA(R2) must be
 
obtained. This parallelogram has sides P1 P2 and P1 P4 (shown as vectors with arrow heads on
Fig 4-18) that can be represented as

Updates at http://www.ece.uah.edu/courses/ee385/ 4-28

EE385 Class Notes 7/6/2015 John Stensby

P2

R2 P3
P1

P4

Figure 4-18: Parallelogram in x-y plane.

      
P1 P4   dz  iˆ   dz  ˆj
 z   z 
, (4-68)
      
P1 P2   dw  iˆ   dw  ˆj
 w   w 

where iˆ and ĵ are unit vectors in the x and y directions, respectively. Now, the vector cross
   
product of sides P1 P4 and P1 P2 is denoted as P1 P4  P1 P2 . And, the area of parallelogram R2
   
is the magnitude P1 P4 P1 P2 sin() =  P1 P4  P1 P2 , where  is the positive angle between
the vectors. Since iˆ  ˆj  kˆ , ˆj  iˆ  kˆ , ˆj  ˆj = iˆ  iˆ = kˆ  kˆ = 0, we write

 iˆ ˆj kˆ 
     
   z
      w 
AREA(R2 )  P1 P4  P1 P2  det  dz dz 0   det   dzdw. (4-69)
z z    
   
     z w 
 dw dw 0 
 w w 

In the literature, the last determinant on the right-hand-side of (4-69) is called the Jacobian of
the transformation (4-64); symbolically, it is denoted as J(xy); instead, the notation
(x,y)/(z,w) may be used. We write

 x x     
 (x, y   z w   z w 
J (x, y)   det    det   . (4-70)
 (z w   y y     
   
 z w   z w 

Updates at http://www.ece.uah.edu/courses/ee385/ 4-29

EE385 Class Notes 7/6/2015 John Stensby

Finally, substitute (4-69) into (4-67), cancel out the dzdw term that is common to both sides, and
obtain the desired result

 (x, y 
f ZW (z, w)  f XY (x, y) , (4-71)
 (z w  
x (z,w)
y  (z,w)

a formula for the density fZW in terms of the density fXY. It is possible to obtain (4-71) directly
from the change of variable formula in multi-dimensional integrals; this fact is discussed briefly
in Appendix 4A.
It is useful to think of (4-69) as

 (x, y 
AREA(R2 )  AREA(R1 ) , (4-72)
 (z w 

a relationship between AREA(R2) and AREA(R1). So, the Jacobian can be thought of as the
“area gain” imposed by the transformation (the Jacobian shows how area is scaled by the
transformation).
By considering the mapping of a rectangle on the x, y plane to a parallelogram on the z,
w plane (i.e., in the argument just given, switch planes so that the rectangle is in the x-y plane
and the parallelogram is in the z-w plane) , it is not difficult to show

 (z, w 
f XY (x, y)  f ZW (z, w) , (4-73)
 (x y 

where (x,y) and (z,w) are related by (4-62) and (4-64). Now, substitute (4-73) into (4-71) to
obtain

Updates at http://www.ece.uah.edu/courses/ee385/ 4-30

EE385 Class Notes 7/6/2015 John Stensby

 (z, w   (x, y 
f ZW (z, w)  f ZW (z, w) , (4-74)
 (x y   (z w 

where (x,y) and (z,w) are related by (4-62) and (4-64).

Equation (4-74) leads to the conclusion

1
 (x, y 
  (z, w  , (4-75)
 (z w 
 (x y 

where (x,y) and (z,w) are related by (4-62) and (4-64).

Sometimes, the Jacobian (z,w)/(x,y) is easier to compute than the Jacobian
(x,y)/(z,w); Equation (4-75) tells us that the former is the numerical inverse of the latter. In
terms of the Jacobian (z,w)/(x,y), Equation (4-71) becomes

f XY (x, y)
f ZW (z, w)   , (4-76)
 (z w  x (z,w)
 (x, y  y  (z,w)

which may be easier to evaluate than (4-71).

Often, the original transformation (4-62) does not have an inverse. That is, it may not be
possible to find unique functions  and  as described by (4-64). In this case, we must solve
(4-62) for its real-valued roots xk(z,w), yk(z,w), 1  k  n, where n > 1. These n roots depend on
z and w; each of the (xk,yk) “covers” a different part of the x-y plane. Note that

z = g(xk,yk), w = h(xk,yk) (4-77)

for each root, 1  k  n. For this case, a simple extension of (4-71) leads to

Updates at http://www.ece.uah.edu/courses/ee385/ 4-31

EE385 Class Notes 7/6/2015 John Stensby

n
 (x, y  
f ZW (z, w)   f XY (x, y) 
 (z w  
, (4-78)
k 1
 (x, y)  (x k , y k )

and the generalization of (4-76) is

 f XY (x, y)
n 
f ZW (z, w)     (z w   . (4-79)
 
k 1   (x, y  
  (x, y)  (x , y )
k k

That is, to obtain fZW(z,w), we should evaluate the right-hand-side of (4-71) (or (4-76)) at each of
the n roots xk(z,w), yk(z,w), 1  k  n, and sum up the results.
Example 4-17: Consider the linear transformation

z = ax + by   z  a b   x 
     ,
 
    
w = cx + dy  w  c d   y

where ad - bc  0. This transformation has an inverse. It is possible to express

1
 x  a b  z x = Az+ Bw
     
     
 y c d w  y = Cz+Dw,

where A, B, C and D are appropriate constants (can you find A, B, C and D??). Now, compute

a b 
 (z w     ad  bc
 det
 (x, y   
c d 

Updates at http://www.ece.uah.edu/courses/ee385/ 4-32

EE385 Class Notes 7/6/2015 John Stensby

If X and Y are random variables described by fXY(x,y), the density function for random variables
Z = aX + bY, W = cX + dY is

f XY (Az  Bw, Cz+Dw)

f zw (z, w)  .
ad-bc


Example 4-18: Consider X, an n1, zero-mean Gaussian random vector with positive definite
 
covariance matrix x. Define Y = AX, where A is an nn nonsingular matrix. Note that

 y1   a11  a1n   x1 
     n
    
         yi   a ik x k , 1 i  n
     k 1
 y n  a n1  a nn   x n 


As discussed previously, the density for X is

 1  
f x (X)  exp   1 X T  x1 X 
(2) n / 2  x
1/ 2  2 


Since A is invertable, we can write fY(Y) as


 f X (X)
f Y (Y)  
 (Y)

 ,

 (X) X=A-1 Y
 

where

Updates at http://www.ece.uah.edu/courses/ee385/ 4-33

EE385 Class Notes 7/6/2015 John Stensby

 y1 / x1 y1 / x 2  y1 / x n   a11 a12  a1n 

    
 (Y)  y 2 / x1 y 2 / x 2  y 2 / x n  a 21 a 22  a 2n 
  det    det  
 (X)          
   
 y n / x1 y n / x 2  y n / x n   a n1 a n2  a nn 

 det[A]

is the absolute value of the determinant of the matrix A. Note that

     T  
   
 
 Y  E  Y Y T   E  AX AX   A E  X X T  A T  A  X A T
 

 
T
 X  A 1 Y A 1

This leads to the result

 1  
f Y (Y)  exp   1 (A 1Y)T  x1 A 1Y 
(2)n / 2  x
1/ 2
det A  2 

  T 1 T 1 1   ,
1
 exp  Y (A )  x A Y 
 1
(2) n/2
x
1/ 2
det A  2  
     Y1 
1/2
Y

a result rewritten as

 1  1   ,
f Y (Y)  exp   1 Y T  Y Y
1/ 2  2 
(2)n / 2  Y


where Y = AxAT is the covariance of Gaussian random vector Y. This example leads to the

Updates at http://www.ece.uah.edu/courses/ee385/ 4-34

EE385 Class Notes 7/6/2015 John Stensby

general, very important result that linear y-axis


transformations of Gaussian random variables

r




produces Gaussian random variables (remember

this!!). x-axis
Example 4-19 (Polar Coordinates): Consider the
transformation

Figure 4-19: Polar coordinate transfor-

mations used in Example 4-19.
r  x 2  y2 , 0r

  Tan 1 (y / x),     

that is illustrated by Figure 4-19. With the limitation of  to the ( ] range, the
transformation has the inverse

x = r cos()
y = r sin()

 cos   r sin  
 (x, y)
 det  r
 (r, )  
sin() r cos  

so that

 (x, y)
f r (r, )  f XY (x, y)
 (r, ) x  r cos 
y  r sin 

 r f XY (r cos , r sin )

for r > 0 and - <   . Suppose that X and Y are independent, jointly Gaussian, zero mean

Updates at http://www.ece.uah.edu/courses/ee385/ 4-35

EE385 Class Notes 7/6/2015 John Stensby

with a common variance 2. For this case, the above result yields

 (x, y) r  {r cos }2  {r sin }2 

f r (r, )  f XY (x, y)  r f XY (r cos , r sin )  exp   
 (r, ) x  r cos  22  2 2 
y  r sin 

1 r  r 2 
 exp  .
2  2 2
 
 2 
 
f ( )

 f r (r)

Note that r and  are independent, r is Rayleigh and  is uniform over (-].
Example 4-20: Consider the random variables Z = g(X,Y) and W = h(X,Y) where

z = g(x, y)  x 2  y 2
. (4-80)
w = h(x, y)  y / x

Transformation (4-80) has roots (x1, y1) and (x2, y2) given by

x1  z(1  w 2 ) 1/ 2 , y1  wx1  wz(1  w 2 )1/ 2

(4-81)
x 2   z(1  w 2 ) 1/ 2 , y 2  wx 2   wz(1  w 2 ) 1/ 2

for -  < w <  and z > 0; the transformation has no real roots for z < 0. A direct evaluation of
the Jacobian leads to

 z z 
 x y   x(x 2  y 2 )½ y(x 2  y 2 ) ½ 
 (z w    det  ,
 det 
 (x, y   w w   2 
    y / x 1/ x 
 x y 

Updates at http://www.ece.uah.edu/courses/ee385/ 4-36

EE385 Class Notes 7/6/2015 John Stensby

which can be expressed as

 (z w 
 (x 2  y 2 )½ 1  y 2 / x 2  . (4-82)
 (x, y 

When evaluate at both (x1, y1) and (x2, y2), the Jacobian yields

 (z w   (z w  1 w2
    . (4-83)
 (x, y  x , y )  (x, y   x , y ) z
1 1 2 2

Finally, application of (4-78) leads to the desired result

z
f ZW (z, w)  f XY (x1, y1)  f XY (x 2 , y2 ) , z  0,   w   , (4-84)
1 w2

where (x1,y1) and (x2,y2) are given by (4-81). If, for example, X and Y are independent, zero-
mean Gaussian random variables with the joint density

1
f XY (x, y) = exp  (x 2 + y 2 ) / 22  , (4-85)
2
2

then we obtain the transformed density

z 1/ 
f ZW (z, w)  exp   z 2 / 22  U(z)  f Z (z)f W (w) (4-86)
2   1 w2

where

Updates at http://www.ece.uah.edu/courses/ee385/ 4-37

EE385 Class Notes 7/6/2015 John Stensby

z
f Z (z)  exp  z 2 / 22  U(z)
 2  
(4-87)
1/ 
f W (w) 
1 w2

Thus, random variables Z and W are independent, Z is Rayleigh, and W is Cauchy.

Linear Transformations of Gaussian Random Variables
Let yi, 1  i  n, be zero mean, unit variance, independent (which is equivalent to being
uncorrelated in the Gaussian case) Gaussian random variables. Define the Gaussian random
    
vector Y = [y1 y2  yn]T. Note that E[Y ] = 0 and the covariance matrix is y = E[ Y Y T ] =
I, an n  n identity matrix. Hence, we have

 1  
f (Y)  exp   1 Y T Y  . (4-88)
(2)n / 2  2 

Now, let A be an n  n nonsingular, real-valued matrix, and consider the linear transformation

 
X  AY . (4-89)

  
The transformation is one-to-one. For every Y there is but one X , and for every X there is but
   
one Y = A-1X . We can express the density of X in terms of the density of Y as


 f y (Y)
f x (X)  (4-90)
abs[J]  
Y  A 1X

where

Updates at http://www.ece.uah.edu/courses/ee385/ 4-38

EE385 Class Notes 7/6/2015 John Stensby

 x1 y1 x1 y 2  x1 y n 

 
 
 x 2 y1 x 2 y 2  x 2 y n 
J  det    det[A]  A  0 . (4-91)
    
 
 
x n y1 x n y 2  x n y n 

Hence, we have

 1 
f x (X)  f Y (A 1X)
A

1  
 exp   1 (A 1X)T A 1X  (4-92)
(2) n / 2 A  2 

1  
 exp   1 X T (A 1 )T A 1X  ,
(2) n / 2 A  2 

which can be written as

 1  
f x (X)  exp   1 X T  x1X  , (4-93)
(2) n / 2  x
1/ 2  2 

where  x1  (A 1 )T A 1 , which leads to the requirement that

x = AAT. (4-94)

Since A is nonsingular (a requirement on the selection of A),  is positive definite. In this

development, we used x = AAT = AAT = A2 so that A = x1/2
 
It is important to note that X = A Y is zero mean Gaussian with a covariance matrix
given by x = AAT. Note that a linear transformation of Gaussian random variables produces

Updates at http://www.ece.uah.edu/courses/ee385/ 4-39

EE385 Class Notes 7/6/2015 John Stensby

Gaussian random variables.


Consider the converse problem. Given zero mean Gaussian vector X with positive
 
definite covariance matrix x. Find a non-singular transformation matrix A so that X = AY,

where Y is zero mean Gaussian with covariance matrix y = I (identity matrix). The implication
 
is profound: Y = A-1X says that it is possible to transform a Gaussian vector with correlated
entries into a Gaussian vector made with uncorrelated (and independent) random variables. We
can remove correlation by properly transfoming the original vector. Clearly, we must find a
matrix A that satisfies

AAT = x. (4-95)

The solution to this problem comes from linear algebra. Given any positive definite
symmetric matrix x, there exists a nonsingular matrix P such that

PTxP = I, (4-96)

which means that x = (PT)-1P-1 = (P-1)TP-1 (we say that x is congruent to I). Compare this to
the result given above to see that matrix A can be found by using

A = (P-1)T = (PT)-1. (4-97)

The procedure for finding P is simple:

1) Use the given x to write the augmented matrix [ x  I ]
2) Do elementary row and column operations until the augmented matrix becomes [ I  PT ] .
The
elementary operations are
i) interchange two rows (columns)

Updates at http://www.ece.uah.edu/courses/ee385/ 4-40

EE385 Class Notes 7/6/2015 John Stensby

ii) multiply a row (column) by a scalar

iii) add a multiple of one row (column) to another row (column).
3) Write the desired A as A = (PT)-1.
Example 4-21: Suppose we are given the covariance matrix

 1 2 
x   .
 
 2 5 

First, write the augmented matrix

 1 2 1 0
 x      
   
 2 5 0 1

1) Add to 2nd row 2first row to obtain

 1 2 1 0
 
 
0 1 2 1

2) Add to 2nd column 2first column to obtain

1 0 1 0 
  = [ I  PT]
 
0 1 2 1 

1 0 
3) P T   
 
2 1

Updates at http://www.ece.uah.edu/courses/ee385/ 4-41

EE385 Class Notes 7/6/2015 John Stensby

 1 0
4) A = (PT )-1   
 
 2 1 

Check Results: is PTxP = I ? (Yes!) Check Results: is AAT = x? (Yes!)

Example 4-22: Consider the covariance matrix

 2 0 3
 
 
x   0 1 0 
 
 3 0 10 

Now, write the augmented matrix

       
 
 
[ x         
 
       

Add to 3rd row 3/2 times 1st row. Add to 3rd column 3/2 times 1st column

      
 
 
     
 
       
   

Multiply 1st row by 1/ 2 . Multiply 1st column by 1/ 2

Updates at http://www.ece.uah.edu/courses/ee385/ 4-42

EE385 Class Notes 7/6/2015 John Stensby

     
  
 
     
 
      
   

Multiply 3rd row by 2 /11 . Multiply 3rd column by 2 /11 .

      
  
 
        [ IP T ]
 
    2  2 
  11 11 

Finally, compute

 2 0 0 
 
 
A  (PT )1   0 1 0 
 
 3 2 0 11 
 2 2 

Check Results: x = AAT ? (YES!)

Updates at http://www.ece.uah.edu/courses/ee385/ 4-43