Sei sulla pagina 1di 12

T

AE4003 Problems 1
X ( t ) = Lt c
³ A cos ( 2π f t + Θ ) dt
T →∞ 2T
−T
Problem 1
=0
Consider a random process consisting of a sine wave with a random phase T
1 2
X ( t + τ ) X ( t ) = Lt ³A cos ( 2π f c t + 2π f cτ + Θ ) cos ( 2π f c t + Θ ) dt
T →∞ 2T
−T
X(t)= Acos(2πfct + Θ)
2 T
A
= Lt c c c
A, fc are constants and Θ is a random variable with a uniform distribution. T →∞ 4T ³ cos ( 4π f t + 2π f τ + 2Θ ) + cos ( 2π f τ ) dt
−T
Evaluate the autocorrelation function and mean of the random process using ensembled
A2
and time averaging and hence show that the random process is ergodic in both. = 0+ cos ( 2π f cτ )
2
Use the autocorrelation function to find the power spectral density function of X(t). What = cos ( 2π f cτ )
is the mean square value of X(t)?
The results are identical for both emsembled and time averaging and hence the random
By ensemble averaging process is ergodic in both its autocorrelation function and mean.
RX (τ ) = E { X ( t + τ ) X ( t )}
Using RX(τ) we find the power spectral density function of X(t) by finding its Fourier
= E { A2 cos ( 2π f c t + 2π f cτ + Θ ) cos ( 2π f c t + Θ )}
Transform. The mean square value of X(t) is simply RX(0) or the integral of the power
A2 A2 spectral density function over the entire frequency axis.
= E {cos ( 4π f c t + 2π f cτ + 2Θ )} + E {cos ( 2π f cτ )}
2 2
2 2 A2
A A RX (τ ) = cos ( 2π f cτ )
= .0 + cos ( 2π f cτ ) 2
2 2
2 A2
A S X ( f ) = ℑ ª¬ RX (τ ) º¼ = ªδ ( f − f c ) + δ ( f + f c ) º¼
= cos ( 2π f cτ ) 4 ¬
2 2
A
E ª¬ X 2 ( t ) º¼ = RX ( 0 ) =
1 2
fΘ = or


∞ A2 A2 A2
mX = A cos ( 2π f c t + Θ ) f Θ ( Θ ) d Θ E ª¬ X 2 ( t ) º¼ = X
³ S ( f ) df = + =
³ −∞
4 4 2
−∞

1 Problem 2
= c
³ A cos ( 2π f t + Θ ) 2π d Θ
0

=0 Consider the problem of generating a new random process Y(t) by mixing (i.e.
multiplying) a given random process X(t) with a sine wave with a random phase. The
By time averaging random phase function of the sine wave has a uniform density. Find the autocorrelation
function of Y(t) in terms of the autocorrelation function RX(τ) of X(t). Find also the
power spectral density function of Y(t) in terms of the power spectral density function
SX(f) of X(t).
Y ( t ) = X ( t ) cos ( 2π f c t + Θ ) Problem 4
RY (τ ) = E {Y ( t + τ ) Y ( t )} A comb filter for a random signal X(t) is formed by combining (taking difference in this
= E { X ( t + τ ) cos ( 2π f c t + 2π f cτ + Θ ) . X ( t ) cos ( 2π f c t + Θ )} case) the direct input signal with its delayed quantity using a delay line which gives a
time delay of T seconds. If the power spectral density function of X(t) is SX(f), what is
= E { X ( t + τ ) X ( t )}.E {cos ( 2π f c t + 2π f cτ + Θ ) cos ( 2π f c t + Θ )} the power spectral density function SY(f) of the comb filter output? Identify the nulls and
1 maximums of the comb filter output as function of frequencies.
= RX (τ ) . E {cos ( 4π f c t + 2π f cτ + 2Θ ) + cos ( 2π f cτ )}
2
The transfer function H(f) of the comb filter is obtained in frequency domain by wrting
1 the output signal Y(f) as the difference of the direct input signal 1.X(f) and the delayed
= RX (τ ) cos ( 2π f cτ )
2 signal e-j2πfT.X(f). H(f) is defined as Y(f)/X(f).
Y(f )
H(f)= = 1 − e− j 2π fT = 1 − cos ( 2π fT ) + j sin ( 2π fT )
SY ( f ) = ℑ ¬ª RY (τ ) ¼º X(f)
1 2 2 2
= ª S X ( f − f c ) + S X ( f + f c ) º¼ H ( f ) = ¬ª1 − cos ( 2π fT ) ¼º + ¬ªsin ( 2π fT ) ¼º

= 2 ª¬1 − cos ( 2π fT ) º¼
= 4sin 2 (π fT )
Problem 3 2
SY ( f ) = H ( f ) S X ( f ) = 4sin 2 (π fT ) S X ( f )
Consider the problem of creating a random process X(t) by adding white noise w(t) to a
sine wave Acos(2πfct + Θ) with a uniformly distributed random phase Θ. Find the Note that SY(f) has nulls at fT=n and maximums at fT=(2n+1)/2, n=0,1,2,...
autocorrelation function RX(τ). How can the sine wave component in X(t) be detected in
practice?

X (t ) = A cos ( 2π f ct + Θ ) + w ( t ) Problem 5
The sine wave and white noise components are independent
X and Y are two random variables whose joint probability density function is defined by
RX (τ ) = E { X ( t ) X ( t + τ )}
­1 0 ≤ y ≤ 2 x, 0 ≤ x ≤ 1
f XY ( x, y ) = ®
= E ª¬ A cos ( 2π f c t + Θ ) + w ( t ) º¼ ª¬ A cos ( 2π f c t + 2π f cτ + Θ ) + w ( t + τ ) º¼
{ } ¯ 0 elsewhere
2
= E {A ª¬ cos ( 2π f c t + Θ ) º¼ ª¬cos ( 2π f c t + 2π f cτ + Θ ) º¼ + E Aw ( t ) ª¬cos ( 2π f c t + 2π f cτ + Θ ) º¼
} { } Find the probability density function fY(y). What is the conditional probability density
+ E { A cos ( 2π f c t + Θ ) w ( t + τ )} + E {w ( t ) w ( t + τ )} fX|Y(x) when y=1/2? Evaluate the expectation (i.e. mean) of X conditioned on y=1/2.
A2 N
= cos ( 2π f cτ ) + 0 + 0 + 0 δ (τ )
2 2
A2 N0
= cos ( 2π f cτ ) + δ (τ )
2 2

The autocorrelation function RX(τ) is a sine wave of amplitude A2/2 and period 1/fc
which can be detected for |τ|≠0.
Problem 6
y
X and Y are two random variables whose joint probability density function is defined by
y=2x
­1
° , x2 + y 2 ≤ 1
fXY=1 f XY ( x, y ) = ® π
y=1/2 °̄ 0, elsewhere
fXY=1

x By finding the probability density function fY(y) and fX(x), show that X and Y are not
x=1 independent. What are the expectations (or means) of X and Y?
By evaluating the cross correlation function between X and Y, show that X and Y are
uncorrelated.

y

y=√(1-x2)
fY ( y ) = ³ f XY ( x, y ) dx
−∞
1 x= -√(1-y2) x=√(1-y2)
y
= ³ 1dx 0 ≤ y ≤ 2
y (x,y) or (r,θ)
2

y
= 1− , 0≤ y≤2 x x
2
f ( x, y ) Unit Circle
f X |Y ( x ) = XY
fY ( y ) y= -√(1-x2)
1 4 1
→ f X |Y ( x ) y = 1 = = , ≤ x ≤1
2
1 3 4
1−
4
1

1
º= x f
y= X |Y ( x ) y = 1 dx
E ª« X
¬ 2 ¼» ³ 1 2
4
1
1
4 4 ª x2 º
= xdx = « »
3 ³1 3¬ 2 ¼1
4 4

4 ª1 1 º 5
= « − »=
3 ¬ 2 32 ¼ 8
Density functions: Problem 7
1− y 2

fY ( y ) = f XY ( x, y ) dx The Gauss-Markov process X(t) has the autocorrelation function RX(τ)= σ2e-β|τ|. A new
³ random process Y(t) is obtained by scaling X(t) and adding a constant component.
− 1− y 2
Y(t)= aX(t) + b, where a,b are constants. Find the autocorrelation function RY(τ) of Y(t)
1− y 2
1 2 and the cross correlation function RXY(τ) between X(t) and Y(t), in terms of the
= ³ dx = 1 − y2 , −1 ≤ y ≤ 1
π π autocorrelation function RX(τ) and the mean mX of X(t).
− 1− y 2

1− x 2
RY (τ ) = E {Y ( t ) Y ( t + τ )} = E ª¬ aX ( t ) + b º¼ ª¬ aX ( t + τ ) + b º¼
{ }
fX ( x) = ³ f XY ( x, y ) dy
2 2
− 1− x 2 = E {a X ( t ) X ( t + τ ) + abX ( t ) + abX ( t + τ ) + b }
1− x 2
1 2 2 = a 2 RX (τ ) + 2abmX + b 2
= ³ dx = 1 − x , −1 ≤ x ≤ 1
π − 1− x 2
π
∴ f XY ( x, y ) ≠ f X ( x ) . fY ( y ) RXY (τ ) = E { X ( t ) Y ( t + τ )} = E X ( t ) ª¬ aX ( t + τ ) + b º¼
{ }
Hence X, Y are not independent.
= E {aX (t ) X ( t + τ ) + bX ( t )}
Expectations (or means) of X and Y:
1
= aRX (τ ) + bmX
3
1 1 2 2
2 1 (1 − x )
1 − x 2 dx =
3
E { X } = ³ xf X ( x ) dx = ³x
−1
π −1
π Problem 8
2 −1

=0 If Z(t) is a random process obtained by adding two Normally distributed random


1 processes X(t) and Y(t), show the Z(t) is also normally distributed. It is given that for
E {Y } = ³ yf ( y ) dyY
random processes related by Z=X+Y, the Fourier Transforms of their density functions
−1 fZ , fY and fY are related by F(fZ) = F(fX) . F(fY).
=0
Cross correlation function:
2
1 x = 1− y 1 x = 1− y 2

E { XY } = ³³ xyf XY ( x, y ) dxdy = ³ ³ xy.1dxdy = ³y ³ xdxdy


( area ) x 2 + y 2 ≤1 −1 x =− 1− y 2 −1 x =− 1− y 2

=0
alternatively
θ = 2π r =1
sin 2θ
E { XY } = r 2 sin (θ ) cos (θ ) f XY ( r , θ ) rdrdθ = ³ r 3 drdθ
=0
r ≤1, 0 ≤ ≤ 2π
³³θ θ 2 r³=0

1 − cos 2θ
= =0
8 2 0

Hence X, Y are not correlated.

This shows that if X and Y are not correlated, they are not necessarily independent.
Given : Problem 9
Z = X +Y with densities f z , f x , f y
The multi-variate Normally distributed random variables are defined by an n-vector X =
ℑ ( f Z ) = ℑ ( f X ) ⋅ ℑ ( fY ) [X1 X2 ...Xn]T with mean mX= [ m1 m2 ... mn]T and variances σ12, σ22 ,... , σn2 . The multi-
→ f Z = ℑ−1 ª¬ ℑ ( f X ) ⋅ ℑ ( fY ) º¼ variate random variables are jointly normal and the joint probability density function is
defined in terms of the covariance matrix C as:
x2 σ X2 ω 2 T
− −
1 2σ X2 2
fX = e → ℑ( f X ) = e [C ] = E {( X − m )( X − m ) }
2πσ X 2
ª .
y2
E ( X 1 − m1 ) E {( X 1 − m1 )( X 2 − m2 )} E {( X 1 − m1 )( X n − mn )} º
σ Y2ω 2
1 − −
« { } »
2σ Y2 2
fY = e → ℑ ( fY ) = e « 2 »
2πσ Y E {( X 2 − m2 )( X 1 − m1 )} E ( X 2 − m2 ) { } . E {( X 2 − m2 )( X n − mn )}»

« »
∴ « . . . . »
2 2 « 2 »
ª − σ X +σ Y ω 2 º E
«¬ {( nX − m X
n )( 1 − m1 )} E X − m
{( n n )( X 2 − m2 )} . E ( X n − mn )
{ } »¼
f Z = ℑ−1 ª¬ ℑ ( f X ) ⋅ ℑ ( fY ) º¼ = ℑ−1 «e 2 »
¬« ¼»
T


z2 1 − 12 ( x − m ) C −1 ( x − m )
{ }
1 2 σ X2 +σ Y2
fX ( x) = n 1 e
( ) 2 2
= e ( 2π ) C
2π (σ X2 + σ Y2 )
Hence Z is also Normally distributed.
For the case of two normal random variables with X= [ X1 X2]T with mean m= [m1 m2]T,
Note the Fourier Transform Pair used above:
variances σ12 and σ22 and correlation between X1 and X2 given by the coefficient ρ,
ω2
2 2 π − obtain the joint probability function fX(x). Show that if the normal random variables are
4 a2
f ( t ) = e − a t ⇔ ℑ ª¬ f ( t ) º¼ = e uncorrelated, they are also independent.
a
Let
2
t ω2 ω2
1 ª 1 − 2 º 1 π − σ2 − σ2
2 2
a2 = 2
→ ℑ« e 2σ » = e =e
2σ «¬ 2πσ »¼ 2πσ 1

ªX º ªm º ªx º
X = « 1» m = « 1» x = « 1 » (realizations of X) Covariance Matrix C is:
¬X2 ¼ ¬ m2 ¼ ¬ x2 ¼
2
ª E {X12 } E {X1 X 2 } E {X1 X 3 } º
ª E ( X 1 − m1 )
{ E {( X 1 − m1 )( X 2 − m2 )}º
} « » ª RX ( 0 ) RX ( 0.5 ) RX (1) º
» « »
C=«
2
« » C= « E {X 2 X 1} E {X 22 } E {X 2 X 3 }» = « RX ( 0.5 ) RX ( 0 ) RX ( 0.5 ) »
E X − m X − m E X − m « »
{( 2 2 )( 1 1 )} ( 2 2 ) { }
¬« ¼» « E {X 3 X 1} E {X 3 X 2 } E {X32 } » «¬ RX (1) RX ( 0.5 ) RX ( 0 ) »¼
2 2
¬ ¼
2 2
E 1 1 1 2 2 2
{( X − m ) } = σ ª 100 100e−1 100e −2 º
E {( X − m ) } = σ
« »
E {( X 1 − m1 )( X 2 − m2 )} = E {( X 2 − m2 )( X 1 − m1 )} = Cov X 1 & X 2 = «100e −1 100 100e −1 »
«100e−2 100e−1 100 »
= ρ Var X 1 Var X 2 = ρσ 1σ 2 ¬ ¼
­ ª x1 º ½
ª σ2 ρσ 1σ 2 º 1 ª σ 22 − ρσ 1σ 2 º ° « »°
− 12 ®[ x1 x2 x3 ]C −1 « x2 » ¾
→C = « 1 »= « » C = (1 − ρ 2 ) σ 12σ 22 1 ° °
¯ ¬« x3 ¼» ¿
¬ ρσ 1σ 2 σ 22 ¼ C ¬ − ρσ 1σ 2 σ 12 ¼ f X1 X 2 X 3 ( x1 , x2 , x3 ) = 3 1 e
2 2
( 2π ) C
ª 1 −ρ º
« 1− ρ 2 σ 2
1 « ( ) 1 (1 − ρ 2 )σ12 »»
C −1 =
C « −ρ 1 » Problem 11
« 2 2 2 2
»
¬« (1 − ρ ) σ 1 (1 − ρ )σ 2 ¼» Find the time averaged autocorrelation function RX(τ) of a random binary sequence X(t)
1 ­° 1 −1
ª x − m1 º ½° with equally probable and discrete output values of +A, -A. Evaluate also the mean and
f X1 X 2 ( x1 , x2 ) = exp ®− [ x1 − m1 x2 − m2 ] C « 1 »¾ variance of X(t). The pulse interval is T sec. Assume that the random variables giving the
2
2πσ 1σ 2 1− ρ ¯° 2 ¬ x2 − m2 ¼ ¿°
outputs in different pulse intervals are independent.
2
1 ­° 1 ª ( x1 − m1 ) 2
= − 2 ρ
( x1 − m1 ) ( x2 − m2 ) + ( x2 − m2 ) º ½°
exp ®− 2
« 2 »¾
2πσ 1σ 2 1 − ρ 2 °¯ 2 (1 − ρ ) «¬ σ 1 σ 1σ 2 σ 22 »¼ °¿ |τ|
If X1 and X 2 are not correlated, ρ =0
2 2
1 °­ 1 ª ( x − m ) ( x − m ) º °½ +A
→ f X1 X 2 ( x1 , x2 ) = exp ® − « 1 2 1 + 2 2 2 » ¾
2πσ 1σ 2 2 ¬« σ 1 σ2
¯° ¼» ¿°
( x1− m1 )2 ( x2 −m2 )2
− 12 − 12
1 σ12 1 σ 22
= e . e = f X1 ( x1 ) . f X 2 ( x2 )
2πσ 1 2πσ 2
Hence X1 and X 2 are independent.

Problem 10
-A
Consider a Gauss-Markov process X(t) with an autocorrelation function RX(τ)=100e-2|τ|
and zero mean. Since X(t) is a Gaussian process with zero mean, we can generate three
normal random variables from X(t) by taking the sample realizations at three instances T
X(0), X(0.5) and X(1.0). The correlation between the three random variables are defined
by the Gauss-Markov autocorrelation function RX(τ). Find the joint probability density
function fX(x) of the multi-variate X= [X(0) X(0.5) X(1.0)] T, mean mX= [0 0 0]T and X(t) has discrete values of +A and –A with equal probability of ½. Hence the mean of X(t)
realizations x= [x1 x2 x3]T. is mX = ½A+½(-A) = 0 and its variance is σX2 = ½(A-0)2+½(-A-0)2 = A2.
When |τ|=0, X(t) and X(t+τ) are fully correlated and you can evaluate RX (τ) simply as |τ|
RX (0) = E[X(t)X(t+0)] = E[X2] = σX2 = A2
And when |τ|=T, X(t) and X(t+τ) are uncorrelated and hence
RX (T) = E[X(t)X(t+T)] = E[X(t)] E[X(t+T)] = mX2 = 0. +A

For 0<|τ|<T, over the fully correlated portion of the pulse period T-|τ|, its contribution to
E[X(t)X(t+τ)] is proportional to RX (0)= A2 and over the uncorrelated portion |τ|, it is
proportional to of RX (T)= 0. Using time averaging over large number of periods T:
NT
1
RX (τ ) = E { X (t ) X (t + τ )} = Lt X ( t ) X ( t + τ ) dt
N →∞ NT ³
0

(T − τ ) . A 2 τ
= + .0
T T 0
A2 ( T − τ
=
) +0
T
T
§ τ ·
= A2 ¨ 1 − ¸
© T ¹

X(t) has discrete values of +A and 0 with equal probability of ½. Hence the mean of X(t)
Problem 12 is mX = ½A+½(0) = ½A (non-zero) and its variance is σX2 = ½(A-A/2)2+½(0-A/2)2 =
A2/4.
Find the time averaged autocorrelation function RX(τ) of a random binary sequence X(t)
with equally probable and discrete output values of +A, 0. Evaluate also the mean and When |τ|=0, X(t) and X(t+τ) are fully correlated and you can evaluate RX (τ) simply as:
variance of X(t). Hence find the spectral density function SX(f) of the random process. RX (0) = E[X(t)X(t+0)] = E[X2] = σX2 + mX2 = A2/2. Directly we can also find E[X2]=
The pulse interval is T sec. Assume that the random variables giving the outputs in ½(A2)+½(02)=A2/2.
different pulse intervals are independent. And when |τ|=T or greater, X(t) and X(t+τ) are uncorrelated and hence
When τ→∞, what is RX(τ)? Explain why it is equal to the square of the mean of X(t). RX (T) = E[X(t)X(t+T)] = E[X(t)] E[X(t+T)] = mX2 = A2/4.

For 0<|τ|<T, over the fully correlated portion of the pulse period T-|τ|, its contribution to
E[X(t)X(t+τ)] is proportional to RX (0)= A2/2 and over the uncorrelated portion |τ|, it is
proportional to of RX (T)= A2/4. Using time averaging over large number of periods T:
NT differential equation for X(t) describing the linear system driven by white noise input u(t).
1
RX (τ ) = E { X (t ) X (t + τ )} = Lt ³ X ( t ) X ( t + τ ) dt The autocorrelation function of the Gauss-Markov process is taken as RX(τ)= σ2e-β|τ|.
N →∞ NT
0
2
(T − τ ) . A τ A2 N0
= + . SN ( s ) = =1
T 2 T 4 2
A2 τ A2 S X ( s ) = G ( s ) G ( −s ) SN ( s ) = G ( s ) G ( −s )
= − .
2 T 4 −β τ 2σ 2 β
S X ( s ) = ℑ ¬ª RX (τ ) ¼º = ℑ ª¬σ 2 e º¼ = 2
A2 A2 § τ · −s + β 2
= + ¨1 − ¸ τ <T
4 4 © T¹ σ 2β σ 2β
= . = G ( s ) G ( −s )
A2 β +s β −s
RX (τ ) = E { X ( t )}.E { X ( t + τ )} = mX2 = τ ≥T
4 σ 2β
→ G (s) =
s+β
Writing R X (τ ) as:
The ODE describing the linear system with transfer function G(s) and
­ A2 A2 ª τ º
° + «1 − » τ < T driven by white noise input u(t) is:
° 4 4 ¬ T¼
R X (τ ) = ® dX ( t )
° A2 + β X ( t ) = 2σ 2 β u ( t )
τ ≥T dt
°̄ 4 Note: This is a linear first order ODE for which analytical solution can be found:
= RX 1 (τ ) + RX 2 (τ ) Homogeneous solution is
A2 X (t ) = X (t0 )e− β (t −t0 )
RX 1 (τ ) =
4 Complete solution to inhomogeneous equation is
A2 ª τ º t
RX 2 (τ ) = «1 − » − β ( t −t0 ) − β ( t −τ )
4 ¬ T¼ X (t ) = X (t0 )e + ³e 2σ 2 β u (τ ) dτ
t0
A2 Setting t = tk +1 , t0 = tk the above solution can be written in the form
Obviously when τ → ∞, RX (τ ) = = mX2
4
X ( tk +1 ) = Φ ( tk +1 , tk ) X ( tk ) + w ( tk +1 , tk )
X ( t ) , X ( t + τ ) are uncorrelated for any X ( t ) when τ → ∞, hence:
or more concisely as the difference equation
RX (τ ) = E { X ( t )}.E { X ( t + τ )} = mX .mX = mX2
X k +1 = Φ k X k + wk
A2 A2T This is the Discrete-Time model of the Gauss-Markov process in which
S X ( f ) = ℑ ª¬ R X (τ ) º¼ = δ ( f )+ sinc 2 ( fT )
4 4 Φ k is the state transition function (or matrix) and w k is the process noise term.

Problem 13

Consider a linear system driven by white noise u(t) whose spectral density SN(f) is
constant N0/2=1. The transfer function of the linear system is G(s) where s=j2πf. If the
output X(t) of the linear system has the spectral density function SX(f) corresponding to a
Gauss-Markov process, what should the transfer function G(s) be? Write down the
Problem 14

The WGS-84 World Geodetic System serves as a user interface for a GPS user who
needs to input the initial user location in the WGS coordinates (latitude φ, longitude λ on r2 z2 2r 2z
Earth’s ellipsoidal surface and height h above the surface datum) which will be converted f ( r, z ) = + − 1 → Normal NS = ∇f ( rn , zn ) = 2n iˆ + 2n ˆj
a2 b2 a b
to the ECEF Cartesian coordinate system for internal processing (position determination
from pseudorange measurements) and subsequently, the output/display of the processed NS is parallel to (ru - rn)i + (zu – zn)j and hence their vector components are proportional.
results (determined position in ECEF coordinates) back to the user in WGS coordinates.
In this exercise we convert the user location (longitude λ=30°, latitude φ=60°, h=10km) ru − rn zu − z n
to ECEF coordinates (xu, yu, zu) using explicit formulae and then back to WGS 2 rn
= 2 zn
→ a 2 ( ru − rn ) zn = b 2 ( zu − zn ) rn
coordinates using a numerical method (root finding of nonlinear equation). The answer a2 b2
should of course be longitude λ=30°, latitude φ=60°, h=10km. b2 2 2
where zn2 = ( a − rn ) = (1 − e2 )( a 2 − rn2 )
Using a=6378.137km, b= 6356.7523142km, 1-e2=(b/a)2=0.9966471892, e=0.081819194. a2
The explicit formulae to convert the WGS coordinates (λ, φ, h) of the GPS user position
S to the ECEF coordinates (xu, yu, zu) are
To solve this non-linear equation for rn (and hence also zn) in terms of the known ru and
ª º zu, we first assume a circular Earth (b = a). The initial estimate for rn is given by:
« a cos λ
+ h cos λ cos φ »
« 1 + 1 − e 2 tan 2 φ »
« ( ) » 2773.104092 zu
ª xu º « » ª º a 2 ( ru − rn ) zn = a 2 ( zu − zn ) rn → ru zn = zu rn or zn = rn
« » a sin λ ru
2 2
u = « yu » = « + h sin λ cos φ » = ««1601.052337 »» km
« » and hence solving for the approximate rn using f (rn , zn ) = 0 :
«¬ zu »¼ « 1 + (1 − e ) tan φ » «¬5509.137288 »¼
2
b2 2 2 z2
« » zn2 = 2 (
a − rn ) = (1 − e2 )( a 2 − rn2 ) = u2 rn2
a
a (1 − e ) sin φ ru
« + h sin φ »
1 − e 2 sin 2 φ 2 2 2
u
«¬ »¼
rn2 =
(1 − e ) r a
2 2 2
u
z + (1 − e ) r u
To convert from the ECEF coordinates of S back to the WGS coordinates, the longitude
→ rn = 3197.079568 km z n = 5500.491578 km
is simply given by the formula :
­ §y ·
° tan −1 ¨ u ¸ xu ≥ 0
° © xu ¹
° From a 2 ( ru − rn ) zn = b 2 ( zu − zn ) rn or ( ru − rn ) zn = (1 − e2 ) ( zu − zn ) rn
° §y ·
λ = ® 180 + tan −1 ¨ u ¸ xu < 0 yu ≥ 0 = 29.99999910466941 we form the residue function :
° © xu ¹
° § · g ( rn ) = ( ru − rn ) zn − (1 − e 2 ) ( zu − zn ) rn = ru zn − (1 − e 2 ) zu rn − e 2 rn zn
°−180 + tan −1 ¨ yu ¸ xu < 0 yu < 0
2 2
°¯ © xu ¹ where zn = (1 − e )( a − rn2 )

We define ru = xu2 + yu2 and consider the 2D vertical ru -z u plane through the Earth's axis . g(rn) can be used to find (numerically) the improved solution to rn (and hence zn) since
The point N where the normal dropped from S(ru, zu) intersects the Earth’s surface can be the more precise root rn should make the residue function g(rn) tend to zero. This is
found numerically. In this 2D plane, let the coordinates of N be (rn, zn). The cross section essentially a root finding problem.
of the Earth is an ellipse defined by f(r,z) = 0, where
g(rn) =1.469743438065052e-009 rn =3.197104759403027e+003 Matlab codes:
The updated fzero function in Matlab 7 is a little different from the
older versions, so some change in the code is necessary if you use
g(rn) = -9.720679372549057e-009 rn =3.197104759403028e+003 older Matlab versions.

Most precise values function r=gfunction(rn,ru,zu)


rn = 3197.1047594 km boa=6356.7523142/6378.137;
a=6378.137;
zn = 5500.4770343 km k1=a*a;
k2=boa*boa;%1-e*e
The latitude φ and the height h are then calculated from the most precisely determined e=sqrt(1-k2);
values of rn and zn using: zn=sqrt(k2*(k1-rn*rn)); % point lies on ellipse
r=ru*zn-k2*zu*rn-e*e*rn*zn % residue error - zero when exactly
2 zn
satisfied
b2 a 2 zn 1 zn rn
tan φ = 2 rn
= =
a2
b 2 rn 1 − e2 rn
→ φ = 59.99999820933881 format long;
2 2
clear
2
h = ( ru − rn ) + ( zu − zn ) % enter the EGW-84 geodetic coordinates, convert to cartesian (ECEF) by
formulae
→ h = 9.99999999999931 km % and convert back to geodetic coordinates using numerical method
phi1=60;
lambda1=30;
h1=10;
d2r=0.017453292;
r2d=57.29577951;
tanphi1=tan(phi1*d2r);
sinphi1=sin(phi1*d2r);
cosphi1=cos(phi1*d2r);
coslambda1=cos(lambda1*d2r);
sinlambda1=sin(lambda1*d2r);
boa=6356.7523142/6378.137;
a=6378.137;
k1=a*a;
k2=boa*boa;%1-e*e
e=sqrt(1-k2);
deno1=sqrt(1+k2*tanphi1*tanphi1);
xu=a*coslambda1/deno1 + h1*coslambda1*cosphi1
yu=a*sinlambda1/deno1 + h1*sinlambda1*cosphi1
ru=sqrt(xu*xu+yu*yu)
zu=a*k2*sinphi1/sqrt(1-e*e*sinphi1*sinphi1) + h1*sinphi1
rn0=sqrt(k2*ru*ru*k1/(zu*zu+k2*ru*ru)) %initial estimate for rn
zn0=zu*rn0/ru %initial estimate for rn
rn=fzero(@(rn)gfunction(rn,ru,zu), rn0) %improved value for rn
zn=sqrt(k2*(k1-rn*rn)) %improved value for zn
lambda=r2d*atan2(yu,xu)
phi=r2d*atan(zn/(k2*rn))
h=sqrt((ru-rn)*(ru-rn)+(zu-zn)*(zu-zn))
Problem 15 In the perifocal PQW coordinate system, the P axis is taken along the line of the
Argument of Perigee. The true anomaly νk is then measured from the P axis. Hence the
The GPS satellite SV15 has the following ephemeris data at the reference GPS time of t0e PQW coordinates for the satellite orbit position estimate are given by:
= 151200 sec during a certain GPS week: a (1 − e2 ) cosν k
Eccentricity e = 0.006778693292 rPk = = a ( cos Ek − e ) = 1.265176 × 107 m
Inclination i = 0.9721164968 rad 1 + e cos vk
Mean anomaly M0 (at t= t0e) = -0.8856059028 rad a (1 − e2 ) sinν k
Argument of Perigee ω = 1.738558535 rad rQk = = a 1 − e 2 sin Ek = −2.325386 × 107 m
1 + e cos vk
Longitude of the Right Ascension of the Ascending Node Ω0 (at GPS time weekly epoch
i.e., t= 0) = -2.8654714 rad
Square root of major axis of orbit = 5153.618444 √km. To find the satellite position in the ECEF coordinates, evaluate the transformation matrix
The ECEF coordinate system is used. The Earth’s rotational speed is between the PQW and the ECEF coordinates:
Ω
 = 7.2921151467 × 10 −5 rad / s and the gravitational constant is GM=3.986004415x105 cos Ω k = 0.327374
e
km3/s2. sin Ω k = −0.944895
Determine the orbit position at GPS time t= tk = 150000 sec during the GPS week in (a) cos ω = −0.166976
the PQW orbit plane coordinate system (P axis is taken along the line of the Argument of sin ω = 0.985961
Perigee in the orbit plane) and (b) the ECEF coordinate system.
cos i = 0.563552
sin i = 0.826080
Major axis a= 26559783.07 m R11 = cos Ω k cos ω − sin Ω k sin ω cos i = 0.470358
2π GM ×109 R12 = − cos Ω k sin ω − sin Ω k cos ω cos i = −0.411692
n= = = 1.458586 ×10−4 rad / s (note: GM in km 3 / s 2 , a in meter)
T a3 R21 = sin Ω k cos ω + cos Ω k sin ω cos i = 0.339677
At t= tk = 150000 sec determine the mean anomaly and the longitude of the RAAN
R22 = − sin Ω k sin ω + cos Ω k cos ω cos i = 0.900824
M k = M 0 + n ( tk − t0 e ) = −1.060636 rad
R31 = sin ω sin i = 0.814483
Ω = Ω −Ω
k 0
 t = −13.803644 rad
e k
R32 = cos ω sin i = −0.137936
Find the approximate true anomaly and eccentric anomaly from the mean anomaly.
In ECEF coordinates, the satellite orbit position is given by:
ν k 0  M k + 2e sin M k + 54 e 2 sin 2 M k = −1.0750647 rad
ª ri º ª R11 R12 º
ν « » ª rP º
Ek 0 = 2 tan −1 ª 11+−ee tan 2k 0 º = −1.0691115 rad
( )
¬ ¼ « rj » = [ R ] « rQ » [ R ] = «« R21 R22 »»
Improve the above solution numerically using Newton’s method «¬ rk »¼ ¬ ¼ «¬ R31 R32 »¼
E − e sin Ek 0 − M k
ΔE = − k 0 ri = R11rP + R12 rQ = 1.552430 × 107 m
1 − e cos Ek 0
rj = R21rP + R22 rQ = −1.665011× 107 m
E − e sin Ek 0 − M k
Ek 1 = Ek 0 + ΔE = Ek 0 − k 0 = −1.0665713461 rad rk = R31rP + R32 rQ = 1.351219 × 107 m
1 − e cos Ek 0
Accept this as sufficiently accurate solution (one iteration).
Ek= Ek1 = -1.0665713461 rad.
The improved value of the true anomaly νk can be determined from
E
vk = 2 tan −1 ª 11+−ee tan 2k º
¬ ¼
but it is not necessary in this problem. If you change the P axis from the line of the
Argument of the Perigee, you will need the true anomaly value.
Matlab codes Problem 16

format long; The GPS satellite SV15 has the known ECEF position given by xi = +15524471.175 m , yi
eccen=0.006778693292; = -16649826.222 m and zi = +13512272.387 m. The initial estimate of the user position
incli=0.9721164968;
Mo=-0.8856059028; in ECEF coordinate system is given by xu = -730000 m, yu = -5440000 m and zu =
wPerigee=1.738558535; +3230000 m. The receiver clock bias is estimated to be equivalent to +10000 m in range
Omegao=-2.8654714; bias. The pseudorange measurement reading, obtained from the C/A code epochs by the
a=26559783.07; GPS receiver, is 89491.971 m. Find the incremental pseudorange Δρi for this
toe=151200;
GPStime=150000; measurement.
GM=3.986004415e+005;
OmegaEdot=7.2921151467e-005;
nrad=sqrt((GM*1.0e+009)/(a*a*a)); Given the GPS satellite SV15 position (ECEF) as xi = +15524471.175 m , yi = -
Mk=Mo+nrad*(GPStime-toe);
16649826.222 m and zi = +13512272.387 m and the initial estimate of the user position
Omegak=Omegao-OmegaEdot*GPStime;
vk=Mk+2.0*eccen*Mk+5.0*eccen*eccen*sin(2.0*Mk)/4.0; (ECEF) as xu = -730000 m, yu = -5440000 m and zu = +3230000 m, the estimated
E=2.0*atan(sqrt((1.0-eccen)/(1.0+eccen))*tan(vk/2.0)); physical range is
error=1.0; 2 2 2
while abs(error)>1.0e-009 Ri = ( xi − xu ) + ( yi − yu ) + ( zi − zu ) = 22261921.80m
E1=E-(E-eccen*sin(E)-Mk)/(1.0-eccen*cos(E)); The estimated TOA is Ri/299792.458 = 74.257778 ms and hence the fractional
error=E1-E;
E=E1; millisecond is 0.257778 ms.
end The partial pseudorange is sum of fractional milli-second range estimate plus the clock
rP=a*(cos(E)-eccen) bias equivalent range. The receiver clock bias is estimated to be equivalent to +10000 m
rQ=a*sqrt(1.0-eccen*eccen)*sin(E) in range bias. Hence the estimated partial pseudorange is
cosOmegak=cos(Omegak);
sinOmegak=sin(Omegak); ρˆi = 0.257778x299792.458 + 10000 = 77279.900 + 10000 = 87279.900 m
coswPerigee=cos(wPerigee); The measured partial pseudo range is ρi = 89491.971m
sinwPerigee=sin(wPerigee);
cosincli=cos(incli); Hence the incremental pseudorange of the user from SV15 is
sinincli=sin(incli); Δρi = ρi − ρˆ i = 89491.971 − 87279.900
R11=cosOmegak*coswPerigee-sinOmegak*sinwPerigee*cosincli;
R12=-cosOmegak*sinwPerigee-sinOmegak*coswPerigee*cosincli; = 2212.071m
R21=sinOmegak*coswPerigee+cosOmegak*sinwPerigee*cosincli;
R22=-sinOmegak*sinwPerigee+cosOmegak*coswPerigee*cosincli;
R31=sinwPerigee*sinincli;
R32=coswPerigee*sinincli;
rx=R11*rP+R12*rQ
ry=R21*rP+R22*rQ
rz=R31*rP+R32*rQ