Professional Documents
Culture Documents
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
Problem Sheet 1
Examples of Random Processes
1. Give examples of situations in which time series can be used for explanation, description,
forecasting and control.
4. Plot a one-dimensional probability density function (PDF) at any discrete time moment ti=iΔt (i is
a positive integer and Δt is a time interval) for
(c) a process of throwing at equal time intervals Δt a biased coin, in which probability of
heads is 0.4 and of tails 0.6.
5. Find mean values and variances at any time moment ti of the processes (a) and (b) from Q. 4.
Illustrate their positions on the plots for PDFs by analogy with examples in lecture notes.
6. Find mean value and variance of a continuous random variable α whose PDF is
! π$
pα (α ) = C cos # a + & with α ∈ #%− π ,0&( and C being some constant. Note, that first of all you
" 4% $ 4 '
need to find the value of constant C .
1
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
Problem Sheet 2
Statistical Characteristics of a Random Process, Stationarity
3. X(t)=Acos(ωt+ϕ)=f(ϕ,t), where A and ω are constants, ϕ is a random variable that with equal
probability takes two values: -π/4 and π/4.
5. Let X(t)=cos(ωt+ϕ), where ω and ϕ are random variables with joint probability density function
p2ωϕ = 1 8π , ω ∈ [8;12], ϕ ∈ [−π; π]
For processes 1 – 5
Find mean value and variance of random process X(t).
Find autocorrelation function of random process X(t).
Is the process wide sense stationary?
2
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
A2 A2
1) X(t) ≡ 0, σ 2X (t) = , K XX (t, τ) = ΨXX (t, τ) = cosωτ
2 2
1 1
2) X(t) ≡ 0, σ 2X (t) = cos2 ωt, K XX (t, τ) = ΨXX (t, τ) = (cos(2ωt + ωτ) + cosωτ)
€ 3 6
A A2 2
3) X(t) = cosωt, σ 2X (t) = sin ωt,
€ 2 2
A2 A2 A2
K XX (t, τ) = cos(ωτ), ΨXX (t, τ) = cos(ωτ) − cos(2ωt + ωτ)
2 4 4
2 2A 2
& A2 4 A2 ) & A2 4 A2 )
4) X(t) = cos(ωt + π /4), σ (t) = ( − 2 + − sin(2ωt)( − 2 +,
X
€ π ' 2 π * 'π π *
A2 A2 & A2 4 A2 ) & A2 4 A2 )
K XX (t, τ) = cos(ωτ) − sin(2ωt + ωτ), ΨXX (t, τ) = ( − 2 + cos(ωτ) − ( − 2 + sin(2ωt + ωτ)
2 π ' 2 π * 'π π *
1 1
5) X(t) ≡ 0, σ 2X (t) = , K XX (t, τ) = ΨXX (t, τ) = (sin(12τ) − sin(8τ))
€ 2 8τ
3
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
Problem Sheet 3
Properties of Autocorrelation and Covariance
1.
2. €
( )(
ΨXX (t,t + τ ) = X(t) − X(t) X(t + τ ) − X(t + τ ) )
= X(t)X(t + τ ) − X(t) X(t + τ )
3.
€
( )(
ΨXY (t,t + τ ) = X(t) − X(t) Y (t + τ ) − Y (t + τ ) )
= X(t)Y (t + τ ) − X(t) Y (t + τ )
4.
€ 2
σ X2 (t) = X 2 ( t ) − X ( t )
4
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
Problem Sheet 4
Ergodicity of a Random Process
3. X(t)=Acos(ωt+ϕ)=f(ϕ,t), where A and ω are constants, ϕ is a random variable that with equal
probability takes two values: -π/4 and π/4.
5. Let X(t)=cos(ωt+ϕ), where ω and ϕ are random variables with joint probability density function
p2ωϕ = 1 8π , ω ∈ [8;12], ϕ ∈ [−π; π]
For processes 1 – 5
• Find if the random process X(t) is ergodic with respect to mean value.
• Find if the random process X(t) is ergodic with respect to variance and covariance.
5
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
A2
1) x(t) = X(t) ≡ 0, ( x(t) − x )( x(t + τ) − x ) = ΨXX (t, τ) = cosωτ
2
Yes, this process is ergodic with respect to mean, variance and covariance.
€
ξ *2
2) x(t) = X(t) ≡ 0, ( x(t) − x )( x(t + τ) − x ) = cosωτ ≠ ΨXX (t, τ)
2
This process is ergodic with respect to mean, but not variance or covariance (compare with results
for PS. 2 Q. 2).
€
A2
3 − 4) x(t) ≡ 0, ( x(t) − x )( x(t + τ) − x ) = cosωτ
2
The mean values and covariances of these processes obtained by averaging over the ensemble of
realizations (see results for PS. 2, Q. 3-4) are different from those obtained by averaging along the
single realization. Thus, the processes are not ergodic either with respect to mean, variance or
€ covariance.
1
5) x(t) = X(t) ≡ 0, ( x(t) − x )( x(t + τ) − x ) = cosω * τ ≠ ΨXX (t, τ)
2
This process is ergodic with respect to mean, but not covariance (compare with results for PS. 2 Q.
5).
€
6
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
Problem Sheet 5
Statistical Characteristics of a Random Process, Stationarity – More Problems
1. Consider random process X(t)=ξ(t)cos(ωt+ϕ), where ω is constant, ξ(t) is random process that is
1st order stationary and does not depend on ϕ. ϕ is random variable. Find the conditions that ϕ
should satisfy to make random process X(t) wide sense stationary. Hint: consider autocorrelation
function of X(t)
3. In problem 2 assume that the only known information about the process X(t) is that it is wide
sense stationary and has autocorrelation function CXX(τ). Find mean value of Y(t). Find
autocorrelation function of Y(t) in terms of CXX(τ). Is Y(t) wide sense stationary?
4. Let X(t)=Acos(ωt)+Bsin (ωt), where ω is constant, A and B are uncorrelated random variables
with zero mean (they may have different distributions). Is X(t) wide sense stationary (WSS)? If the
process is not WSS, what additional conditions should be imposed on A and B to make it WSS?
7
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
1
2. KYY (t, τ) = exp(−a | τ |)cosωτ .
2
€
1
3. Y (t) = 0 KYY (t, τ) = CXX (τ)cosωτ . The process is WSS.
2
€ % A2 B2 ( % A2 B2 (
4. X(t) = 0, K XX (t, τ) = cos(2ωt + ωτ)'' − ** + cosωτ'' + ** . The process is not WSS. To
& 2 2) & 2 2)
€ make the process WSS, variances of A and B should be equal.
€ 1 1 1 1
5. X(t) ≡ 0, σ 2X (t) = t 2 + σ 2ξ , K XX (t, τ) = ΨXX (t, τ) = t(t + τ) + σ 2ξ exp(−λ | τ |)cosωτ . The
3 2 3 2
process is not WSS.
€ €
8
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
Problem Sheet 6
Properties of the Fourier Transform (FT)
1. Prove the time-scaling property of the Fourier Transform (FT), i.e. prove the following. Let
F(ω) be the FT of the function f(t). Let the argument t of f(t) be multiplied by some real
constant a, i.e. let the time be scaled. Then the FT of the function f(at) is equal to 1/|a|
F(ω/a).
Find Fourier amplitude and phase spectra for the realizations x(t) in problems 3-7. In all problems
sketch the solution. It might be convenient to sketch x(t) first.
4. x(t)=Asin(ω t)
0
5. Sum of two cosines with different frequencies x(t) = A1 cos (ω1 t) + A2 cos (ω 2 t) .
6. Sum of a cosine and a sine with different frequencies x(t) = A1 cos (ω1 t) + A2 sin (ω 2 t) .
7. Cosine with periodically modulated amplitude:
9
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
1.
2.
3.
ψ
F(ω) = Aπ[δ(ω − ω 0 ) + δ(ω + ω 0 )], ϕ(ω) = ω
ω0
4. F(ω) = Aπ[δ(ω − ω 0 ) + δ(ω + ω 0 )], ϕ(ω) = − π ω
2 ω0
€
5. F(ω) = A1 π [δ(ω − ω1 ) + δ(ω + ω1 )] + A2 π [δ(ω − ω 2 ) + δ(ω + ω 2 )], ϕ(ω) ≡ 0
€ 6.
m
F(ω) = π [δ(ω − ω 2 ) + δ(ω + ω 2 )] + π [δ(ω − (ω1 + ω 2 )) + δ(ω + (ω1 + ω 2 ))] +
2
m
+ π[δ(ω − (ω1 − ω 2 )) + δ(ω + (ω1 − ω 2 ))], ϕ(ω) ≡ 0
2
In the right-hand side of the figure the amplitude of FT is shown for the parameters taken as in the
left-hand side.
€
' ∞ * ∞
8. ℑ) ∑ δ(t − nT), = ∑ e− jωnT
(n=−∞ + n=−∞
10
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
Problem Sheet 7
Fourier Transform and Wiener-Khintchine Theorem
#1, 0 ≤ t ≤ T,
p(t) = $
%0, t < 0 or t > T.
a) Calculate the Fourier Transform (FT), F(ω ) , of p(t). Find and sketch the amplitude
Fourier spectrum, F(ω ) , of p(t) for T=10.
€
b) Consider x(t) = p(t)cosω 0 t . Calculate the FT, F1 (ω ) , of x(t) directly using the
definition. Compare your result with the one obtained by application of frequency-
€
shifting property of the FT.
€ 2
c) Calculate
€ the squared amplitude Fourier spectrum,
€ F1 (ω ) , of x(t). Compare your result
with the squared amplitude Fourier spectrum of y(t) = p1 (t)cosω 0 t , where
$ T T
&1, €− 2 ≤ t ≤ 2 ,
p1 (t) = %€
&0, T
t > ,
' 2
that was obtained in lecture.
2
d) Sketch carefully the F1 (ω ) for T=10, ω0=5.
€
2. a) Calculate the Fourier Transform (FT), F2 (ω ) , of z(t) = p(t)sin ω 0 t , where p(t) is the same
as in Q. 1. Obtain the same result by two methods:
€
• from the direct definition of FT
€ €
• using the frequency-shifting property of FT
2
b) Calculate the squared amplitude Fourier spectrum, F2 (ω ) , of z(t) from a). Compare your
result with the squared amplitude Fourier spectra of x(t) and y(t) of Q. 1.
3. Which of the following functions could be a valid power spectral density, given that a, b, c,
€ in Hz,
and f0 are real positive numbers? f is frequency
af a a + cf a a a
a) 2 , b) 2, c) , d) , e) , f ) δ( f ) + ,
b+ f b+ f b+ f 2 2
f −b a + f + f2
2
b+ f 2
a cos(3 f )
g) , h) aδ ( f − f 0 ) − aδ ( f + f 0 ) , i) .
b + jf 2 1+ f 2
€ € € € € €
4. Consider a random process that is 1st order stationary with zero mean and the power spectral
density
€
€ €
$ W 0 = const, | ω |≤ ω 0
S(ω ) = % ,
& 0, | ω |> ω 0
6. Consider stationary random process X(t) with zero mean and autocorrelation function
ΨXX (τ ) = σ X2 e−α |τ | , α=const>0.
Find power spectral density of this process and sketch it. Find the power of the process.
[Hint: you can use the following identity:
ax e ax
∫e cos(bx)dx = 2
a + b2
(acos(bx) + bsin(bx)) + C . ]
8. Consider a random process X(t) that is wide sense stationary with zero mean and with the
€
power spectral density SX(ω)
W0
%W (1 − ω ), ω ≤ 1,
SX (ω ) = & 0
' 0, ω > 1.
Find autocorrelation function and power of this process. Sketch autocorrelation function.
€ 9. Consider a random process X(t) that is wide sense stationary with zero mean and with the
autocorrelation function
+ $ τ'
- A 1 − , − T ≤ τ ≤ T,
K XX (τ ) = , 0 &% T )(
-. 0, elsewhere.
Find power spectral density and power of this process. Sketch power spectral density.
€
12
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
1.
2 4 #
2 T
& 2 #T &
a) F(ω ) = 2 sin % ω (, F(ω ) = sin% ω (
ω $2 ' ω $2 '
€ $ (ω − ω 0 )T ' $ (ω + ω 0 )T '
sin 2 & ) sin 2 & )
2 % 2 ( % 2 ( cos(ω 0T )
c) F1 (ω ) = 2 + 2 +
ω 2 − ω 02
[ ]
cos(ω 0T ) − cos(ωT )
(ω − ω 0 ) (ω + ω 0 )
d)
€
2
The graph for F1 (ω ) is symmetric with respect to zero.
13
€
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
2. a)
1 * $ sin((ω + ω 0 )T ) sin((ω − ω 0 )T ) ' $ cos((ω − ω 0 )T ) −1 cos((ω + ω 0 )T ) −1' -
F2 (ω ) = , j && − )) + && − )) /
2 ,+ % ω +ω0 ω −ω0 ( % ω −ω0 ω +ω0 ( /.
$ (ω − ω 0 )T ' $ (ω + ω 0 )T '
sin 2 & ) sin 2 & )
2 % 2 ( % 2 ( cos(ω 0T)
€ b) F2 (ω ) = 2 + 2 − × [cos(ω 0T) − cos(ωT)]
(ω − ω 0 ) (ω + ω 0 ) ω 2 − ω 02
3.
€ (a) no, since it is not even.
(b) yes since it is real, even and positive for all f.
(c) no, since it is not even.
(d) no, since it can be negative at f 2<b
(e) no, since it is not even
(f) yes, since it is real, even and positive for all f.
(g) no, since it is not real
(h) no, since it is negative at f=-f0
(i) no, since it is negative when cos(3f) is
negative.
4.
W0
K XX (τ ) = sin(ω 0τ ),
πτ
W 0ω 0 sin(ω 0τ ) W 0ω 0
P = K XX (0) = lim = .
τ→0 π ω 0τ π
A 2π A 02 π A 02 π
5. S(ω ) = €0 (δ (ω − ω 0 ) + δ (ω + ω 0 )) δ(ω + ω0 ) δ(ω − ω0 )
2 2 2
6.
2ασ X2
S(ω ) =
α2 +ω2
P = σ X2
€ 14
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
% 1 1 (
7. S(ω ) = 2' 2 + 2*
& 4 + (4 π + ω ) 4 + (4 π − ω ) )
2
€ W 0 $ sin(τ /2) '
8. K XX (τ ) = & )
2π % (τ /2) (
€ The “width” of S(ω) is roughly proportional to the difference between the two values of ω closest to
0, at which S(ω)=0:
ω1,2T 2π 4π
= ±π ⇒ ω1,2 = ± ; ω1 − ω 2 =
2 T T
Thus, the wider the ACF (i.e. the larger the T), the narrower the power spectral density is.
€
15
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
Problem Sheet 8
Nyquist Theorem, Aliasing
$ 27 27
& 2 + , f ≤ 10,
SX ( f ) = % 9 + ( f − 3) 9 + ( f + 3) 2
&' 0, f > 10.
A realization of this process is sampled with sampling frequency 12 Hz. Sketch the best power
spectral density that could possibly be estimated numerically from this realization if observation
€ infinity.
time could tend to
2. Find the smallest sampling frequency necessary to discretize realizations of the random process
X(t) = A(t)cos(ω 0 t + ϕ) , where ϕ is a random variable uniformly distributed in [−π; π], and A(t) is
a stationary band-width limited random process whose power spectral density SA(f) has the
following form:
%W (1− ω ), ω ≤ 1,
€ SA (ω ) = & 0
' 0, ω > 1.
A(t) and ϕ are statistically independent. Sketch power spectral density SA(f) of the process A(t) and
power spectral density SX(f) of X(t), and on the latter plot show the position of the frequency
sought. €
$8(1− f ), f ≤ 1,
SA ( f ) = %
& 0, f > 1.
A(t) and ϕ are statistically independent. Sketch the power spectral density of the process A(t), SA(f).
Find the power spectral density of the process X(t), SX(f). Sketch SX(f) and on this plot show the
smallest allowed sampling frequency. Find the power of the process X(t).
€
4. Consider a random process X(t) = A(t) sin(8πt + ϕ) , where ϕ is a random variable uniformly
distributed in [0; π], A(t) is a wide-sense stationary random process whose autocorrelation function
K AA (τ ) is given by
+ $ τ'
€ -2&1 − ), τ ≤ 8,
K AA (τ ) = , % 8 (
-. 0, τ > 8.
€
Assume that realizations of this process are sampled with sampling step Δt=0.125 sec. Sketch the
best power spectral density that could possibly be estimated numerically from a realization of this
€ could tend to infinity.
process if observation time
16
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
5. Assume that a realization of a random process is x(t)=cos(6πt). Let this signal be recorded during
time interval [-T; T]. In order to reduce the leakage of power due to finite observation time, one
uses a triangular window G(t) before estimating Fourier Spectrum of the original infinitely long
signal. The equation of the window function and of the actual signal y(t) from which Fourier
Transform is calculated are:
$
&1 − t , t ≤ T,
G(t) = % T y(t) = x(t)G(t).
&' 0, t > T,
What would be the best estimate of Fourier Transform of x(t), if the sampling step could be taken
infinitely small?
€
6. For all processes considered in Q. 4-9 of Problem Sheet 7 estimate roughly what would be a
reasonable sampling frequency for their realizations. You are not required to specify the exact
numbers, but only to show a suitable cut-off frequency fc on the sketches of the power spectral
densities.
17
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
ω0 +1
2. f N = 2 ,
2π
1 1
3. SX ( f ) = SA ( f + 3) + SA ( f − 3), P = 4, f N = 8Hz
4 4
18
! RANDOM'PROCESSES'AND'TIME'SERIES'ANALYSIS.'PART'II:'RANDOM'PROCESSES' !
! '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''Problem'Sheets' !
1 1 1 − cos(16πf )
4. SX ( f ) = SA ( f + 4) + SA ( f − 4), SA ( f ) =
4 4 8π 2 f 2
19