EE2S31 Signal Processing – Stochastic Processes

Lecture 5: Stochastic processes – Ch. 13

Alle-Jan van der Veen

23 May 2022
Today: Ch. 13 Stochastic Processes
Stochastic: random
Process: sequence of variables where the ordering is of importance.

Random variable: Mapping from an outcome s in the sample space

to a real number x(s).
Stochastic process: Mapping from an outcome s in the sample
space to a function x(t, s), which depends on an ordering variable like
time or space: a random signal.
5. stochastic processes 2 / 46
Stochastic Processes
The stochastic process is denoted by X (t).
Sample function x(t, s1 ) is one particular realization (outcome s1 ) of
this process.

The ensemble of a stochastic process is the set of all possible time

functions that can result from an experiment.
5. stochastic processes 3 / 46
Problem 13.1.3
Consider the transmission of 3 bits in a BPSK system (binary phase
shift keying):
S ∈ {000, 001, 010, 011, 100, 101, 110, 111}

The ensemble consists of 8 possible sample functions:

5. stochastic processes 4 / 46
Example: speech

Experiment ‘s’: pronounce ‘ssss’

Sample function: one realization of the waveform x(t, s)

5. stochastic processes 5 / 46
Example: arrival times of data packets

5. stochastic processes 6 / 46
Example: binary bit pattern

A stochastic description can be used for image compression.

5. stochastic processes 7 / 46
Example: handwritten digits

A stochastic description (features!) is used for pattern recognition.

5. stochastic processes 8 / 46
Example: linear filtering of a stochastic process

How can we describe Y (t) when X (t) is a stochastic process?

Statistical descriptions of X (t): Statistical descriptions of Y (t):

R
mean µX mean µY = µX t h(t)dt
Autocorrelation function RX (τ ). RY (τ ) = h(τ ) ∗ h(−τ ) ∗ RX (τ ).

This will be the topic of the Supplement (next weeks)

5. stochastic processes 9 / 46
Description of a random process

How to describe a stochastic process at one time instance: e.g.,

X (t1 )?
How to describe a stochastic process at multiple time instances: e.g.,
[X (t1 ), X (t2 )]T ?

5. stochastic processes 10 / 46
Description of a random process
Similar as for RVs, we can use the PDF/PMF to describe stochastic
processes:

At any (fixed) time tk the stochastic process can be regarded as a

random variable:
X (tk ) ∼ fXtk (xtk )
This PDF may be different for each tk !

The joint behavior for multiple time instances t, i.e., t1 , · · · , tk is

given by the joint PDF:

[X (t1 ), · · · , X (tk ), · · · ]T ∼ fXt1 ,Xt2 ,··· ,Xtk ,··· (xt1 , xt2 , · · · , xtk , · · ·)

5. stochastic processes 11 / 46
Example: rectified sinusoid with random amplitude

Let X (t) = R| cos(ωt)| with

(
1 −r /10
10 e r ≥0
fR (r ) =
0 otherwise

Calculate fX (t) (x)!

Approach

First calculate the CDF FX (t) (x)

d
Then calculate the PDF fX (t) (x) = dx FX (t) (x)

5. stochastic processes 12 / 46
Example: rectified sinusoid with random amplitude

FX (t) (x) = P[X (t) ≤ x]

= P[R |cos(ωt)| ≤ x]
= P [R ≤ x/| cos(ωt)|] if cos(ωt) 6= 0
Z x/| cos(ωt)|
= fR (r ) dr
0
x
− 10| cos(ωt)|
= 1−e if x ≥ 0

If cos(ωt) 6= 0:
(
0 x <0
FX (t) (x) = x
− 10| cos(ωt)|
1−e x ≥ 0.

5. stochastic processes 13 / 46
Example: rectified sinusoid with random amplitude

If cos(ωt) 6= 0:
(
dFX (t) (x) 0 x <0
fX (t) (x) = = 1
x
− 10| cos(ωt)|
dx 10| cos(ωt)| e x ≥ 0.

If cos(ωt) = 0, then X (t) = 0 (constant) and fX (t) (x) = δ(x).

5. stochastic processes 14 / 46
Problem 13.2.1 (similar: 13.2.4)
Let W be an exponential random variable with PDF
(
e −w w ≥0
fW (w ) =
0 otherwise

Find the CDF FX (t) (x) of the time-delayed ramp process X (t) = t − W .

5. stochastic processes 15 / 46
Problem 13.2.1 (similar: 13.2.4)
Let W be an exponential random variable with PDF
(
e −w w ≥0
fW (w ) =
0 otherwise

Find the CDF FX (t) (x) of the time-delayed ramp process X (t) = t − W .

P[X (t) ≤ x] = P[t − W ≤ x] = P[W ≥ t − x] .

Since W ≥ 0, if x ≥ t then P[W ≥ t − x] = 1. When x < t,

Z ∞
P[W ≥ t − x] = fW (w )dw = e −(t−x) .
t−x

5. stochastic processes 15 / 46
Problem 13.2.1 (cont’d)
Combining the facts, we have
(
e −(t−x) x <t
FX (t) (x) = P[W ≥ t − x] =
1 x ≥t

We note that the CDF contains no discontinuities. Taking the

derivative of the CDF with respect to x, we obtain the PDF
(
e −(t−x) x <t
fX (t) (x) =
0 x ≥t

• Treat t as a constant
• Calculate CDF FX (t) (x)
• Then calculate PDF fX (t) (x)

5. stochastic processes 16 / 46
Description of a random process
Notice that

[X (t1 ), · · · , X (tk ), · · · ]T ∼ fXt1 ,Xt2 ,··· ,Xtk ,··· (xt1 , xt2 , · · · , xtk , · · ·)

resembles a vector random variable,

but can be of infinite dimensionality,

and ordering (in time) of the X (tk ) is essential.

Generally, the joint PDF is very difficult to acquire. Exceptions:

Independent identically distributed (iid) random sequence/process

Gaussian stochastic process
Poisson process (skipped), Brownian motion process (skipped)

5. stochastic processes 17 / 46
Independent identically distributed (iid) random sequences
For an iid random sequence

all X (tk ) are mutually independent random variables for all tk ,

all X (tk ) have the same PDF for all tk .

Let Xn denote a (time discrete) iid random sequence with sample vector
X = [Xn1 , · · · , Xnk ]T .

For a discrete valued Xn , the joint PMF is

k
Y
PX (x) = PX1 (x1 ) · · · PXk (xk ) = PX (xi )
i=1

For a continuous valued Xn , the joint PDF is

k
Y
fX (x) = fX1 (x1 ) · · · fXk (xk ) = fX (xi )
i=1
5. stochastic processes 18 / 46
Example: Bernoulli process (time discrete)
One realization of a Bernoulli process (e.g., a bit sequence)

PMF for one time instance k:


p xk = 1 (
p xk (1 − p)1−xk

xk = 0, 1
PXk (xk ) = 1 − p xk = 0 ⇔ PXk (xk ) =
 0 otherwise
0 otherwise


5. stochastic processes 19 / 46
Example: Bernoulli process (cont’d)
(
p xk (1 − p)1−xk xk = 0, 1
PXk (xk ) =
0 otherwise

For two time instances t1 and t2 we obtain (iid process!)

PX1 ,X2 (x1 , x2 ) = PX1 (x1 )PX2 (x2 ) = p x1 (1 − p)1−x1 p x2 (1 − p)1−x2

= p x1 +x2 (1 − p)2−x1 −x2

For k time instances we obtain

k
Y
PX (x) = p xi (1 − p)1−xi = p x1 +···+xk (1 − p)k−(x1 +···+xk )
i=1

5. stochastic processes 20 / 46
Gaussian Process
Gaussian processes occur quite often in nature (remember the central
limit theorem!)
The process X (t) is a Gaussian stochastic process (sequence) if and
only if X = [X (t1 ) · · · X (tk )]T is a Gaussian random vector for any
integer k > 0 and any set of time instances t1 , t2 , · · · , tk .

5. stochastic processes 21 / 46
Gaussian Process
Remember that for Gaussian random vectors X = [X1 , X2 , · · · , XN ]T we
have: h i
exp − 21 (x − E[X ])T CX−1 (x − E[X ])
fX (x) =
(2π)N/2 det(CX )1/2

For a Gaussian stochastic process X (t), the distribution of

X = [Xt1 , Xt2 , ..., Xtk ]T is thus also given by fX (x).

For each collection of sample times t1 , · · · , tk , we need to specify

The mean: µX = E[X ]; specify E[X (ti )] for all i;

The (auto)covariance: cov[X , X ] = CX ;
specify cov[X (ti ), X (tj )] for all i, j.

5. stochastic processes 22 / 46
Expected value of a stochastic process

Expected value of X (tk ) at time tk :

Z ∞
E[X (t)] = x fX (t) (x) dx
−∞

5. stochastic processes 23 / 46
Autocovariance of a stochastic process

cov[X (t1 ), X (t2 )] indicates how much the process is likely to change
from t1 to t2 .
Large covariance: sample function unlikely to change.
Zero covariance: sample function expected to change rapidly.

5. stochastic processes 24 / 46
Autocovariance and autocorrelation
The covariance of a stochastic process at two different time instances is
called “autocovariance”:

CX (t, τ ) = cov[X (t), X (t + τ )]

= E [(X (t) − E[X (t)]) (X (t + τ ) − E[X (t + τ )])]
= E[X (t)X (t + τ )] −E[X (t)]E[X (t + τ )]
| {z }
Similar to crosscorrelation E[XY ]

The autocorrelation is similarly defined as

RX (t, τ ) = E[X (t)X (t + τ )]

CX (t, τ ) = RX (t, τ ) − E [X (t)] E [X (t + τ )]

CX (t, 0) = E[X (t)2 ] − E[X (t)]2 = var[X (t)]

5. stochastic processes 25 / 46
The Autocorrelation function

Notice that exact formulation of the autocorrelation depends on

whether time and amplitudes are discrete or continuous:
ZZ
RX (t, τ ) = E[X (t)X (t + τ )] = x y fX (t),X (t+τ ) (x, y ) dxdy
XX
RX (t, τ ) = E[X (t)X (t + τ )] = xy P[X (t) = x, X (t + τ ) = y ]
x y

ZZ
RX [n, k] = E[Xn Xn+k ] = x y fXn ,Xn+k (x, y ) dxdy
XX
RX [n, k] = E[Xn Xn+k ] = xy P[Xn = x, Xn+k = y ]
x y

5. stochastic processes 26 / 46
Example: sinusoidal process

Random process: X (t) = A sin(ωt + Φ)

Amplitude A and phase Φ are independent random variables, where

A is uniformly distributed on [−1, +1]

Φ is uniformly distributed on [0, 2π]

5. stochastic processes 27 / 46
Example: sinusoidal process

RX (t, τ ) = E[X (t)X (t + τ )]

= E[A2 sin(ωt + Φ) sin(ω(t + τ ) + Φ)]
= E[A2 ] E[sin(ωt + Φ) sin(ω(t + τ ) + Φ)]
Z 1 2
a
= da E[sin(ωt + Φ) sin(ω(t + τ ) + Φ)]
−1 2
1
= E[sin(ωt + Φ) sin(ω(t + τ ) + Φ)]
3
1
= E[cos(ωτ ) − cos(2ωt + ωτ + 2Φ)]
6
1 1
= cos(ωτ ) − E[cos(2ωt + ωτ + 2Φ)]
6 6| {z }
=0
1
= cos(ωτ )
6

5. stochastic processes 28 / 46
Example: sinusoidal process
1
RX (t, τ ) = 6 cos(ωτ )

High correlation for ωτ = 0, ±2π, · · ·

Zero correlation for ωτ = ± 21 π, · · ·
Very negative correlation for ωτ = ±π, · · ·

5. stochastic processes 29 / 46
Uncorrelated and orthogonal processes

If all pairs X (t), X (t + τ ) are uncorrelated, i.e.,

(
var[X (t)] ∀ t and τ = 0
CX (t, τ ) =
0 ∀ t and τ =
6 0

then X (t) is called an uncorrelated process.

If all pairs X (t), X (t + τ ) are orthogonal, i.e.,

(
E[X 2 (t)] ∀ t and τ = 0
RX (t, τ ) =
0 ∀ t and τ =
6 0

then X (t) is called an orthogonal process

5. stochastic processes 30 / 46
Problem 13.7.2
For the time-delayed ramp process X (t) = t − W from Problem 13.2.1,
find for any t ≥ 0:
(a) The expected value µX (t),
(b) The autocovariance function CX (t, τ ).
Hint: E[W ] = 1 and E[W 2 ] = 2.

5. stochastic processes 31 / 46
Problem 13.7.2
For the time-delayed ramp process X (t) = t − W from Problem 13.2.1,
find for any t ≥ 0:
(a) The expected value µX (t),
(b) The autocovariance function CX (t, τ ).
Hint: E[W ] = 1 and E[W 2 ] = 2.

We know but don’t need:

(a) The mean is (
µX (t) = E[t − W ] = t − E[W ] = t − 1. e x−t x < t
fX (t) (x) =
0 o.w.
(b) The autocovariance is

CX (t, τ ) = E[X (t)X (t + τ )] − µX (t)µX (t + τ )

= E[(t − W )(t + τ − W )] − (t − 1)(t + τ − 1)
= t(t + τ ) − E[(2t + τ )W ] + E[W 2 ] − (t − 1)(t + τ − 1)
= −(2t + τ )E[W ] + 2 + 2t + τ − 1 = 1
5. stochastic processes 31 / 46
Stationary Process

A stochastic process is stationary if and only if every joint-PDF is shift

invariant:
fX (t1 ),X (t2 ),··· ,X (tk ) (x1 , x2 , · · · , xk )
= fX (t1 +∆t),X (t2 +∆t),··· ,X (tk +∆t) (x1 , x2 , · · · , xk )

Consequence I The marginal PDF’s are independent of t:

fX (t) (x) = fX (t+∆t) (x) = fX (x)

The marginal PDF’s are identical for all tk !

⇒ Expected value and variance are time independent.

5. stochastic processes 32 / 46
Problem 13.8.2
X = [X1 X2 ]T has expected value E[X ] = 0 and covariance matrix
 
2 1
CX =
1 1

Does there exist a stationary process X (t) and time instances t1 and t2
such that X is actually a pair of observations [X (t1 ) X (t2 )]T of the
process X (t)?

5. stochastic processes 33 / 46
Problem 13.8.2
X = [X1 X2 ]T has expected value E[X ] = 0 and covariance matrix
 
2 1
CX =
1 1

Does there exist a stationary process X (t) and time instances t1 and t2
such that X is actually a pair of observations [X (t1 ) X (t2 )]T of the
process X (t)?

The short answer is No. For the given process X (t),

var[X (t1 )] = C11 = 2, var[X (t2 )] = C22 = 1 .

However, stationarity of X (t) requires var[X (t1 )] = var[X (t2 )], which is
a contradiction.

5. stochastic processes 33 / 46
Stationary process

Consequence II The 2D joint-PDF is shift invariant

fX (t1 ),X (t2 ) (x1 , x2 ) = fX (t1 +∆t),X (t2 +∆t) (x1 , x2 )

= fX (0),X (t2 −t1 ) (x1 , x2 )

⇒ only the “distance” τ between t2 and t1 matters.

RX (t, τ ) = RX (τ )
CX (t, τ ) = CX (τ ) = RX (τ ) − E[X ]2

5. stochastic processes 34 / 46
Stationary process
Examples of stationary processes

iid process; e.g. Bernoulli process

Poisson process (random arrival process, ch. 13.4; we skip this)

Remember the PMF for the Bernoulli stochastic process:

k
Y
PX (x) = p xi (1 − p)1−xi = p x1 +···+xk (1 − p)k−(x1 +···+xk )
i=1

which does not depend on the actual time.

Non-stationary processes are difficult to model and to handle in practice.

5. stochastic processes 35 / 46
Problem 13.8.4
Let X (t) be a stationary continuous-time random process. By sampling
X (t) every ∆ seconds, we obtain the discrete-time random sequence
Yn = X (n∆). Is Yn a stationary sequence?

5. stochastic processes 36 / 46
Problem 13.8.4
Let X (t) be a stationary continuous-time random process. By sampling
X (t) every ∆ seconds, we obtain the discrete-time random sequence
Yn = X (n∆). Is Yn a stationary sequence?

Since Yni +k = X ((ni + k)∆) for a set of time samples n1 , · · · , nm

fYn1 +k ,··· ,Ynm +k (y1 , · · · , ym ) = fX ((n1 +k)∆),··· ,X ((nm +k)∆) (y1 , · · · , ym ) .

Since X (t) is a stationary process,

fX ((n1 +k)∆),··· ,X ((nm +k)∆) (y1 , · · · , ym ) = fX (n1 ∆),··· ,X (nm ∆) (y1 , · · · , ym ) .

Since X (ni ∆) = Yni , we see that

fYn1 +k ,··· ,Ynm +k (y1 , · · · , ym ) = fYn1 ,··· ,Ynm (y1 , · · · , ym )
Hence, Yn is a stationary sequence.
5. stochastic processes 36 / 46
Wide-Sense Stationary (WSS) Processes

To show that a process is stationary, we need the overall joint-PDF.

– Quite impossible to get, except for special cases.

However, we can often estimate the process’

– Expected value
– Autocorrelation function

If only the expected value and the autocorrelation function satisfy the
property of stationarity, we call this process wide sense stationary
(WSS).

– Hence, we don’t know anything about other properties of the

process!

5. stochastic processes 37 / 46
Wide-Sense Stationary (WSS) Processes
A process is wide-sense stationary, if and only if
– The expected value E[X (t)] does not depend on time:
E[X (t)] = c.
– The autocorrelation function only depends on the time difference
τ and not the absolute time t:
RX (t, τ ) = RX (τ )
or
RX [n, k] = RX (k)

Example: sinusoidal random process, X (t) = sin(ωt + Φ) where Φ is

uniformly distributed on [0, 2π]: derive that
1
CX (τ ) = RX (τ ) = cos(ωτ )
2
5. stochastic processes 38 / 46
Problem 13.9.3 (important for later!)

True or False: If Xn is a wide sense stationary random sequence with

E[Xn ] = c, then Yn = Xn − Xn−1 is a wide sense stationary random
sequence.

5. stochastic processes 39 / 46
Problem 13.9.3 (important for later!)

True or False: If Xn is a wide sense stationary random sequence with

E[Xn ] = c, then Yn = Xn − Xn−1 is a wide sense stationary random
sequence.

True: First we observe that E[Yn ] = E[Xn ] − E[Xn−1 ] = 0, which does

not depend on n. Second, we verify that

RY [n, k] = E[Yn Yn+k ]

= E[(Xn − Xn−1 )(Xn+k − Xn+k−1 )]
= E[Xn Xn+k ] − E[Xn Xn+k−1 ] − E[Xn−1 Xn+k ] + E[Xn−1 Xn+k−1 ]
= RX [k] − RX [k − 1] − RX [k + 1] + RX [k] ,

which does not depend on n. Hence, Yn is WSS.

5. stochastic processes 39 / 46
Problem 13.9.6
X (t) and Y (t) are independent wide sense stationary processes.
Determine if W (t) = X (t)Y (t) is wide-sense stationary.

5. stochastic processes 40 / 46
Problem 13.9.6
X (t) and Y (t) are independent wide sense stationary processes.
Determine if W (t) = X (t)Y (t) is wide-sense stationary.
True: Independence of X (t) and Y (t) implies

E[W (t)] = E[X (t)Y (t)] = E[X (t)] E[Y (t)] = µX µY

and

RW (t, τ ) = E[W (t)W (t + τ )]

= E[X (t)Y (t) X (t + τ )Y (t + τ )]
= E[X (t)X (t + τ ) Y (t)Y (t + τ )]
= E[X (t)X (t + τ )] E[Y (t)Y (t + τ )] (by independence)
= RX (τ )RY (τ )

Since W (t) has constant expected value and the autocorrelation

depends only on the time difference τ , W (t) is wide-sense stationary.
5. stochastic processes 40 / 46
WSS Processes and the autocorrelation function
Important properties of RX (τ ) for WSS processes:

RX (0) ≥ 0
RX (τ ) = RX (−τ )
RX (0) ≥ |RX (τ )|

Example:
1
RX (t, τ ) = cos(ωτ )
6

5. stochastic processes 41 / 46
Cross correlation for Stochastic Processes
In addition to the autocorrelation function, we can also define the
cross-correlation between two stochastic processes:
RXY (t, τ ) = E[X (t)Y (t + τ )]
RXY [n, k] = E[Xn Yn+k ]

Two random processes X (t) and Y (t) are jointly wide sense
stationary, if X (t) and Y (t) are wide sense stationary and
RXY (t, τ ) = RXY (τ ).

If X (t) and Y (t) are jointly WSS, then

RXY (τ ) = RYX (−τ ).
(and of course similar for time-discrete processes)

5. stochastic processes 42 / 46
Example
Xn is a zero mean WSS stochastic process. Let Yn = (−1)n Xn .

E[Yn ] = (−1)n E[Xn ] = 0, since Xn is zero mean.

RY [n, k] = E[Yn Yn+k ]
= E[(−1)n Xn (−1)n+k Xn+k ]
= (−1)2n+k E[Xn Xn+k ] = (−1)k RX [k]
Process Yn is WSS as RY [n, k] only depends on k.

RXY [n, k] = E[Xn Yn+k ]

= E[Xn (−1)n+k Xn+k ]
= (−1)n+k E[Xn Xn+k ] = (−1)n+k RX [k]
Xn and Yn are not jointly WSS as their cross-correlation function
depends on both n and k.

5. stochastic processes 43 / 46
Problem 13.10.2
X (t) is a wide sense stationary random process. Let
(a) Y (t) = X (t + a)
Are Y (t) and X (t) jointly wide sense stationary?

5. stochastic processes 44 / 46
Problem 13.10.2
X (t) is a wide sense stationary random process. Let
(a) Y (t) = X (t + a)
Are Y (t) and X (t) jointly wide sense stationary?
Since E[Y (t)] = E[X (t + a)] = µX and
RY (t, τ ) = E[Y (t)Y (t + τ )]
= E[X (t + a)X (t + τ + a)] = RX (τ ) ,
we have verified that Y (t) is wide sense stationary. Next we calculate
the cross correlation:
RXY (t, τ ) = E[X (t)Y (t + τ )]
= E[X (t)X (t + τ + a)] = RX (τ + a) .
Since RXY (t, τ ) depends on the time difference τ but not on the
absolute time t, we conclude that X (t) and Y (t) are jointly wide sense
stationary.
5. stochastic processes 44 / 46
Problem 13.10.2 (cont’d)
Now repeat for
(b) Y (t) = X (at)

5. stochastic processes 45 / 46
Problem 13.10.2 (cont’d)
Now repeat for
(b) Y (t) = X (at)

Since E[Y (t) = E[X (at)] = µX and

RY (t, τ ) = E[Y (t)Y (t + τ )]
= E[X (at)X (a(t + τ ))]
= E[X (at)X (at + aτ )] = RX (aτ ) ,
we have verified that Y (t) is wide sense stationary. Now we calculate
the cross correlation:
RXY (t, τ ) = E[X (t)Y (t + τ )]
= E[X (t)X (a(t + τ ))] = RX ((a − 1)t + aτ ) .
Except for the trivial case where a = 1 and Y (t) = X (t), RXY (t, τ )
depends on both the absolute time t and the time difference τ : we
conclude that X (t) and Y (t) are not jointly wide sense stationary.
5. stochastic processes 45 / 46
To do for this lecture:

Make some selected exercises:

13.1.1, 13.3.1, 13.7.1, 13.7.3, 13.7.5, 13.9.3, 13.9.5, 13.9.7, 13.10.1,
13.10.3

Next lecture, we’ll wrap up Ch.13 (ergodicity), and start with

Supplement Sections 1 and 2.

5. stochastic processes 46 / 46