Stochastic Processes

Lecture-6

Dr. Men Zhirong, Prof. Xu Huaping

School of Electronics and Information Engineering
[email protected]
Stochastic Process
◼Homework
Suppose 𝑿 and 𝒀 are independent with E=0 and D=1, Z(t)=X+Yt, then calculate m(t),
D[Z(t)], R(t1,t2) and C(t1,t2).
Stochastic Process
◼Homework
Suppose 𝑿 and 𝒀 are independent with E=0 and D=1, Z(t)=X+Yt, then calculate m(t),
D[Z(t)], R(t1,t2) and C(t1,t2).
Stochastic Process
◼Considering more than 2 stochastic processes,

Figure 2.7 Input and output process of a linear system

Stochastic Process
◼Cross-correlation function and cross-covariance function of 𝑋 𝑡 𝑎𝑛𝑑 𝑌 𝑡

The 2-nd joint moment about zero

∞ ∞
𝑅𝑋𝑌 𝑡1 ，𝑡2 = 𝐸 𝑋 𝑡1 𝑌 𝑡2 = න න 𝑥𝑦𝑓 𝑥, 𝑡1 ; 𝑦, 𝑡2 𝑑𝑥 𝑑𝑦
−∞ −∞
The 2-nd central joint moment

𝐶𝑋𝑌 𝑡1 ，𝑡2 = 𝐸 𝑋 𝑡1 − 𝑚𝑋 𝑡1 𝑌 𝑡2 − 𝑚𝑌 𝑡2
Stochastic Process
◼The relationship and properties between two stochastic processes of 𝑋 𝑡 𝑎𝑛𝑑 𝑌 𝑡
1. For any 𝑡1 , 𝑡2 , ⋯ , 𝑡𝑛 and 𝑡1′ , 𝑡2′ , ⋯ , 𝑡𝑚
′
, if

𝑓𝑋𝑌 = 𝑥1 , 𝑥2 , ⋯ , 𝑥𝑛 ; 𝑡1 , 𝑡2 , ⋯ , 𝑡𝑛 ; 𝑦1 , 𝑦2 , ⋯ , 𝑦𝑚 ; 𝑡1′ , 𝑡2′ , ⋯ , 𝑡𝑚

′

= 𝑓𝑋 𝑥1 , 𝑥2 , ⋯ , 𝑥𝑛 ; 𝑡1 , 𝑡2 , ⋯ , 𝑡𝑛 𝑓𝑌 𝑦1 , 𝑦2 , ⋯ , 𝑦𝑚 ; 𝑡1′ , 𝑡2′ , ⋯ , 𝑡𝑚 ′

𝑋 𝑡 𝑎𝑛𝑑 𝑌 𝑡 are independent, and we have

𝑓𝑋𝑌 𝑥, 𝑡1 ; 𝑦, 𝑡2 = 𝑓𝑋 𝑥, 𝑡1 𝑓𝑌 𝑦, 𝑡2

Then the cross-correlation function and cross-covariance function are

∞ ∞
𝑅𝑋𝑌 𝑡1 ，𝑡2 = 𝐸 𝑋 𝑡1 𝑌 𝑡2 = න 𝑥𝑓 𝑥, 𝑡1 𝑑𝑥 න 𝑦𝑓 𝑦, 𝑡2 𝑑𝑦 = 𝑚𝑋 𝑡1 𝑚𝑌 𝑡2
−∞ −∞

𝐶𝑋𝑌 𝑡1 ，𝑡2 = 𝐸 𝑋 𝑡1 − 𝑚𝑋 𝑡1 𝑌 𝑡2 − 𝑚𝑌 𝑡2
= 𝐸 𝑋 𝑡1 − 𝑚𝑋 𝑡1 𝐸 𝑌 𝑡2 − 𝑚𝑌 𝑡2
=0
Stochastic Process
◼The relationship and properties between two stochastic processes of 𝑋 𝑡 𝑎𝑛𝑑 𝑌 𝑡
2. For any 𝑡1 𝑎𝑛𝑑 𝑡2 , if the cross-covariance is 0, then 𝑋(𝑡) 𝑎𝑛𝑑 𝑌(𝑡) are not related.

𝐶𝑋𝑌 𝑡1 ，𝑡2 = 0

3. For any 𝑡1 𝑎𝑛𝑑 𝑡2 , if the cross-correlation is 0, then 𝑋(𝑡) 𝑎𝑛𝑑 𝑌(𝑡) are orthogonal.

𝑅𝑋𝑌 𝑡1 ，𝑡2 = 0
 Stochastic Process
◼Classification: according to statistical characteristics

Stationary stochastic process

𝐹n 𝑥1 , 𝑥2 , ⋯ , 𝑥𝑛 ; 𝑡1 , 𝑡2 , ⋯ , 𝑡𝑛 = 𝐹n 𝑥1 , 𝑥2 , ⋯ , 𝑥𝑛 ; 𝑡1 + 𝜏, 𝑡2 + 𝜏, ⋯ , 𝑡𝑛 +𝜏 𝑛 = 1,2 ⋯

The statistical characteristics doesn’t change with time.

E.g. System thermal noise in steady state;

Non-stationary stochastic process

𝐹n 𝑥1 , 𝑥2 , ⋯ , 𝑥𝑛 ; 𝑡1 , 𝑡2 , ⋯ , 𝑡𝑛 ≠ 𝐹n 𝑥1 , 𝑥2 , ⋯ , 𝑥𝑛 ; 𝑡1 + 𝜏, 𝑡2 + 𝜏, ⋯ , 𝑡𝑛 +𝜏 𝑛 = 1,2 ⋯
The statistical characteristics depend on the start time.

E.g. Movement of the earth surface caused by an earthquake;

 Stochastic Process
◼Classification: according to memory characteristics

Memoryless stochastic process 𝑛

𝐹n 𝑥1 , 𝑡1 ; 𝑥2 , 𝑡2 ; ⋯ ; 𝑥𝑛 , 𝑡𝑛 = ෑ 𝐹1 𝑥𝑗 , 𝑡𝑗
𝑗=1
E.g. Gaussian noise

Markov stochastic process

First order

𝐹n 𝑥𝑛 , 𝑡𝑛 𝑥𝑛−1 , 𝑡𝑛−1 ; 𝑥𝑛−2 , 𝑡𝑛−2 ; ⋯ ; 𝑥2 , 𝑡2 ; 𝑥1 , 𝑡1 = 𝐹n 𝑥𝑛 , 𝑡𝑛 𝑥𝑛−1 , 𝑡𝑛−1

Second order
𝐹n 𝑥𝑛 , 𝑡𝑛 𝑥𝑛−1 , 𝑡𝑛−1 ; 𝑥𝑛−2 , 𝑡𝑛−2 ; ⋯ ; 𝑥2 , 𝑡2 ; 𝑥1 , 𝑡1 = 𝐹n 𝑥𝑛 , 𝑡𝑛 𝑥𝑛−1 , 𝑡𝑛−1 ; 𝑥𝑛−2 , 𝑡𝑛−2
 Stochastic Process - Stationary stochastic process
◼Strict (strong) stationary process
Definition: Suppose {𝑋 𝑡 , 𝑡 ∈ 𝑇} is a stochastic process, for any positive integer n, real numbers
𝑡1 , 𝑡2 , ⋯ , 𝑡𝑛 and 𝜏, the n-th distribution function of random variables 𝑋 𝑡1 , 𝑋 𝑡2 , …, 𝑋 𝑡𝑛 is the same as
the n-th distribution function of random variables 𝑋 𝑡1 + 𝜏 , 𝑋 𝑡2 + 𝜏 , …, 𝑋 𝑡𝑛 + 𝜏 , then 𝑋 𝑡 is a strict
stationary process.

𝐹n 𝑥1 , 𝑥2 , ⋯ , 𝑥𝑛 ; 𝑡1 , 𝑡2 , ⋯ , 𝑡𝑛 = 𝐹n 𝑥1 , 𝑥2 , ⋯ , 𝑥𝑛 ; 𝑡1 + 𝜏, 𝑡2 + 𝜏, ⋯ , 𝑡𝑛 +𝜏 𝑛 = 1,2 ⋯

Also means

𝑓n 𝑥1 , 𝑥2 , ⋯ , 𝑥𝑛 ; 𝑡1 , 𝑡2 , ⋯ , 𝑡𝑛 = 𝑓n 𝑥1 , 𝑥2 , ⋯ , 𝑥𝑛 ; 𝑡1 + 𝜏, 𝑡2 + 𝜏, ⋯ , 𝑡𝑛 +𝜏

𝑓1 𝑥1 , 𝑡1 = 𝑓1 𝑥1 , 𝑡1 + 𝜏 = 𝑓1 𝑥1 , −𝜏 + 𝜏 = 𝑓1 𝑥1 , 0 𝑓1 𝑥, 𝑡 = 𝑓1 𝑥
The one-dimensional density function of strict stationary process is independent of time.
Stochastic Process - Stationary stochastic process
◼Strict (strong) stationary process
∞
𝐸𝑋 𝑡 = න 𝑥 𝑓1 𝑥 𝑑𝑥 = 𝑚𝑋
−∞

∞
𝐷𝑋 𝑡 = 𝐸 𝑋 𝑡 − 𝑚𝑋 2 =න 𝑥 − 𝑚𝑋 2 𝑓1 𝑥 𝑑𝑥 = 𝜎 2
−∞

𝑓2 𝑥1 , 𝑥2 ; 𝑡1 , 𝑡2 = 𝑓2 𝑥1 , 𝑥2 ; 𝑡1 − 𝑡2 = 𝑓2 𝑥1 , 𝑥2 ; 𝜏
∞ ∞
𝑅 𝑡, 𝑡 − 𝜏 = 𝐸 𝑋 𝑡 𝑋 𝑡 − 𝜏 =න න 𝑥1 𝑥2 𝑓2 𝑥1 , 𝑥2 ; 𝜏 𝑑 𝑥1 𝑑𝑥2 = 𝑅 𝜏
−∞ −∞

𝐶 𝜏 = 𝐸 𝑋 𝑡 − 𝑚𝑋 𝑋 𝑡 − 𝜏 − 𝑚𝑋 = 𝑅 𝜏 − 𝑚𝑋2

𝐶 0 = 𝑅 0 − 𝑚𝑋2 = 𝜎 2
 Stochastic Process - Stationary stochastic process
◼Wide-sense (weak) stationary process
Definition: Suppose {𝑋 𝑡 , 𝑡 ∈ 𝑇} is a stochastic process with E 𝑋 2 𝑡 < ∞, if

𝐸𝑋 𝑡 = 𝑚𝑋 = 𝐶

and
𝑅 𝑡1 , 𝑡2 = 𝐸 𝑋 𝑡 𝑋 𝑡 − 𝜏 = 𝑅 𝜏 , 𝜏 = 𝑡1 − 𝑡2

then 𝑋 𝑡 is a wide-sense stationary process.

For the wide-sense stationary process, only the first-order and second order moment are discussed,
which is not as strict as the strong stationary process.
 Stochastic Process - Stationary stochastic process
◼Example 1

Let the stochastic process 𝑋 𝑡 = 𝐴𝑡. 𝐴 is a random variable which is uniformly distributed between [0,1]. Is 𝑋(𝑡) a
stationary stochastic process?
 Stochastic Process - Stationary stochastic process
◼Example 2

Let the stochastic process 𝑍 𝑡 = 𝑋𝑐𝑜𝑠𝑡 + 𝑌𝑠𝑖𝑛𝑡, −∞ < 𝑡 < ∞. 𝑋 and 𝑌 are two independent random variables,
taking the values -1 and 2 with the probability of 2/3 and 1/3 respectively. Discuss the stationarity of the stochastic
process 𝑍 𝑡 .
 Stochastic Process - Stationary stochastic process
◼Example 2

Let the stochastic process 𝑍 𝑡 = 𝑋𝑐𝑜𝑠𝑡 + 𝑌𝑠𝑖𝑛𝑡, −∞ < 𝑡 < ∞. 𝑋 and 𝑌 are two independent random variables,
taking the values -1 and 2 with the probability of 2/3 and 1/3 respectively. Discuss the stationarity of the stochastic
process 𝑍 𝑡 .
 Stochastic Process - Stationary stochastic process
◼Example 3

Set random phase cosine wave 𝑋 𝑡 = 𝑐𝑜𝑠 𝜔0 𝑡 + 𝛩 . 𝜔0 is a constant and 𝛩 is a random variable uniformly
distributed in the interval [0,2π], as:
1
𝑓 𝜃 = ቐ2𝜋 , 0 ≤ 𝜃 ≤ 2𝜋
0, 𝑒𝑙𝑠𝑒
Analyze the stationarity of the stochastic process 𝑋 𝑡 .
 Stochastic Process - Stationary stochastic process
◼Example 3

Set random phase cosine wave 𝑋 𝑡 = 𝑐𝑜𝑠 𝜔0 𝑡 + 𝛩 . 𝜔0 is a constant and 𝛩 is a random variable uniformly
distributed in the interval [0,2π], as:
1
𝑓 𝜃 = ቐ2𝜋 , 0 ≤ 𝜃 ≤ 2𝜋
0, 𝑒𝑙𝑠𝑒
Analyze the stationarity of the stochastic process 𝑋 𝑡 .
Stochastic Process - Stationary stochastic process
◼Second-order moment
Definition: Suppose {𝑋 𝑡 , 𝑡 ∈ 𝑇} is a stochastic process with E 𝑋 2 𝑡 < ∞, then 𝑋 𝑡 is called second-
order moment process.

For a second-order moment process, 𝑅𝑋 𝑡1 , 𝑡2 is the correlation function, then

𝑅𝑋 𝑡2 , 𝑡1 = 𝑅𝑋 ∗ 𝑡1 , 𝑡2 𝑡1 , 𝑡2 ∈ 𝑇
Stochastic Process - Stationary stochastic process
◼Properties of autocorrelation function of stationary stochastic process
∞ ∞
𝑅 𝑡, 𝑡 − 𝜏 = 𝐸 𝑋 𝑡 𝑋 𝑡 − 𝜏 =න න 𝑥1 𝑥2 𝑓2 𝑥1 , 𝑥2 ; 𝜏 𝑑 𝑥1 𝑑𝑥2 = 𝑅 𝜏
−∞ −∞

1. 𝑅 𝜏 is an even function 𝑅 𝜏 = 𝑅 −𝜏 𝐶𝑋 𝜏 = 𝐶𝑋 −𝜏

2. 𝑅 𝜏 has the maximum value when 𝜏 = 0 𝑅 𝜏 ≤𝑅 0 𝐶 𝜏 ≤𝐶 0

Stochastic Process - Stationary stochastic process
◼Properties of autocorrelation function of stationary stochastic process
∞ ∞
𝑅 𝑡, 𝑡 − 𝜏 = 𝐸 𝑋 𝑡 𝑋 𝑡 − 𝜏 =න න 𝑥1 𝑥2 𝑓2 𝑥1 , 𝑥2 ; 𝜏 𝑑 𝑥1 𝑑𝑥2 = 𝑅 𝜏
−∞ −∞

3. When the time interval is big enough, the states at 𝑡1 and 𝑡2 are unrelated.

lim 𝑅 𝜏 = 𝑅 ∞ = 𝑚2 𝐶 ∞ = lim 𝐶 𝜏 = 0
𝜏→∞ 𝜏→∞

4. When 𝜏 = 0
𝑅 0 = 𝐸 𝑋2 𝑡 =𝐷 𝑋 𝑡 + 𝑚2 = 𝜎 2 + 𝑚2
𝜎 2 = 𝑅 0 − 𝑚2 = 𝑅 0 − 𝑅 ∞
𝐶 𝜏 = 𝑅 𝜏 − 𝑚2 = 𝑅 𝜏 − 𝑅 ∞
 Stochastic Process - Stationary stochastic process
◼Example 4

Let 𝑅 𝜏 be the autocorrelation function of a real stochastic process 𝑋 𝑡 . 𝑋 𝑡 is the generalized stationary process.
Prove that if for any 𝜏1 ≠ 0 , 𝑅 𝜏1 = 𝑅 0 , then 𝑅 𝜏 must be periodic.
 Stochastic Process - Stationary stochastic process
◼Example 4

Let 𝑅 𝜏 be the autocorrelation function of a real stochastic process 𝑋 𝑡 . 𝑋 𝑡 is the generalized stationary process.
Prove that if for any 𝜏1 ≠ 0 , 𝑅 𝜏1 = 𝑅 0 , then 𝑅 𝜏 must be periodic.
 Stochastic Process - Stationary stochastic process
◼Example 5
The typical sample function of the random telegraph process 𝑋 𝑡 is shown in Figure 2.9. The
duration of the process is −∞ to ∞. At any time 𝑡, the sample function has only two values, “1”
and “0” and the probability of each occurrence is 1/2. The moment when 𝑋 𝑡 exchanges
from “1” to “0” and “0” to “1” is random. In any given time period 𝜏, the probability of the
number of exchanges 𝑘 obeys the Poisson distribution:
𝜆𝜏 𝑘 −𝜆𝜏
𝑅 𝑘, 𝜏 = 𝑒
𝑘!
In the formula, 𝜆 is the average number of transformations per unit time. Try to calculate the
mathematical expectation and correlation function of the stochastic process 𝑋 𝑡 .

The typical sample function of the random telegraph process

Stochastic Process - Stationary stochastic process
◼Example 5 𝜆𝜏 𝑘 −𝜆𝜏
𝑅 𝑘, 𝜏 = 𝑒
𝑘!
Stochastic Process - Stationary stochastic process
◼Example 5 𝜆𝜏 𝑘 −𝜆𝜏
𝑅 𝑘, 𝜏 = 𝑒
𝑘!
Stochastic Process - Stationary stochastic process
◼Example 5 𝜆𝜏 𝑘 −𝜆𝜏
𝑅 𝑘, 𝜏 = 𝑒
𝑘!
Stochastic Process - Stationary stochastic process
◼Properties of cross-correlation function of stationary stochastic process

𝑅𝑋𝑌 𝜏 = 𝐸 𝑋 𝑡 𝑌 𝑡 − 𝜏

𝑅𝑋𝑌 𝜏 is not just an even function of 𝜏.

𝑅𝑋𝑌 −𝜏 = 𝑅𝑌𝑋 𝜏

𝐶𝑋𝑌 −𝜏 = 𝐶𝑌𝑋 𝜏

𝑅𝑋𝑌 𝜏 2 ≤ 𝑅𝑋 0 𝑅𝑌 0
𝐶𝑋𝑌 𝜏 2 ≤ 𝜎𝑋2 𝜎𝑌2
 Stochastic Process - Stationary stochastic process
◼Example 6

Suppose the relationship between 𝑋(𝑡) and 𝑌(𝑡) of a stationary stochastic process is 𝑌 𝑡 = 𝑏𝑋 𝑡 + 𝑐. In the
formula, 𝑏 and 𝑐 are both constants. Calculate the cross-covariance.
 Stochastic Process - Stationary stochastic process
◼Example 6

Suppose the relationship between 𝑋(𝑡) and 𝑌(𝑡) of a stationary stochastic process is 𝑌 𝑡 = 𝑏𝑋 𝑡 + 𝑐. In the
formula, 𝑏 and 𝑐 are both constants. Calculate the cross-covariance.
Stochastic Process - Stationary stochastic process
◼Correlation coefficient and correlation time

When describing the correlation of 𝑋 𝑡 − 𝑚𝑋 and 𝑋 𝑡 − 𝜏 − 𝑚𝑋 , the

covariance function 𝐶𝑋 𝜏 = 𝑋 𝑡 − 𝑚𝑋 𝑋 𝑡 − 𝜏 − 𝑚𝑋 can be used, but
not enough. The correlation would be small even if they are strongly correlated,
when 𝑋 𝑡 − 𝑚𝑋 and 𝑋 𝑡 − 𝜏 − 𝑚𝑋 are small.
Hence, the intensity of should be removed, which is named correlation
coefficient:
𝐶𝑋 𝜏 𝑅𝑋 𝜏 − 𝑚𝑋2
𝑟𝑋 𝜏 = =
𝐶𝑋 0 𝜎𝑋2

Typical 𝑟𝑋 𝜏 curve
Stochastic Process - Stationary stochastic process
◼Correlation coefficient and correlation time
When 𝜏 → ∞, 𝑟𝑋 𝜏 = 𝑟𝑋 ∞ = 0, which means that 𝑋 𝑡 and 𝑋 𝑡 − 𝜏 are unrelated any more. In practice, for any 𝜏 > 𝜏0 ,
if 𝑋 𝑡 and 𝑋 𝑡 − 𝜏 are unrelated, then 𝜏0 is called correlation time.

∞
𝜏0 = න 𝑟𝑋 𝜏 𝑑𝜏
0

For engineering, 𝑟𝑋 𝜏0 ≤ 0.05 Correlation time

 Stochastic Process - Stationary stochastic process
◼Homework
The covariance functions of stochastic processes 𝑋(𝑡) and 𝑌(𝑡) are:

1 −2𝜆 𝜏 𝑠𝑖𝑛𝜆𝜏
𝐶𝑋 𝜏 = 𝑒 , 𝐶𝑌 𝜏 =
4 𝜆𝜏
(1) Discuss the fluctuation rate of the two processes;
𝜋
(2) Discuss the correlation of the two processes at 𝜏 = ;
𝜆
𝜋
(3) Discuss the correlation of process 𝑌(𝑡) at 𝜏 = 0 and 𝜏 = .
𝜆