Revision as of 16:43, 24 February 2017 by Zhongc (Talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)


ECE Ph.D. Qualifying Exam

Communication, Networking, Signal and Image Processing (CS)

Question 1: Probability and Random Processes

August 2013



Part 2

Let $ X_1,X_2,... $ be a sequence of jointly Gaussian random variables with covariance

$ Cov(X_i,X_j) = \left\{ \begin{array}{ll} {\sigma}^2, & i=j\\ \rho{\sigma}^2, & |i-j|=1\\ 0, & otherwise \end{array} \right. $

Suppose we take 2 consecutive samples from this sequence to form a vector $ X $, which is then linearly transformed to form a 2-dimensional random vector $ Y=AX $. Find a matrix $ A $ so that the components of $ Y $ are independent random variables You must justify your answer.


Solution 1

Suppose

$ A=\left(\begin{array}{cc} a & b\\ c & d \end{array} \right) $.

Then the new 2-D random vector can be expressed as

$ Y=\left(\begin{array}{c}Y_1 \\ Y_2\end{array} \right)=A\left(\begin{array}{c}X_i \\ X_j\end{array} \right)=\left(\begin{array}{c}aX_i+bX_j \\ cX_i+dX_j\end{array} \right) $


Therefore,

$ \begin{array}{l}Cov(Y_1,Y_2)=E[(aX_i+bX_j-E(aX_i+bX_j))(cX_i+dX_j-E(cX_i+dX_j))] \\ =E[(aX_i+bX_j-aE(X_i)-bE(X_j))(cX_i+dX_j-cE(X_i)-dE(X_j))] \\ =E[acX_i^2+adX_iX_j-acX_iE(X_i)-adX_iE(X_j)+bcX_iX_j+bdX_j^2-bcX_jE(X_i)\\ -bdX_jE(X_j)-acX_iE(X_i)-adX_jE(X_i)+acE(X_i)^2+adE(X_i)E(X_j)\\ -bcX_iE(X_j)-bdX_jE(X_j)+bcE(X_i)E(X_j)+bdE(X_i)^2]\\ =E(ac(X_i-E(X_i))^2+(ad+bc)(X_i-E(X_i)(X_j-E(X_j))+bd(X_j-E(X_j))^2]\\ =(ac)Cov(X_i,X_i)+(ad+bc)Cov(X-i,X_j)+(bd)Cov(X_j,X_j)\\ =ac\sigma^2+(ad+bc)\rho\sigma^2+bd\sigma^2 \end{array} $

Let the above formula equal to 0 and $ a=b=d=1 $, we get $ c=-1 $.

Therefore, a solution is

$ A=\left(\begin{array}{cc} 1 & 1\\ -1 & 1 \end{array} \right) $.

Comments on solution 1

More procedures and explanations would be better.


Solution 2

Assume

$ Y=\left(\begin{array}{c}Y_i \\ Y_j\end{array} \right)=A\left(\begin{array}{c}X_i \\ X_j\end{array} \right)=\left(\begin{array}{c}a_{11}X_i+a_{12}X_j \\ a_{21}X_i+a_{22}X_j\end{array} \right) $.

Then

$ \begin{array}{l}E(Y_iY_j)=E[(a_{11}X_i+a_{12}X_j)(a_{21}X_i+a_{22}X_j)]\\ =a_{11}a_{21}\sigma^2+a_{12}a_{22}\sigma^2+(a_{11}a_{21}+a_{22}a_{11})E(X_iX_j) \end{array} $

For $ |i-j|\geq1 $, $ E(X_i,X_j)=0 $. Therefore, $ a_{11}a_{21}+a_{12}a_{22}=0 $.

One solution can be

$ A=\left(\begin{array}{cc} 1 & -1\\ 1 & 1 \end{array} \right) $.


Critique on Solution 2:

1. $ E(Y_iY_j)=0 $ is not the condition for the two random variables to be independent.

2. "For $ |i-j|\geq1 $, $ E(X_i,X_j)=0 $" is not supported by the given conditions.


Solution 3

$ Y=\left(\begin{array}{c}Y_1 \\ Y_2\end{array} \right)=AX=\left(\begin{array}{cc} a & b\\ c & d \end{array} \right)\left(\begin{array}{c}X_i \\ X_j\end{array} \right)=\left(\begin{array}{c}aX_i+bX_j \\ cX_i+dX_j\end{array} \right) $

We know that the sum of two independent Gaussian distributed random variables is still Gaussian distributed.

Thus, $ Y_1,Y_2 $ are Gaussian distributed random variables. If they are uncorrelated, then they are also independent.

$ r = \frac{COV(Y_1,Y_2)}{\sigma1\sigma2} = 0 $

which indicates that

$ E(Y_1Y_2) - E(Y_1)E(Y_2) = 0 $

we know that $ |i-j|=1 $

$ E(Y_1Y_2) = E((aX_i+bX_j)(cX_i+dX_j)) = E(acX_iX_i+adX_iX_j+bcX_jX_i+bdX_jX_j) \\ = ac({\sigma}^{2}+{E(X_i)}^2)+ad(\rho{\sigma}^{2}+E(X_j)E(X_i))+bc(\rho{\sigma}^{2}+E(X_j)E(X_i))+bd({\sigma}^{2}+{E(X_j)}^2) = 0 $

$ E(Y_1)E(Y_2) = E((aX_i+bX_j))E((cX_i+dX_j)) = (aE(X_i)+bE(X_j))(cE(X_i)+dE(X_j)) \\ = ac{E(X_i)}^{2}+adE(X_i)E(X_j)+bcE(X_i)E(X_j)+bd{E(X_j)}^{2} $

Therefore,

$ E(Y_1Y_2) - E(Y_1)E(Y_2) = ac{\sigma}^{2}+ad\rho{\sigma}^{2}+bc\rho{\sigma}^{2}+bd{\sigma}^{2} = 0 $

$ \left\{ \begin{array}{ll} ac = -bd\\ ad=-bc \end{array} \right. $

Thus, one of the possible solutions is $ a = 1,b = -1,c = 1, d = 1 $

Similar Question

Let $ X_1,X_2,... $ be a sequence of jointly Gaussian random variables with the same mean $ u $ and with covariance function where $ |\rho|<1 $

$ Cov(X_i,X_j) = \left\{ \begin{array}{ll} {\sigma}^2, & i=j\\ \rho{\sigma}^2, & |i-j|=1\\ 0, & otherwise \end{array} \right. $

Find the mean and variance of $ S_n = X_1 + ...+ X_n $


Back to QE CS question 1, August 2013

Back to ECE Qualifying Exams (QE) page

Alumni Liaison

Basic linear algebra uncovers and clarifies very important geometry and algebra.

Dr. Paul Garrett