up
PDF (letter size)
PDF (legal size)

ECE 3341 formulas (Stochastic processes) Northeastern Univ. Boston

Nasser M. Abbasi

1993   Compiled on November 16, 2018 at 11:19am  [public]

Contents

1 Statistical averages
2 Random sequences

1 Statistical averages

expectation, the expected or mean\[ E\left [ X\right ] \equiv \mu _x=\int _{-\infty }^\infty x\;f_x\left ( x\right ) \;dx \]

if \(x\) is discrete then\[ E\left [ X\right ] \equiv \mu _x=\sum _nx_nP_x\left ( x_n\right ) \]

for a normalized system, i.e. total weight = 1, then \(\mu _x\) can be considered to be the center of gravity.

expected value of \(Y=G\left ( x\right ) \)

\[ E\left [ g\left ( x\right ) \right ] \equiv \mu _y=\int _{-\infty }^\infty y\;f_Y\left ( y\right ) \;dy=\int _{-\infty }^\infty g\left ( x\right ) f_Y\left ( y\right ) \;dy \]

\[ \mu _y=\int _{-\infty }^\infty g\left ( x\right ) f_X\left ( x\right ) \;dx \]

 

theorm\[ f_Y\left ( y\right ) =\frac{f_X\left ( x=g^{-1}\left ( y\right ) \right ) }{\left | g^{^{\prime }}\left ( x\right ) \right | } \]

\[ dy=\left | g^{^{\prime }}\left ( x\right ) \right | \;dx \]

conditonal expectation\[ E\left [ Y|B\right ] \equiv \int _{-\infty }^\infty y\;f_{Y|B}\left ( y|b\right ) \;dy \]

moments, nth moment

\(n^{th}\) moment of \(X\) denoted by \(\epsilon _n\)\[ \epsilon _n\equiv \int _{-\infty }^\infty x^n\;f_X\left ( x\right ) \;dx \]

\[ \begin{array}{l} \epsilon _0=1 \\ \epsilon _1=\mu _x\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\; \text{mean value}\; \\ \epsilon _2=E\left [ X^2\right ] \;\;\;\;\;\;\;\;\;\;\; \text{mean squared value}\end{array} \]

central moments

\[ m_n\equiv \int _{-\infty }^\infty \left ( u-\mu _x\right ) ^n\;f_X\left ( x\right ) \;dx \]

\[ \begin{array}{lll} m_0 & = & 1 \\ m_1^{} & = & 0 \\ m_2 & = & E\left [ \{X-\mu _X\}^2\right ] \;\;\;\;\;\;\;\;\; \text{spread or variance =}\sigma _x^2 \\ m_3 & = & E\left [ \{X-\mu _X\}^3\right ] \;\;\;\;\;\;\;\;\;\text{skew}\end{array} \]

 

standard deviation

\[ \sigma _x=\sqrt{m_2} \]

 realtionships between moments

\[ \sigma ^2=m_2=\epsilon _2-\epsilon _1^2=E\left [ X^2\right ] -\left \{ E\left [ X\right ] \right \} ^2 \]

2 Random sequences

mean sequence

\[ \mu _X\left ( n\right ) \equiv E\left [ X_n\right ] =\int _{-\infty }^\infty x_n\;f\left ( x_n\right ) \;dx_n \]

autocorrelation Bisequence\[ R_X\left ( m,n\right ) \equiv E\left [ X_mX_n^{*}\right ] =\int \int x_mx_n^{*}\;f\left ( x_m,x_n\right ) \;dx_mdx_n \]

Auto covariance Bisequence\[ K_X\left ( m,n\right ) \equiv E\left [ \left \{ X_m-\mu _X\left ( m\right ) \right \} \left \{ X_m^{*}-\mu _X^{*}\left ( m\right ) \right \} \right ] \]

relation\[ K_X\left ( m,n\right ) =R_X\left ( m,n\right ) -\mu _X\left ( m\right ) \mu _X^{*}\left ( n\right ) \]

definitions

uncorrelated random sequence

\[ \begin{array}{l} if \\ K_x(m,n)=0\;\;\;\;\;\;\;\;\;\forall m,n\;\;\;m\neq n \\ \;\;\;\;\;\;\;\;\;\;\;\;\;\;=\sigma _x^2\;\;\;\;\;\;\;\;m=n\end{array} \]

or\[{\bf R_X\left ( m,n\right ) =\mu _X\left ( m\right ) \mu _X^{*}\left ( n\right ) \;\;\;\;\;\;\;\forall m,n\;\;\;m\neq n\;} \]

then the sequence is called uncorrelated random sequence

 

orthogonal random sequence

if\[ \begin{array}{l} R_X\left ( m,n\right ) =0\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\forall m,n\;\;\;m\neq n \\ \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;=E\left [ x_n^2\right ] \;\;\;\;\;\;\;\;\;\;\;\;m=n\end{array} \]

then the sequence is called an orthogonal random sequence

Gausian random sequence

if all kth order distributions of a random sequence \(X_n\) are jointly Gaussian then it is called a Gaussian random seq.

strict sense stationary SSS

if the kth order probability functions do not depend on the index n, then it is SSS.

Wide sense stationary WSS

if the mean function is constant and the autocorrelation (covariance) is shift-invariant then it is WSS.

i.e.\[ \mu _x\left ( n\right ) =\mu _x \]

and\[ R_X\left ( m,n\right ) =R_X\left ( m-n\right ) \]

\(^{}\)

usefull identity\[ \int _0^\infty x^ne^{-x}\;dx=n! \]

when adding 2 i.i.d R.V., their f’s convolve and their characterstic functions is multiplied

if X is an i.i.d, then at each n it is the same RV, and they are independent RV’s

 

convolution

for a discrete, linear time invariant

\[ y\left ( n\right ) =\sum _{m=-\infty }^\infty h(n-m)u(m) \]

\[ y=h*u \]

for a continouse liner time invariant \[ y\left ( t\right ) =\int _{-\infty }^\infty h\left ( t-\tau \right ) u\left ( \tau \right ) \;d\tau \]

\[ y=h*u \]