lecture 5,6,7: random variables and signals aliazam abbasfar
DESCRIPTION
Random variables (RV) PDF, CDF f X (x) = d/dx [ F X (x) ] Mean, variance, momentsE[x], Var[x], E[x n ] Functions of RVs Y = g(X) Several RVs Joint PDF, CDF Conditional probability Sum Independent RVs Correlation of 2 RVsE[x y] Example : Binary communication with noiseTRANSCRIPT
Lecture 5,6,7: Random variables and signals
Aliazam Abbasfar
OutlineRandom variables overview
Random signals
Signals correlation
Power spectral density
Random variables (RV)PDF, CDF
fX(x) = d/dx [ FX(x) ]
Mean, variance, moments E[x], Var[x], E[xn]
Functions of RVs Y = g(X)
Several RVs Joint PDF, CDF Conditional probability Sum
Independent RVsCorrelation of 2 RVs E[x y]
Example : Binary communication with noise
(x)y)/f(x,fx)X|(yf XXYY
dx )(xfdy (y)f iXY
dx x)-z(x,f(z)f XYZ
x y
XYXYXY dydx y)(x,fy)(x,F y);(x,f
Binomial distributionX = # of successes in N independent trials
p : success probability (1-p : failure)
Sum of N binary RVs : X = xi
If N is large, it becomes a Gaussian PDF x =Np x
2 =Npq
Example : Error probability in binary packets
N
0k
k-NkX k)δ(xqp
kN
(x)f
Gaussian RVs and the CLT PDF (mean and variance)
CDF defined by error function (erf(•))
Central Limit Theorem: X1,…,Xn i.i.d Let Y=iXi, Z=(Y-Y)/Y As n, Z becomes Gaussian, x=0, x
2=1.
Uncorrelated Gaussian RVs are independent
]/σ)μ[(x
XX
2X
2Xe
σ2π1(x)f
x
x
N(x,x2) Z~N()
Tails decreaseexponentially
2x/erfc21Q(x)erf(u),1erfc(u),
σ2μxerf1
21(x)Fx)p(X
X
XX
Random ProcessesEnsemble of random signals (sample functions)
Deterministic signals with RVs Voltage waveforms Message signals Thermal noise
Samples of a random signal x(t) ; a random variable
E[x(t)], Var[x(t)] x(t1), x(t2) joint random variables
CorrelationCorrelation = statistic similarity
Cross correlation of two random signals RXY(t1,t2)=E[x(t1)y(t2)] Uncorrelated/Independent RSs
Autocorrelation R(t1,t2)=E[x(t1)x(t2)] RX(t,t) = E[x2(t)] = Var[x(t)]+E[x]2
Average power P = E[Pi] = E[<xi
2(t)>] = <RX(t,t)> Most of RSs are power signals ( 0< P < )
Wide Sense Stationary (WSS)A process is WSS if
E[x(t)]=X
RX(t1,t2)= E[x(t1)x(t2)]=RX(t2-t1)= RX()RX(0)=E[x2(t)]<Stationary in 1st and 2nd moments
AutocorrelationRX()= RX(-)|RX()| RX()RX()=0 : samples separated by uncorrelated
Average power P = <E[x2(t)]> = Rx(0)
Ergodic processTime average of any sample function =
Ensemble average ( any i and any g)<g(xi(t))> = E[g(x(t))]
Ensemble averages are time-independent DC : <xi(t)> = E[ x(t) ] = mx Total power : <xi
2(t)> = E[ x2(t) ] = (sx)2 + (mx)2
Average power : P = E[<xi
2(t)>] = Pi
Use one sample function to estimate signal statistics Time-average instead of ensemble average
ExamplesSinusoid with random phase
DC signal with random level
Binary NRZ signaling
Power spectral densityTime-averaged autocorrelation
Power spectral density
Average power
)](R[(f)G xxx F ]
T(f)X
lim[E](f)G[E(f)G2
T
Txx i
df(f)G P-
x
]τ)-(t(t)xxE[τ)-t(t,R)(R iixxxx
ExamplesY(t) = X(t) cos(wct)
WSS ?RY() and GY(f)
Correlations for LTI systems
If x(t) is WSS, x(t) and y(t) are jointly WSS
mY = H(0) mX RYX() = h() Rxx()RXY() = RYX(-)= h(-) Rxx()RYY() = h() h(-) Rxx()
GY(f) = |H(f)|2 GX(f)
Sum processz(t) = x(t) + y(t)
RZ() = RX() + RY() + RXY() + RXY(-) GZ(f) = GX(f) + GY(f) + 2 Re[GXY(f)]
If X and Y are uncorrelated RXY() = mX mY
GZ(f) = GX(f) + GY(f) + 2 mX mY (f)
ReadingCarlson Ch. 9.1, 9.2
Proakis&Salehi 4.1, 4.2, 4.3 4.4