Professional Documents
Culture Documents
PowerSpectralDensity PDF
PowerSpectralDensity PDF
Derivation
In applying frequency-domain techniques to the analysis of
random signals the natural approach is to Fourier transform
the signals.
( ( )) ∫ X ( t ) e
F XT t = T
− j 2 π ft
dt , T < ∞
−∞
() ( ( ))
2 2
∫ XT t dt = ∫ F XT t df
−T / 2 −∞
Dividing through by T ,
T /2 ∞
1
() 1
( ( ))
2
∫ X T t dt = ∫ F X T t
2
df
T −T / 2 T −∞
Derivation
T /2 ∞
1
() 1
( ( )) Average signal power
2
∫ X T t dt = ∫ F X T t df ←
2
T −T / 2 T −∞ over time, T
If we let T approach infinity, the left side becomes the average
power over all time. On the right side, the Fourier transform
is not defined in that limit. But it can be shown that even
though the Fourier transform does not exist, its expected value
does. Then
⎛ 1 T /2 2 ⎞ ⎛1 ∞ ⎞
() ( ( ))
2
E⎜ ∫
⎝ T −T / 2
X T t dt ⎟ = E ⎜ ∫ F X T t
⎠ ⎝ T −∞
df ⎟
⎠
Derivation
Taking the limit as T approaches infinity,
T /2 ∞
1
( ) 1 ⎡
( ( )) ⎤
2
T →∞ T ∫
lim E X dt = lim ∫ E ⎢ F X T t
2
⎥⎦ df
−T / 2
T →∞ T
−∞ ⎣
⎡
( ( )) ⎤
2
⎢ F XT t ⎥
∞
( )
E X 2
= ∫ lim E ⎢
T →∞ T ⎥ df
−∞
⎢⎣ ⎥⎦
The integrand on the right side is identified as power spectral
density (PSD).
⎛
( ( )) ⎞
2
F XT t
GX ( )
f = lim E
T →∞ ⎜
⎜
T
⎟
⎟
⎜⎝ ⎟⎠
Derivation
∞
∫ G ( f ) df
−∞
X { ( )}
= mean − squared value of X t
∫ G ( f ) df
−∞
X { ( )}
= average power of X t
1
2π ∫2 π
( )
G X Ω dΩ = mean − squared value of X ⎡⎣ n ⎤⎦ { }
where the Fourier transform is the discrete-time Fourier transform (DTFT) defined by
( X ( F )) = ∫ X ( F ) e ( ) ∑ x ⎡⎣ n ⎤⎦ e
∞
x ⎡⎣ n ⎤⎦ = F −1
1
j 2 π Fn F
( )
dF ←⎯→ X F = F x ⎡⎣ n ⎤⎦ = − j 2 π Fn
n= −∞
( ( )) ( ) ∑ x ⎡⎣ n ⎤⎦ e
∞
1
x ⎡⎣ n ⎤⎦ = F −1
X Ω =
2π ∫2π
( )
X Ω e jΩn F
( )
dΩ ←⎯→ X Ω = F x ⎡⎣ n ⎤⎦ = − jΩn
n= −∞
PSD and Autocorrelation
( )
GX f = F RX τ( ( )) or G ( F ) = F ( R
X X
⎡⎣ m ⎤⎦ )
White Noise
White noise is a CT stochastic process whose PSD is constant.
( )
GX f = A
Signal power is the integral of PSD over all frequency space.
Therefore the power of white noise is infinite.
∞
( ) ∫ Ad f → ∞
E X2 =
−∞
Properties:
()
R XY t ←⎯
F
( )
→ G XY f or R XY ⎡⎣ n ⎤⎦ ←⎯
F
→ G XY F ( )
G XY ( f ) = G ( f ) or G ( F ) = G ( F )
*
YX XY
*
YX
(
Re G XY ( f )) and Re (G ( f )) are both even
YX
Two Random
Signals
Cross Power Spectral Density
Crosscorrelation
and CPSD of
Two Random
Signals
Cross Power Spectral Density
Two Random
Signals Plus
Narrowband
Interference
Cross Power Spectral Density
Crosscorrelation
and CPSD of
Two Random
Signals Plus
Narrowband
Interference
Measurement of Power Spectral
Density
A natural idea for estimating the PSD of an ergodic stochastic
CT process is to start with the definition,
⎛
( ( )) ⎞
2
F XT t
GX ( )
f = lim E ⎜
T →∞ ⎜ T
⎟
⎟
⎜⎝ ⎟⎠
( ( ))
2
F XT t
( )
Ĝ X f =
T
Unfortunately this approach yields an estimate whose variance
does not decrease with increasing time T .
Measurement of Power Spectral
Density
Another approach to estimating PSD is to first estimate
autocorrelation and then Fourier transform that estimate.
We can estimate autocorrelation from
T −τ
1
R̂ X ()
τ =
T −τ ∫ X (t ) X (t + τ ) dt , 0 ≤ τ << T
0