Scribd
Upload
23 views24 pages

Cepstrum and Homomorphic Filtering

The document discusses various signal processing techniques, including cepstrum analysis, homomorphic filtering, and spectral analysis of ECG signals. It explains concepts such as power spectral density, auto and cross-correlation, and the use of complex cepstrum for deconvolution in applications like seismic signal processing. Additionally, it highlights the clinical significance of cross spectral coherence in detecting brain midline shifts using EEG data.

Uploaded by

46.Swetha M
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
23 views24 pages

Cepstrum and Homomorphic Filtering

The document discusses various signal processing techniques, including cepstrum analysis, homomorphic filtering, and spectral analysis of ECG signals. It explains concepts such as power spectral density, auto and cross-correlation, and the use of complex cepstrum for deconvolution in applications like seismic signal processing. Additionally, it highlights the clinical significance of cross spectral coherence in detecting brain midline shifts using EEG data.

Uploaded by

46.Swetha M
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 24

Cepstrum &

Homomorphic
Filter
Power Spectral Density

Auto correlation by Ensemble


average
rxx(k) = E[x(n)x(n-k)]
Auto correlation by Time average
Cross Power Spectral
Density

Cross correlation by Ensemble


average
rxy(k) = E[x(n)y(n-k)]
Cross correlation by Time
average
Clinical Lead Placement
 Einthoven Limb Leads:

28/01/2021 6
Spectral and Cross-spectral analysis
of the ECG

28/01/2021 7
PSD of standard 12 leads of 10 seconds
of an ECG of sinus rhythm.(256 point
Welch periodogram with Hamming
window with 256 points overlapped

28/01/2021 8
The Spectral content for each lead is
highly similar regardless of the lead
configuration
28/01/2021 9
Cross Spectral
Coherence (CSC)

2
Cxy | Sxy |
 
1
| Sxx | | Syy |
 where Sxx Syy are the power spectral estimates of x and y
 Sxy are the power spectral estimate of x and y

 CSC has values between 0 and 1


CSC between any pair of leads is
greater than 0.9 at most
physiologically significant frequencies
( 1 to 10 Hz)
C1xy between Lead I and Lead II
C2xy between adjacent 10
seconds segments of Lead I
Detection of Brain Midline shift
 EEGs from right and left
hemispheres {x(n) and y(n) }

 CSC values give the amount of


midline shift
CEPSTRUM
Consider a real sequence x(n) having a z
transform X(z).

We assume that x(n) is a stable sequence


so that X(z) converges on the unit circle.

The complex cepstrum of the sequence


x(n) is defined as the sequence c(n), which is
the inverse z-transform of Cx (z), where
Cx(z) = ln X(z)

Complex Cepstrum , c(n) = Z-1 [Cx (z)]


CEPSTRUM
The complex cepstrum exists if Cx(z) converges
in the annular region r1< |z| <r2, where 0<r1<1 and r2 >1.

Within this region of convergence, Cx (z) may be


represented by the Laurent series

Cx (z) = ln X(z) =  cx (n) z-n
n=-

where cx (n)= 1/2j c ln X(z) zn-1 dz


c is a closed contour encircling the origin
and lies within the region of convergence.
The complex cepstrum sequence {cx(n)}
is stable.
Furthermore,if the complex cepstrum
exists, Cx(z) converges on the unit circle and
hence we have

Cx() = ln X() =  cx(n)e-jn
n= - 

where {cx(n)} is the sequence obtained from the


inverse Fourier transform of ln X() that is,
1 
cx(n) = ----  ln X()ejn d
2 -
If we express X() in terms of its magnitude and
phase, say
X() = |X()|ej()
ln X() = ln|X()| + j()

We obtain the complex cepstrum in the form


1 
cx(n) = ----  [ln |X()|+j () ] e jn d
2 -
cx(n) = cm(n) + j c(n)
1 
cm(n) = ----  ln |X()|ejn d
2 - 
1 
c(n) = ----   ()ejn d
2 -
Homomorphic Deconvolution
The complex cepstrum is a useful tool for
performing deconvolution in some applications
such as seismic signal processing.
x(n) H(Z) y(n)= x(n) * h(n)
Time
h(n) domain
Y(z) = X(z)H(z)

The logarithm of Y(z) is


Cy(z) = ln Y(z) = ln X(z) + ln H(z)
= Cx (z) + Ch (z)
Consequently, the complex cepstrum of the
output sequence {y(n)} is expressed as the sum of the
cepstrum of {x(n)} and {h(n)},that is, Cepstral
domain
cy(n) = cx(n) + ch(n)
System is known as Homomorphic system
Homomorphic System

Y(Z) lnY(z) Cy(n)


y(n)
Complex Inverse
Z-transform
Logarithm z-transform

Suppose ch(n) has its main components in the vicinity of


small values of n and cx(n) at large values of n
Then we can separate ch(n) and cx(n) by using
appropriate windows
ch(n) = cy(n) wlp(n)

Low Pass Window


wlp(n) =  1, |n|  N1
 0, Otherwise

cx(n) = cy(n) whp(n)

whp(n) =  0, |n|  N1
High Pass Window
 1, |n| > N1
Inverse Homomorphic System

cx(n) Cx(Z) X(Z) x(n)


Complex Inverse
Z-transform Exponential z-transform
ch(n)
Ch(Z) h(n)
H(Z)

Inplace of Z-transform we can use discrete Fourier


transform and its inverse.
Homomorphic Filtering
Example- Voice Processing u(n) Vocal chord
x(n)=u(n)*h(n) h(n) Vocal tract
X(k) ln X(k) cx(n)
DFT ln ( IDFT
)

ch(n)
Low Pass window

cu(n) High Pass Window

Cu(k) u(n)
DFT Exp( ) IDFT

Ch(k) h(n)
DFT Exp( ) IDFT

You might also like