Scribd
Upload
64 views27 pages

Adaptive Filtering For Biomedical Applications

Adaptive filters for biomedical signals

Uploaded by

Ambareesh Veeraragavan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
64 views27 pages

Adaptive Filtering For Biomedical Applications

Adaptive filters for biomedical signals

Uploaded by

Ambareesh Veeraragavan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 27

ADAPTIVE

F I LT E R I N G F O R
BIOMEDICAL
A P P L I C AT I O N S

MUKIL KUMAR S – 110119066


A AT H I T H YA N K K – 110 119002
SUNDAR R – 110119116
SIVADHARSHINI – 110119108
W H AT I S A S TAT I O N A RY
SIGNAL?
• A signal is said to be stationary if it’s frequency or
spectral contents is not changing with time.
W H AT I S A N O N - S TAT I O N A RY
SIGNAL?
• A signal is said to be non-stationary if it’s frequency or
spectral contents is changing with time.
W H AT I S C O R R E L AT I O N ?
• In general, correlation describes the mutual relationship which
exists between two or more things.
• Correlation between signals indicates the measure up to which
the given signal resembles another signal.
• In other words, if we want to know how much similarity exists
between the signals 1 and 2, then we need to find out the
correlation of Signal 1 with respect to Signal 2 or vice versa.
CONVOLUTION VS
C O R R E L AT I O N
• It is well-known that the output of a linear time (or space) invariant system can be
expressed as a convolution between the input signal and the system impulse response
function.

• Correlation of two signals is the convolution between one signal with the functional
inverse version of the other signal.
CONVOLUTION VS
C O R R E L AT I O N
( C O N T D . . )

AUTO-
C O R R E L AT I O N

CROSS-
C O R R E L AT I O N
O P T I M A L F I LT E R I N G
• Optimal filtering is a means of adaptive extraction of a weak desired
signal in the presence of noise and interfering signals.
• The common methods for designing optimal filter under constrains in the
transfer function are, the Wiener method, the discrete-time Fourier
transform (DFT) method, the least mean square (LMS) method, and
the recursive least squares (RLS) method.
• Wiener filter are mean square error – optimal stationary linear filter for signal
degraded by additive noise. Wiener filter is a filter used to produce an
estimate of a desired or target random process by linear time-
invariant (LTI).
• A FIR filter whose output y[n] best approximates the desired signal d[n] in the
sense that the mean square norm of the error is minimised is called the
optimum FIR Wiener filter.

• Optimum filter is used in the detection system usually it contains a pre-


detection filter whose function is to enhance the strength of the signal
relative to that of the noise.
W H Y T H E A D A P T I V E F I LT E R I N G
REQUIRED?
• Adaptive filters for interference removal scenario:
 Power line interference with varying power frequency
 ECG signal of fetus & mother
 VAG(vibroarthrogram) and VMG(vibromyogram)signal.
A D A P T I V E F I LT E R I N G
• Adaptive filtering methods in general scenarios are usually required to
change the type of filtering dynamically, and hence there are perfect for non-
stationary signals
• A basic representation of an adaptive filtering circuit would be:
A D A P T I V E F I LT E R I N G ( C O N T D … )

• In the previous slide, the signal x(n) is the signal extracted from a patient’s
body that must be filtered. This signal consists of two components, one is the
desired component (i.e.) the signal we want to measure: v(n), and the other is
the noise component which we want to remove: m(n)
• The signal r(n) is the reference input signal which is fed to the adaptive filter
model to get an output signal y(n). The model is constructed in such a way
that the output signal y(n) is as close as possible to the noise component of
x(n) (i.e.) m(n)
• Now by subtracting y(n) from x(n) the error signal e(n) will be close to the
desired component of the measured signal (i.e.) v(n) and in this way we filter
out the noise from the measured signal
A D A P T I V E F I LT E R I N G ( C O N T D … )

• But there are 2 important conditions for the adaptive filter to function
properly
• The first condition is that the desired component v(n) and the noise
component m(n) must be uncorrelated
• The second condition is that the reference signal r(n) must be uncorrelated
with the desired component v(n), but it should be correlated with the noise
component m(n).
• So in the end, our desired signal cam be estimated as:

• Where is the approximate desired component of the input signal x(n). We use
here since no filter is perfect
A D A P T I V E F I LT E R I N G ( C O N T D … )

• Now the aim of adaptive filtering, which is based on optimal filtering is to reduce
the mean square error, so for the adaptive filter model:

• The final term becomes zero since v(n) is uncorrelated with y(n) and has zero mean
A D A P T I V E F I LT E R I N G ( C O N T D … )

• Now since our aim is to minimize the noise, we try and minimize the mean
square error
A D A P T I V E F I LT E R I N G ( C O N T D … )

• Now that we’ve reduced the mean square error, we will take a closer look at
the adaptive filter
A D A P T I V E F I LT E R I N G ( C O N T D … )

• Now from the previous slide, we can represent all those elements in an Mx1
matrix form for computational ease
Least Mean
Squares (LMS)
TYPES OF
ADAPTIVE
F I LT E R I N G
Recursive Least
Squares (RLS)
L M S A D A P T I V E F I LT E R I N G

The weights in the above equation are updated using steepest descent algorithm:

µ is the step size and delta is the gradient of the squared error with respect to the weight

The weights are updated at each time


step to ensure optimal filtering of the
non-stationary signal
A D VA N TA G E S O F L M S
• Simple yet effective
• It performs just M multiplications and M additions
• Hence it is very useful especially at higher sampling frequencies as it is
computationally light
• Imposes only one condition of inputs being uncorrelated over time and is
hence mathematically robust to different types of signals
• Stable for a band of values of µ (must be tuned well)
• µ is dynamically changed to avoid slow updating of weights initially (ie it is
kept high) and to avoid noisy convergence at the end (ie it is kept low)
R L S A D A P T I V E F I LT E R I N G
λ is the forgetting factor
It assigns more weightage to recent errors

This is the equation used to


update weights (where r is the reference
input and x is the actual input)

The parameters phi and theta are written in a


recursive style to frame the update equations
for each

Equation 3 gives the weight equation which is


computationally heavy due to matrix
inversions and vector multiplications
RLS CONTD.
Using the ABCD Lemma, we
represent phi inverse in this
fashion

For ease of notation, we coin two


variables P and k and rewrite
equation 4

Cross-multiplying the
denominator in the previous k
equation and rearranging terms
RLS CONTD.

If the weight updating process


happens slowly, then alpha can be
approximated to the present error
which is v(n)
In the above weight equation, we only have to compute k
as everything else is known to us (hence decreasing the
computation required)
A P P L I C AT I O N S
OF ADAPTIVE
F I LT E R S
• Now let us look at an application

• The image shown here represents a system


for data collection for ECG signals

• This particular setup has both wet electrodes


and capacitive electrodes with an
accelerometer

• The reasons for this setup will be explained in


the upcoming slides

• The aim in this application is to remove


motion artifacts
A P P L I C AT I O N S
OF ADAPTIVE
F I LT E R S ( C O N T D … )

• This setup represents a classical adaptive


filtering method for motion artifacts

• Here the signal from the accelerometer is used


as the reference signal for the adaptive filter

• The measured signal is also passed to a low


pass filter to remove and non overlapping
noises and then the adaptive filter is used to
remove overlapping noises

• By using the accelerometer signal which is


corelated with motion artefacts as a reference
signal, we are able to remove motion artefacts
A P P L I C AT I O N S
OF ADAPTIVE
F I LT E R S ( C O N T D … )

• In the image shown the input signal is the


signal extracted from a patient ()

• This signal is passed through two different


types of filters (low pass and band pass) to
remove the non-overlapping noise and to
attain a reference signal

• Then these signals are fed through the


summer and the adaptive filter to attain the
required results
A P P L I C AT I O N S O F A D A P T I V E
F I LT E R S ( C O N T D … )

Before filtering After filtering


REFERENCES
• https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=8573141
• https://www.researchgate.net/publication/286219429_Use_of_power-line_inter
ference_for_adaptive_motion_artifact_removal_in_biopotential_measurements
NPTEL Courses:
• https://www.youtube.com/watch?v=YG7PPTPr-YM
• https://www.youtube.com/watch?v=zqTWehjIh-8
• https://www.youtube.com/watch?v=MJApbY2XW6g

You might also like