SIGNAL PROCESSING

Wearability issues -physical shape and placement of

sensor, Technical challenges - sensor design, signal
acquisition, Constraint on sampling frequency for
reduced energy consumption, light weight signal
processing, Rejection of irrelevant information, Data
mining
 Wearability Issues
• Design for wearable BSNs focuses on specific
and important issues for developing
wearable computing systems that take into
account
• the physical shape of the sensors
and their active relationship with the human
form.
• Design for wearability requires unobtrusive
sensor node placement on the human body
based on application-specific criteria.
• Criteria for placement can vary with the needs of
functionality and convenience.
• Functionality criteria constrains node placement
to regions where relevant data can be sensed.
The number of nodes required to capture all
relevant data can vary based on the quality
of information sensed at individual locations.
• Convenience criteria include:
(1) physical interference with movement,
(2) difficulty in removing and placing nodes,
(3) social and fashion concerns,
(4) frequency and difficulty of maintenance
(charging and cleaning)
• For example, in continuous healthcare monitoring, patients
will be expected to charge the sensors or replace the
batteries on a regular basis, as they do with cell
• phones and other electronics.
• However, the frequent need to charge and the bulk of
• the battery can frustrate the users, causing them to no
longer wear the sensors.
• Furthermore, batteries are the heaviest component in the
system. By decreasing power usage, the size and weight of
each sensor node can decrease, thus increasing
• patient comfort and device wearability.
• This makes energy usage a primary constraint in
designing BSNs, limiting everything from data sensing
rates and link bandwidth, to node size and weight.
• Thus, one of the important goals in designing BSNs is to
minimize energy consumption while preserving an
acceptable quality of service.
• Energy consumption can be decreased by lower
sampling frequency, decreasing processing power, and
simplifying signal processing.
• Another effective technique is deactivating nodes that
are unnecessary for specific tasks
 Signal Processing for ECG signal
acquisition
• ECG Signal acquisition
• Noises
– Baseline wandering (Motion artifact)
– Powerline Interference
– Physiological interference (EMG, Maternal ECG in
fetal ECG)
 Baseline wandering
• Baseline wandering is a low frequency effect
• To eliminate baseline wandering, a high pass
filter with cutoff frequency of 0.5Hz is used.

FIR High pass filter

ECG Signal Fc=0.5 Hz Filtered ECG
x(n) h(n) signal
y(n)=x(n)*h(n)
 Powerline interference
• 50Hz noise in ECG signal
• Eliminated by notch filter with fc=50Hz

FIR Notch filter

ECG Signal Fc= 50 Hz Filtered ECG
x(n) h(n) signal
y(n)=x(n)*h(n)
 Muscular noise
• High frequency noise

FIR Low pass filter

ECG Signal Fc= 100 Hz Filtered ECG
x(n) h(n) signal
y(n)=x(n)*h(n)
High frequency noise
 FIR Bandpass filter • FIR Notch filter
ECG Signal
x(n)
FL =0.5 Hz fu= 100 • Fc= 50 Hz
h(n) • h(n)
 Optimum Filters
• The desired signal ( speech, EEG, Radar signal)
may be noisy and distorted.
• Classical filters such as lowpass, highpass,
bandpass will not be optimum in finding the best
estimate of the signal
• Wiener pioneered research in design of filter that
would produce the optimum estimate of a signal
from a noisy measurement or observation
Wiener filter
 Adaptive Filters
• Filters with fixed characteristics (tap weights or coefficients are
suitable when the characteristics of the signal and noise (random or
structured) are stationary and known.

• Design of frequency-domain filters requires detailed knowledge of

the spectral contents of the signal and noise.

• Such filters are not applicable when the characteristics of the signal
and/or noise vary with time,
• That is, when they are nonstationary. They are also not suitable
when the spectral contents of the signal and the interference overlap
significantly.
 Adaptive Filters
• Filter that can learn and adapt to the characteristics of
the interference, estimate the interfering signal, and
remove it from the mixture to obtain the desired signal.

• This requires the filter to automatically adjust its impulse

response (and hence its frequency response) as the
characteristics of the signal and/or noise vary.
Selection of Adaptive Filtering
Adaptive filter

23
 Model of adaptive filter Weight updation for
over signal adaptive filter

Estimate of New filter Old Update of

Filter Shifted form
desired coefficients coefficient filter filter
of observed
signal coeffici coefficient in
varying in signal
ent time
time
Solution of the filter coefficients – Wiener-
Hopf equations
Autocorrelation of Cross-correlation of
observed signal desired and observed
Vector of Filter signal
coefficients 24
 Least Mean Square (LMS) algorithm
Principle: To reduce the mean square of error

Mean

Square of
Weight updation estimation error
New tap-weight
vector equation Parameter for
stability and
convergence

Old tap-weight Reference

vector Error signal
signal 25
Adaptive
filters in
cascade to
remove power
line, ECG and
EOG artifacts
from EEG
signal

26
EEG signal before and after adaptive filtering

EEG signal with

power line,
ECG and EOG
artifacts

EEG signal without

power line,
ECG and EOG artifacts

27
QRS Detection using Pan Tompkins
algorithm
Moving Average Filters
Magnitude and Phase response of the Hanning Filter
ECG Signal with high frequency noise after filtering
by the 8 point MA filter
• Highpass filtered signal= Original signal-
lowpass filtered signal
Signal Processing to
compute Heart rate from
PPG signal
Power spectrum of PPG signal
Power line interference and motion artifacts in
PPG signal
Preprocessing of PPG signal
 Heart rate = 60/ Peak to peak interval

Artery stiffness

h- subject height
Pulse oximeter