ECG Signal Analysis - ECG Acquisition and Automated Remote Processing - page 19

multiplications which result in long computation time. The method described in
[ 12 ] enhances the computational efficiency by reducing the number of impulse
response coefficients and using the symmetry of the impulse response. The pre-
scribed method reduced the number of coefficients by a factor of 5 compared to
conventional NRFIR filter and halved the number of multiplications. Application
of adaptive filtering for noise cancelation is described in [ 13 ]. Different types of
adaptive filter structures, viz., basic adaptive filter, least mean square (LMS)
algorithm, and adaptive recurrent filter (ARF), are applied for reduction in power
line interference, EMG, motion artifact, and BW in ambulatory ECG monitoring.
A cascaded adaptive filter for removal of BW is described [ 14 ], and its perfor-
mance is compared with cubic spline approach [ 15 ] while applied to MIT-BIH
arrhythmia database. The proposed method used an adaptive zero frequency notch
filter followed by adaptive impulse correlated filter (AICF). Conventional BW
removal methods using band-pass filter (0.05-100 Hz) suffer from the disadvan-
tages that they distort the ECG at two distinct frequencies, viz., 0 Hz (ideal
baseline voltage) and 0.8 Hz [ 16 ]. A new adaptive filter, which is a combination of
time-sequenced adaptive filter (TSAF) and AICF, is proposed in [ 17 ]. Morpho-
logical operators have been widely used in the signal and image processing
domains because of their robust and adaptive performance in extracting the shape
information in addition to their simple and quick set computation. Morphological
filters are based on some mathematical structures which capture the structural
property of the signal by applying a set of structural element on the dataset.
A modified Morphological operator-based ECG filtering technique is described in
[ 18 ]. Adaptive noise cancelation provides a means for a no priory knowledge of
the signal or the noise characteristics. In this technique [ 19 ], two input signals are
fed to the noise canceler block. The first input is called the 'primary,' containing
the corrupted signal, i.e., signal plus noise. The second one is called 'reference,'
contains the noise correlated with some way with the primary noise. This noise is
filtered to make a close replica with the primary noise. The outputs are subtracted
to produce the noise-free signal. An adaptive Kalman filtering technique is pro-
posed in [ 20 ] for baseline removal from the ECG signal. The ECG is simulated as
a piecewise linear triangular function, smoothed by third-order Savitzky-Golay
FIR filter. The baseline is generated by a second-order polynomial.
PCA is a statistical tool that decorrelates the different signal components from
ECG data [ 21 ], while ICA considers the noise signals as independent entity in the
ECG signal. [ 22 ] deals with PCA followed by different versions of ICA application
for ECG segmentation and QRS detection, noise reduction in CSE database. In
[ 23 ], a modification of classical PCA, named 'projection pursuit approach,' is used
for ECG enhancement with the objective of analysis of ECG beat variability.
An artificial neural network (ANN) is an interconnected chain of computational
units (nodes) that simulates a human brain for its ability of learning from the data.
A neural net is trained with known similar datasets which it used to gather
knowledge about the data. With an unknown dataset, the network responds better
from its acquired knowledge. An application of NN-based adaptive filter in
wavelet domain for noise removal is reported in [ 24 ]. A wavelet-based optimal