72
J. Kumar V. and K. A. Reddy
a PPG is the moving average method [9, 39]. The moving average method works well
only for a limited range of artefacts [60]. When the spectra of motion artefact and that
of the PPG signal overlap significantly, motion artefact becomes in-band noise [61].
In-band noise can be reduced to a large extant with the use of adaptive filters [62–67].
However, to apply adaptive filter technique, a reference signal that is required. The
reference signal must be strongly correlated with the signal but uncorrelated with
artefact. It is also possible to remove artefact with a reference signal that is strongly
correlated with artefact but uncorrelated with the signal. It is possible to obtain a
reference signal correlated with motion artefact by employing additional motionsensing hardware. Synthetic reference signal estimated from the artefact-free part of
a PPG signal [63] can be used for reducing motion artefact. The Masimo SET
® [67]
uses the fact that any motion artefact affects both the arterial pulsations and venous
blood volume change in a similar manner. Utilizing this fact, a reference signal is
extracted from the venous component of a PPG. Such a technique avoids the necessity
of extra hardware to sense motion artefact. A signal processing technique that uses
multi-rate filter bank and a matched filter provides better performance compared to
the moving average and adaptive filtering approaches [67]. Biological signals such
as the PPG are generally non-stationary and their properties change substantially
over time, mostly due to their dynamic nature. Thus use of time-frequency methods
like the wavelet transforms [68, 69] and smoothed pseudo-Wigner-Ville distribution [70] that are best suited for processing non-stationary signals have been applied
to process PPG signals to obtain significant improvements compared to traditional
approaches. An artefact reduction methodology that uses a physical artefact model
coupled with an inversion technique (nonlinear optical receiver) has been proposed
[71]. The model-based approach for reduction of motion artefacts requires an additional source-detector pair, resulting in the three-wavelength probe [72]. Exploiting
the independence between a PPG signal and motion artefact signal, it is possible
to reduce motion artefact using the Independent component analysis (ICA) technique. It has been shown that a third-order ICA applied on the time-derivative of
a PPG signal provides artefact suppression for pulse oximetry [73–75]. Compared
to ICA alone, ICA in conjunction with block interleaving and low pass filtering
provides better performance [73]. But a study showed that motion artefacts, arterial
and venous components of a PPG signal are not statistically independent [74]. It
was also shown that wavelet transform and adaptive filtering techniques introduce
phase shifts in the processed PPG signals and hence these methods have limited
application in restoring motion artefact corrupted PPG signals, especially when PPG
signals are used for estimation of heart rate (HR) and pulse transit time (PTT) [75]. A
motion artefact reduction method based on singular value decomposition (SVD), that
extracts clean artefact-free PPG signals from artefact riddled PPG signals preserving
all the essential morphological features required is described next [76].
J. Kumar V. and K. A. Reddy
a PPG is the moving average method [9, 39]. The moving average method works well
only for a limited range of artefacts [60]. When the spectra of motion artefact and that
of the PPG signal overlap significantly, motion artefact becomes in-band noise [61].
In-band noise can be reduced to a large extant with the use of adaptive filters [62–67].
However, to apply adaptive filter technique, a reference signal that is required. The
reference signal must be strongly correlated with the signal but uncorrelated with
artefact. It is also possible to remove artefact with a reference signal that is strongly
correlated with artefact but uncorrelated with the signal. It is possible to obtain a
reference signal correlated with motion artefact by employing additional motionsensing hardware. Synthetic reference signal estimated from the artefact-free part of
a PPG signal [63] can be used for reducing motion artefact. The Masimo SET
® [67]
uses the fact that any motion artefact affects both the arterial pulsations and venous
blood volume change in a similar manner. Utilizing this fact, a reference signal is
extracted from the venous component of a PPG. Such a technique avoids the necessity
of extra hardware to sense motion artefact. A signal processing technique that uses
multi-rate filter bank and a matched filter provides better performance compared to
the moving average and adaptive filtering approaches [67]. Biological signals such
as the PPG are generally non-stationary and their properties change substantially
over time, mostly due to their dynamic nature. Thus use of time-frequency methods
like the wavelet transforms [68, 69] and smoothed pseudo-Wigner-Ville distribution [70] that are best suited for processing non-stationary signals have been applied
to process PPG signals to obtain significant improvements compared to traditional
approaches. An artefact reduction methodology that uses a physical artefact model
coupled with an inversion technique (nonlinear optical receiver) has been proposed
[71]. The model-based approach for reduction of motion artefacts requires an additional source-detector pair, resulting in the three-wavelength probe [72]. Exploiting
the independence between a PPG signal and motion artefact signal, it is possible
to reduce motion artefact using the Independent component analysis (ICA) technique. It has been shown that a third-order ICA applied on the time-derivative of
a PPG signal provides artefact suppression for pulse oximetry [73–75]. Compared
to ICA alone, ICA in conjunction with block interleaving and low pass filtering
provides better performance [73]. But a study showed that motion artefacts, arterial
and venous components of a PPG signal are not statistically independent [74]. It
was also shown that wavelet transform and adaptive filtering techniques introduce
phase shifts in the processed PPG signals and hence these methods have limited
application in restoring motion artefact corrupted PPG signals, especially when PPG
signals are used for estimation of heart rate (HR) and pulse transit time (PTT) [75]. A
motion artefact reduction method based on singular value decomposition (SVD), that
extracts clean artefact-free PPG signals from artefact riddled PPG signals preserving
all the essential morphological features required is described next [76].