Electronics 2014, 3, 282-302; doi:10.3390/electronics3020282
`
`OPEN ACCESS
`
`electronics
`
`ISSN 2079-9292
`www.mdpi.com/journal/electronics
`
`Review
`Wearable Photoplethysmographic Sensors—Past and Present
`Toshiyo Tamura 1,*, Yuka Maeda 2, Masaki Sekine 1 and Masaki Yoshida 1
`
`1 Osaka Electro-Communication University, Faculty of Biomedical Engineering /18-8, Hatsu-Cho,
`Neyagawa, Osaka 572-8530, Japan; E-Mails: m-sekine@isc.osakac.ac.jp (M.S.);
`yoshida@isc.osakac.ac.jp (M.Y.)
`2 Faculty of Engineering, Information and Systems, University of Tsukuba/1-1-1, Tennodai,
`Tsukuba, Ibaraki 305-8573, Japan; E-Mail: maeda@iit.tsukuba.ac.jp
`
`* Author to whom correspondence should be addressed; E-Mail: tamurat@isc.osakac.ac.jp;
`Tel.: +81-72-824-1131; Fax: +81-72-824-0014.
`
`Received: 25 February 2014; in revised form: 15 April 2014 / Accepted: 18 April 2014 /
`Published: 23 April 2014
`
`Abstract: Photoplethysmography (PPG) technology has been used to develop small,
`wearable, pulse rate sensors. These devices, consisting of infrared light-emitting diodes
`(LEDs) and photodetectors, offer a simple, reliable, low-cost means of monitoring the
`pulse rate noninvasively. Recent advances in optical technology have facilitated the use of
`high-intensity green LEDs for PPG, increasing the adoption of this measurement
`technique. In this review, we briefly present the history of PPG and recent developments in
`wearable pulse rate sensors with green LEDs. The application of wearable pulse rate
`monitors is discussed.
`
`Keywords: photoplethysmography; pulse rate; reflectance; transmittance; green light;
`infrared light; adaptive filter; least mean square algorithm
`
`1. Introduction
`
`It is important to monitor the perfusion of the circulation. The most important cardiopulmonary
`parameter is blood pressure, but monitoring it is complicated. A second important parameter is blood
`flow, which is related to blood pressure. We can monitor the blood perfusion in large vessels using
`ultrasound devices, but it is not practical to use these routinely. Several devices for monitoring blood
`MASIMO 2006
`Apple v. Masimo
`IPR2020-01526
`
`

`

`283
`
`Electronics 2014, 3
`
`perfusion have been developed [1], but, unfortunately, it is difficult to find a practical device.
`However, the perfusion of blood flow and blood pressure can be determined easily using a pulse
`rate monitor.
`Wearable pulse rate sensors based on photoplethysmography (PPG) have become increasingly
`popular, with more than ten companies producing these sensors commercially. The principle behind
`PPG sensors is optical detection of blood volume changes in the microvascular bed of the tissue. The
`sensor system consists of a light source and a detector, with red and infrared (IR) light-emitting diodes
`(LEDs) commonly used as the light source. The PPG sensor monitors changes in the light intensity via
`reflection from or transmission through the tissue. The changes in light intensity are associated with
`small variations in blood perfusion of the tissue and provide information on the cardiovascular system,
`in particular, the pulse rate. Due to the simplicity of this device, wearable PPG pulse rate sensors have
`been developed. This review describes the basic principles of PPG, previous and current developments
`in wearable pulse rate monitors with a light source, and the elimination of motion artifacts.
`
`2. Photoplethysmography (PPG)
`
`2.1. Principle of PPG
`
`The principle of PPG has been reviewed previously [2–4], and is explained briefly here. Light
`travelling though biological tissue can be absorbed by different substances, including pigments in the
`skin, bone, and arterial and venous blood. Most changes in blood flow occur mainly in the arteries and
`arterioles (but not in the veins). For example, arteries contain more blood volume during the systolic
`phase of the cardiac cycle than during the diastolic phase. PPG sensors optically detect changes in the
`blood flow volume (i.e., changes in the detected light intensity) in the microvascular bed of tissue via
`reflection from or transmission through the tissue.
`Figure 1 shows an example of a photoplethysmographic waveform, consisting of direct current
`(DC) and alternating current (AC) components. The DC component of the PPG waveform corresponds
`to the detected transmitted or reflected optical signal from the tissue, and depends on the structure of
`the tissue and the average blood volume of both arterial and venous blood. Note that the DC
`component changes slowly with respiration. The AC component shows changes in the blood volume
`that occurs between the systolic and diastolic phases of the cardiac cycle; the fundamental frequency of
`the AC component depends on the heart rate and is superimposed onto the DC component.
`
`Figure 1. Variation in light attenuation by tissue.
`
`
`
`
`
`

`

`Electronics 2014, 3
`
`2.2. Light Wavelength
`
`284
`
`The interaction of light with biological tissue can be quite complex and may involve scattering,
`absorption and/or reflection. Anderson and Parrish examined the optical characteristics and penetration
`depth of light in human skin [5]; within the visible region, the dominant absorption peak corresponded
`to the blue region of the spectrum, followed by the green-yellow region (between 500 and 600 nm)
`corresponding to red blood cells. The shorter wavelengths of light are strongly absorbed by melanin.
`Water absorbs light in the ultraviolet and longer IR regime; however, red and near-IR light pass easily.
`As a result, IR wavelengths have been used as a light source in PPG sensors.
`Blood absorbs more light than the surrounding tissue. Therefore, a reduction in the amount of blood
`is detected as an increase in the intensity of the detected light. The wavelength and distance between
`the light source and photodetector (PD) determine the penetration depth of the light. Green light is
`suitable for the measurement of superficial blood flow in skin. Light with wavelengths between 500
`and 600 nm (the green-yellow region of the visible spectrum) exhibits the largest modulation depth
`with pulsatile blood absorption. IR or near-IR wavelengths are better for measurement of deep-tissue
`blood flow (e.g., blood flow in muscles). Thus, IR light has been used in PPG devices for some time [6].
`However, green-wavelength PPG devices are becoming increasingly popular due to the large intensity
`variations in modulation observed during the cardiac cycle for these wavelengths [7–9].
`A green LED has much greater absorptivity for both oxyhaemoglobin and deoxyhaemoglobin
`compared to infrared light. Therefore, the change in reflected green light is greater than that in
`reflected infrared light when blood pulses through the skin, resulting in a better signal-to-noise ratio
`for the green light source.
`Several green-light-based photoplethysmographs are available commercially. For example, MIO
`Global has developed the MIO Alpha in cooperation with Philips; this measures the electrocardiogram
`(ECG) with 99% accuracy, even while cycling at speeds of up to 24 kmph [10]. For daily use, Omron
`has developed a green light pulse rate monitor (HR-500U, OMRON, Muko, Japan).
`Furthermore, the use of video cameras using the signal based on the red green blue (RGB) colour
`space has been considered, as shown in Section 3.3. The green signal was found to provide the
`strongest plethysmographic signal among camera RGB signals [11,12]. Haemoglobin absorbs green
`light better than red and green light penetrates tissue to a deeper level than blue light. Therefore, the
`green signal contains the strongest plethysmographic signal.
`
`2.3. Reflected and Transmitted Signals
`
`The wearable PPG has two modes—transmission and reflectance—as shown in Figure 2. In
`transmission mode, the light transmitted through the medium is detected by a PD opposite the LED
`source, while in reflectance mode, the PD detects light that is back-scattered or reflected from tissue,
`bone and/or blood vessels.
`
`
`
`

`

`Electronics 2014, 3
`
`
`285
`
`Figure 2. Light-emitting diode (LED) and photodetector (PD) placement for transmission-
`and reflectance-mode photoplethysmography (PPG).
`
`
`
`The transmission mode is capable of obtaining a relatively good signal, but the measurement site
`may be limited. To be effective, the sensor must be located on the body at a site where transmitted
`light can be readily detected, such as the fingertip, nasal septum, cheek, tongue, or earlobe. Sensor
`placement on the nasal septum, cheek or tongue is only effective under anesthesia. The fingertip and
`earlobe are the preferred monitoring positions; however, these sites have limited blood perfusion. In
`addition, the fingertip and earlobe are more susceptible to environmental extremes, such as low
`ambient temperatures (e.g., for military personnel or athletes in training). The greatest disadvantage is
`that the fingertip sensor interferes with daily activates.
`Reflectance mode eliminates the problems associated with sensor placement, and a variety of
`measurement sites can be used (as discussed in the following section). However, reflection-mode PPG
`is affected by motion artifacts and pressure disturbances. Any movement, such as physical activity,
`may lead to motion artifacts that corrupt the PPG signal and limit the measurement accuracy of
`physiological parameters. Pressure disturbances acting on the probe, such as the contact force between
`the PPG sensor and measurement site, can deform the arterial geometry by compression. Thus, in the
`reflected PPG signal, the AC amplitude may be influenced by the pressure exerted on the skin.
`Reflectance PPG sensors such as the MaxFast (Nellcor™, Mansfield, MA, USA) have been used
`clinically to measure continuous oxygen saturation non-invasively. Anecdotally, it gives false-positive
`readings occasionally. Further research is needed in this area.
`
`3. Factors Affecting PPG Recordings
`
`Previous research has identified several factors that affect PPG recordings, including the measurement
`site (i.e., probe attachment site), the contact force, mechanical movement artifacts, subject posture, and
`breathing, as well as ambient temperature. This chapter briefly discusses these factors.
`
`3.1. Measurement Site of Probe
`
`The location of the LED and PD is an important design issue that affects the signal quality and
`robustness against motion artifacts. Therefore, suitable measurement sites must be located to optimize
`sensor performance. PPG sensors are commonly worn on the fingers due to the high signal amplitude
`that can be achieved in comparison with other sites [13]. However, this configuration is not well suited
`to pervasive sensing, as most daily activities involve the use of the fingers.
`In recent years, different measurement sites for PPG sensors have been explored extensively,
`including the ring finger [14], wrist [15,16], brachia [9,17,18], earlobe [19–21], external ear
`cartilage [22–24], and the superior auricular region [25–27]. In addition, the esophageal region has
`been used in clinical practice [28–30]. Commercial clinical PPG sensors commonly use the finger,
`earlobe and forehead [31]. In addition, use of a glass-type wireless PPG has been examined [32].
`
`
`

`

`Electronics 2014, 3
`
`
`286
`
`The perfusion values of 52 anatomical sites in healthy subjects showed that the fingers, palm, face,
`and ears offer much higher perfusion values compared with other measurement sites [33]; the transmitted
`PPG signal amplitude from the earlobe provides the largest perfusion value. In addition, earlobe
`sensors are easy to fabricate, and have become popular as pulse rate monitors (Table 1). However, a
`spring-loaded ear-clip, although effective, can become painful over extended monitoring periods.
`There was little improvement in the wearable earlobe PPG sensor design until the development of
`micro-electromechanical system (MEMS) technology. MEMS facilitated the fabrication of a
`lightweight, comfortable, fully integrated, self-contained sensor earpiece. For example, an earring PPG
`sensor with magnetic attachment to the earlobe was developed that allowed good contact for
`monitoring during physical activity [22].
`
`Table 1. Key features of a wearable ear photoplethysmography (PPG) device.
`
`In-mount
`Features
`Year published 2009 [20]
`Auditory
`canal
`Otoplastic
`insertion
`
`Sensing site
`
`Probe
`attachment
`Wireless
`communication
`Motion
`cancellation
`
`CUHK
`2008 [26]
`Inferior
`auricle
`
`e-AR
`2007 [19]
`
`Superior auricle
`
`Imperial
`2009 [27]
`Superior
`auricle
`
`Earhook
`
`Earhook
`
`Tape
`
`MIT
`2010 [22]
`
`Earlobe
`
`Magnetic
`earring
`
`MIT
`2012 [24]
`External ear
`cartilage
`
`Earphone
`
`Pulsear
`2004 [23]
`External ear
`cartilage
`Earcup
`headphones
`
`Yes
`
`Yes
`
`No
`
`No
`
`Yes
`
`Yes
`
`None
`
`None
`
`Passive motion
`cancellation
`
`None
`
`
`cancellation
`
`Automatic noise Automatic noise
`cancellation
`
`No
`
`PCA
`
`Samsung
`2009 [21]
`
`Earlobe
`
`Spring-loaded
`clip
`
`No
`
`Automatic noise
`cancellation
`
`Earphone/earbud PPG sensors are also available and provide greater comfort for the user. In this
`design, a reflective photosensor is embedded into each earbud, as shown in Figure 3. The sensor
`earbuds are inserted into the ear and positioned against the inner side of the tragus to detect the amount
`of light reflected from the subcutaneous blood vessels in the region. The PPG sensor earbuds look and
`work like a regular pair of earphones, requiring no special training for use [24].
`
`Figure 3. Earpiece PPG sensor (with permission [24]).
`
`
`
`
`
`

`

`Electronics 2014, 3
`
`
`287
`
`A headset with an ear-clip, transmission-type PPG sensor allows continuous, real-time monitoring
`of heart rate while listening to music during daily activities. In addition, the proposed headset is
`equipped with a triaxial accelerometer, which enables the measurement of calorie consumption and
`step-counting. However, over the course of a variety of daily activities (e.g., walking, jogging, and
`sleeping), the PPG sensor signal may become contaminated with motion artifacts [20].
`The most common commercially available PPG sensor is based on finger measurement sites. Finger
`sites are easily accessed and provide good signal for PPG sensor probes. For example, a ring sensor
`can be attached to the base of the finger for monitoring beat-to-beat pulsations. Data from the ring
`sensor are sent to a computer via a radiofrequency transmitter, as shown in Figure 4. To minimize
`motion artifacts, a double ring design was developed to reduce the influence of external forces,
`acceleration and ambient light, and to hold the sensor gently and securely to the skin, so that the blood
`circulation in the finger remained unobstructed. Experiments have verified the resistance of the ring
`sensor to interfering forces and acceleration acting on the ring body. Benchmark testing with
`FDA-approved PPG and ECG sensors revealed that the ring sensor is comparable in the detection of
`beat-to-beat pulsations, despite disturbances [14,34].
`
`Figure 4. PPG ring sensor.
`
`
`
`Wristwatch-type pulse oximetry and blood pressure sensors have been developed and
`commercialized by several companies. These devices, although much easier to wear, are not usually
`used in clinical settings, due to several technical issues. However, a novel PPG array sensor module
`with a wristwatch-type design has been developed. The proposed module measures the PPG signal
`from the radial artery and the ulnar artery of the wrist, whereas previous methods obtained signals
`from the capillaries in the skin. Phototransistors and IR-emitting diodes were placed in an array format
`to improve the PPG signal sensitivity and level of accuracy. Various arrays were considered for
`optimization. A conductive fiber wristband was used to reduce external noise. In the experiments, the
`proposed module was assessed and compared with the commercially available product produced by
`BIOPAC [16].
`A reflective brachial PPG sensor has also been examined. Although the pulse amplitude is lower
`than those from the finger and earlobe, the PPG pulse waveforms from regions in the vicinity of a
`human artery could be detected and measured easily [17].
`Forehead sensors have shown greater sensitivity to pulsatile signal changes under low perfusion
`conditions, compared with other peripheral body locations [31]. The thin-skin layer of the forehead,
`coupled with a prominent bone structure, helps to direct light back to the PD. Sensor placement
`on the forehead has been shown to result in decreased motion artifacts during certain types of
`physical activity.
`
`
`
`

`

`Electronics 2014, 3
`
`
`288
`
`Glass-type PPG sensors have also been investigated. A reflectance mode-PPG sensor, equipped
`with an accelerometer for detecting kinetic signals and a wireless controller for transmitting both
`signals to the receiver installed on the frame of the glass, was used to obtain PPG data from a user
`performing alternating sitting and sit-to-stand motions—the correlation between the peak-to-peak
`intervals in the signals of a BIOPAC device and the developed device was 97.5% and 87% for sitting
`and the sit-to-stand motion, respectively, given 100% data transmission without error [32].
`
`3.2. Probe Contact Force
`
`In reflectance-type PPG, the PPG signal waveform may be affected by the contact force between
`the sensor and the measurement site. The waveform of the obtained PPG signal differs depending on
`the PPG probe contact pressure. A number of studies have suggested that the PPG waveform coincides
`with arterial stiffness and vascular reactivity. Several studies have shown that moderate pressure on the
`sensor can improve the PPG signal. Ideally, the best PPG signal can be obtained under conditions of
`transmural pressure, defined as the pressure difference between the inside and outside of the blood
`vessel (i.e., the pressure across the wall of the blood vessel). Insufficient pressure results in inadequate
`contact and consequently low AC signal amplitude. However, PPG signal recording under excessive
`pressure conditions can also lead to low AC signal amplitude, as well as distorted waveforms caused
`by the occluded artery beyond the PPG probe. Optimal contact pressure corresponds to the maximal
`pulsatile amplitude; this occurs when the transmural pressure approaches zero (i.e., under maximal
`arterial compliance).
`Despite numerous attempts, no generally accepted standards have been adopted for clinical or
`fundamental PPG measurements of contact pressure. Changes in the AC pulse amplitude, DC
`amplitude, ratio of AC/DC amplitudes and normalized pulse area of the reflected PPG signal have
`been investigated [35]. In one study, a contact force ranging from 0.2 to 1.8 N was applied to the
`finger [36]. As a result, the AC amplitude increased and then decreased with increasing contact force.
`For different arterial stiffness with gender and age, the pulse amplitude and the AC/DC ratio peaked at
`different contact forces ranging from 0.2 to 1.0 N; most subjects achieved their maximum pulse
`amplitude within 0.2–0.4 N. Note that a calibration load of 0.65 N was applied to the fingertip skin to
`provide a uniform pressure force during the experiment. A similar contact force study was conducted
`on the upper arm [37]; an average compression pressure of 30 mmHg (4 kPa) produced a PPG peak
`amplitude for IR and green light sources.
`Studies have shown that contact pressures ranging from 8 to 12 kPa (60–90 mmHg) resulted in the
`largest PPG amplitude for a reflectance sensor attached to the forehead region above the eye, although
`the signal-to-noise ratio (SNR) did not improve significantly [38]. Note that in SpO2 studies, the ratio
`between the relative change in the red and IR light intensities (R/IR) was used to indicate accurate
`oxygen saturation; R/IR values varied from 0 to 80 mmHg for the forehead of a newborn infant, with a
`portion of the PPG signal derived from tissue underneath the skull [39–41].
`Transmission-type PPG also affects the transmural pressure. Increasing the amplitude of arterial
`pulsations improves the PPG SNR. For example, when measuring the pulsatile amplitude of a
`finger-based PPG sensor, as the cuff pressure increases, the PPG amplitude increases and then
`decreases to zero when the arterial blood vessel becomes occluded. The largest PPG amplitude (i.e.,
`
`
`
`

`

`289
`
`Electronics 2014, 3
`
`approaching zero transmural pressure), in general, corresponds to the mean arterial pressure. However,
`this pressure is too high for application over any significant period. To prevent the collapse of capillary
`beds, the cuff pressure must be on the order of 10 mmHg, which is too low to obtain sufficient PPG
`amplitude. Thus, even for transmission-type photoplethysmographs, care must be taken in how the
`sensor is attached to the finger.
`In one particular study, blood pressure and finger PPG waveforms were simultaneously measured in
`healthy volunteers under different contact pressures ranging from 0 to 200 mmHg [42]. The results of
`PPG regression indicated that the amplitude ratio was highest for a contact pressure of 60 mmHg.
`The optimal contact pressure of the probe is important, particularly with regard to arterial stiffness.
`Arterial stiffness, estimated from the PPG pulse-wave second-derivative parameter, b/a, is strongly
`inconsistent when recorded at non-optimal probe contact pressures. To determine the optimal contact
`pressure, one group obtained in vivo PPG readings from conduit artery sites in five healthy subjects
`recorded with probe contact pressures ranging from 0 to 15 kPa [43]. In this study, an 880-nm
`reflectance-type sensor and force transducer derived the optimal contact pressure criterion from the
`PPG AC signal area; the b/a values showed high repeatability (coefficient of variability < 5%) for a
`contact pressure of 10 kPa (75 mmHg), indicating optimal contact [43].
`
`3.3. Video Plethysmography
`
`A video camera can be used as a PD; this is termed video plethysmography. There are two types of
`video plethysmography. One is non-contact for long distance measurements. This uses a charge
`coupled device (CCD) camera, such as a mobile phone camera or web camera, and examines the light
`reflected from skin. Plethysmographic signals can be measured remotely (>1 m) using ambient
`light and PD with either a simple consumer-level digital camera in movie mode or a cellular
`phone [11,44–46]. Although the green channel features the strongest plethysmographic signal,
`corresponding to an absorption peak by (oxy-) hemoglobin, the red and blue channels also contain
`plethysmographic information. These results suggest that ambient light PPG may be useful for medical
`purposes, such as the characterization of vascular skin lesions or remote sensing of vital signs for
`triage or sports purposes. The other is the contact type for short-distance measurements, such as
`placing the finger on the mobile phone camera to cover the entire camera view and using the phone
`camera flash LED light (white light) to illuminate the finger [47]. This type of PPG measurement
`method has been commercialized by Azumio [48].
`
`3.4. Signal Processing
`
`3.4.1. Introduction
`
`Motion artifacts mainly consist of random low-frequency interference. Therefore, most artifact
`reduction is accomplished by signal processing. The signal-processing algorithm basically assumes
`that the original PPG signal has power only at certain frequencies and that the rest is noise. Several
`signal processing techniques can be applied, including those that use referencing from an acceleration
`signal or those that minimize the motion artifact with synthetic noise generation. However, attempts to
`minimize motion artifacts reported to date do not appear to correlate well with real-world noise
`
`
`
`

`

`Electronics 2014, 3
`
`sources. Moreover, studies have shown that a high degree of randomness is not necessarily correlated
`with true motion artifacts in PPG signals. This section briefly reviews current signal processing
`methods used to reduce motion artifacts.
`
`290
`
`3.4.2. Moving Average Filter
`
`The moving average method is commonly used to reduce motion artifacts and works well for a
`limited artifact range. However, this method does not account for sudden changes. The periodic
`moving average filter (PMAF), based on the quasi-periodicity of the PPG signals, segments the PPG
`signal into periods and resamples each period. Thus, with PMAF, the motion artifacts are removed
`without degrading the signal [49].
`In-band noise occurs when the spectra of the motion artifact and that of the PPG signal overlap
`significantly. However, it is not advisable to use fixed-frequency filtering techniques to eliminate
`motion artifacts due to in-band interference and the overlapping frequency spectra in the PPG signal.
`Instead, motion artifacts can be removed using a filter bank and a matched filter consisting of several
`frequency bands [50]. In this case, the adaptive filter exhibited many variations and errors due to
`amplitude variations during convergence when measuring PPG in real-time, whereas the moving
`average filter exhibited a more stable output. Compared with traditional adaptive filtering methods,
`the ratio variation was 50% lower than that of the moving average filter, allowing more stable
`measurement of oxygen saturation, despite the patient’s movement [50].
`
`3.4.3. Fourier Analysis
`
`A Fourier series is applicable only to periodic signals, and, therefore, cannot be directly applied to a
`PPG signal, which is non-stationary and quasi-periodic. To overcome this problem, Fourier series
`analysis can be applied on a cycle-by-cycle basis. In this case, the acquired data are first filtered using
`a Savitzky-Golay (SG) smoothing filter to remove high-frequency noise. Once the noise has been
`removed, the proposed cycle-by-cycle Fourier series (CFS) analysis is carried out, and the PPG signals
`(IR and red) are reconstructed on a cycle-by-cycle basis. Experimental results have verified the
`efficacy of the proposed method [51]. Moreover, the results showed that artifacts induced by patient
`movement were attenuated by at least 35 dB, reducing the measurement error of the PPG signal
`from 37% to 3% using the technique described above.
`
`3.4.4. Adaptive Filter
`
`The basic function of a filter is to remove the unwanted signal from the signal of interest. A
`commonly used filtering technique separates the signal from noise using the peak frequencies of the
`signal and artifact. In addition, the mean square of the error signal (the difference between the desired
`response and actual response of the filter) can be minimized. However, when the peak frequency of the
`pulse rate is similar to the noise frequency, the pulse rate cannot be separated from the noise. In PPG
`measurements, active noise cancellation with acceleration-based adaptive filters has been attempted by
`several groups (Figure 5). Obtaining the best design usually requires a priori knowledge of certain
`statistical parameters (such as the mean and correlation functions) within the useful signal. With this
`
`
`
`

`

`Electronics 2014, 3
`
`information, an optimal filter can be designed that minimizes the unwanted signals according to
`statistical criteria.
`
`291
`
`Figure 5. Block diagram of an adaptive filter with an accelerometer.
`
`
`
`Adaptive noise cancellation (ANC) is an alternative technique used to estimate signals corrupted by
`additional noise or interference. The adaptive filter is inherently self-designed through the use of a
`recursive algorithm that updates the filter parameters. This approach can be used to obtain the desired
`level of noise rejection, without a priori estimates of the signal or noise. Two inputs are required for
`this method: (i) the primary input containing the corrupted signal and (ii) a reference input containing
`some potential noise correlation to the primary noise; note that an acceleration sensor can be
`used as the reference input. Figure 5 shows a motion-tolerant wearable biosensor using a MEMS
`accelerometer [52,53], which operates based on adaptive noise cancellation utilizing an acceleration
`reference. This development has motivated new and improved PPG sensor design.
`
`3.4.4.1. Least Mean Square Adaptive Algorithm
`
`The least mean square (LMS) adaptive algorithm removes motion artifact noise by estimating the
`synthetic noise reference signal and adapting the filter coefficients based on filter order. LMS
`algorithms are a class of adaptive filters used to mimic the desired filter by finding the filter
`coefficients that relate to producing the least mean square of the error signal.
`Many adaptive techniques have been applied to the reduction of motion artifacts from PPG signals,
`including the normalized least mean squares (NLMS) method, recursive least squares (RLS) filter,
`time varying step-size LMS (TVS-LMS), adaptive step-size LMS (AS-LMS), and Kalman filters. A
`LMS filter with automatic step-size control was used to mitigate the effects of motion artifacts in PPG
`recordings for long-term patient monitoring [54]. The experimental results indicated that the proposed
`variable step-size LMS filter provided better performance than the LMS filter with a fixed step-size.
`In another study, a two-dimensional (2-D) active noise cancellation algorithm was applied to
`compensate for motion-distorted signals using directional accelerometer data [55]; a NLMS adaptive
`filter (fourth-order) was used in the algorithm, resulting in a reduction in signal distortion from 52.34%
`to 3.53% over the frequency range of 1–2.5 Hz (the frequency range for daily motion, such as walking
`and jogging). In addition, this wearable health-monitoring device, equipped with a motion artifact
`
`
`
`

`

`292
`
`Electronics 2014, 3
`
`reduction algorithm, can be integrated as a terminal in a so-called ubiquitous healthcare system, to
`provide continuous health monitoring without interrupting daily life.
`In another study, a triaxial MEMS accelerometer was attached to a PPG sensor to detect wrist
`movement [56]. In this case, a fast transversal RLS algorithm was used to reduce the computational
`complexity of the adaptive filter, by providing an estimate of the linear motion-to-artifact transfer
`function. The resulting output represented an estimation of the noise, which was then minimized by the
`RLS algorithm. Experimental results showed that this device produced more reliable signals that were
`stable against motion artifact corruption under typical types of movement, such as the swinging of
`the arms [56].
`A Laguerre series was implemented to compactly represent the system dynamics for joggers using a
`few parameters, such as the heart rate [57]. This study determined that the standard artifact reduction
`scheme does not work when the physiological signal is correlated with wearer motion. Thus, adaptive
`blind-source separation techniques can be used to recover the physiological signal. However, the
`success of this method is currently limited.
`In one study, the correlations among six representative daily motions and patterns of motion
`artifacts were examined to obtain more precise analysis [58]. An artifact reduction algorithm was
`designed for the analysis using the motion data. In addition, in this study, the short operating times and
`small number of variables for the portable device required the use of an LMS adaptive filter. The
`results using this filter were compared with those using other noise reduction algorithms. For validation
`purposes, a real-time motion artifact reduction experiment was performed using a wearable device
`during the motions. The results indicated that the motion artifacts of wearable PPG devices can be
`effectively reduced using the proposed method and that such devices can potentially be used in daily
`life with any type of motion.
`The TVS-LMS algorithm offers a fast convergence rate. However, performance studies have shown
`that the AS-LMS algorithm provides not only a fast convergence rate, but also minimal mean square
`error (MSE), as indicated by the high SNR values.
`A simple, efficient approach, based on the AS-LMS adaptive filter, was applied to reduce motion
`artifacts in corrupted PPG signals [59]. In this study, a synthetic noise reference signal, representing
`motion artifact noise, was generated internally from the motion artifact-corrupted PPG signal itself,
`and therefore no additional hardware (e.g., accelerometer or source-detector pair) was used. The
`generated noise reference signal was then filtered through the AS-LMS adapt