A new method based on complex EMD for motion artifacts reduction in PPG signals for pulse oximeter application

Transcription

1 Journal of Engineering Technology (ISSN: ) Volume 6, Special Issue on Technology Applications and Innovation, PP A new method based on complex EMD for motion artifacts reduction in PPG signals for pulse oximeter application Ashoka Reddy Komalla Departmet of ECE, kakatiya Institute of Technology & Science, Warangal, Telangana, India Received: June 18, 17; Revised: September 9, 17; Accepted: October 13, 17 Abstract. A method based on complex empirical mode decomposition (CEMD) is presented for reduction of motion artifacts (MA) in photoplethysmogram (PPG) signals for pulse oximeter applications. The Pulse oximeter is used in the intensive care units (ICU) for continuous monitoring of pulse rate and arterial blood oxygen saturation (SpO ). Reliability of SpO estimations by pulse oximeters will be affected by the frequently encountered MAs. A prototype analog front-end is developed to record intended experimental test data from volunteers. PPGs are acquired with intentionally created MAs (Horizontal motion of finger, Vertical motion of finger and Bending of finger). Experimental results demonstrated the efficacy of CEMD method in restoring PPG morphology and its performance is evaluated by computing the statistical and quality measures indicating the signal reconstruction. Further, the CEMD processed PPGs resulted in stable SpO estimation readings for making it suitable for pulse oximeter application. Keywords: Pulse Oximeter, PPG, Motion Artifacts, Arterial blood oxygen saturation (SpO ), EMD, Complex EMD 1 Introduction During anesthetic procedures and in post operative intensive care units, the pulse oximeter has become an essential equipment to monitor arterial blood oxygen saturation (SpO ) and pulse rate (PR) [1]. The PR and SpO are estimated using photoplethysmogram (PPG) signal, called pleth. In pulse oximeter, pleths at red (R) and infrared (IR) wavelengths are acquired using an opto-electronic PPG probe. The PPG signal consists of pulsatile AC and non-pulsatile DC components. Pulsatile component is mostly due to the component of light passing through pulsatile arteries caused by heartbeat and non-pulsatile component is due to the constant absorption of light when passing through skin-tissue-bone []. Reliable and accurate estimation of SpO requires R and IR PPG signals with clearly separable pulsatile and nonpulsatile components. However, there are important issues such as motion artifacts (MA), skin pigmentation, ambient light effect, hypo-perfusion etc., which affect the reliability of pulse oximeter estimations. MAs are induced in to PPGs due to intentional or unintentional movements of patient. It results in various errors such as reduced accuracy in SpO estimation, loss of signal, creating false alarms and missed hypoxemic events. The PPG signals when afflicted with severse MAs look like ones as shown in fig1.the frequency spectrum of noise due to MA (.1 Hz and more) has every chance of overlapping with that of useful PPG signal (.5-4. Hz) and results in in-band noise referred as artifact. Motion induced artifacts in the recorded PPG data are less deterministic in nature and its compensation is critical. The spectra of a typical PPG signal corrupted with MA and MA reduced PPG is 187

2 Magnitude Amplitude (V) Journal of Engineering Technology (ISSN: ) Volume 6, Special Issue on Technology Applications and Innovation, PP depicted in Fig.. The frequency content of PPG include the pulsatile cardiac synchronoous AC portion (.5-4 Hz), respiratory activity (.-.35 Hz) and MA noise component (.1 Hz or more). In general the effect of MAs can be reduced by displaying the average value of several previous SpO readings, during MA episodes. 6 IR PPG RED PPG Time (S) Figure. 1. Typical R and IR PPG signals with severe MAs 1 (a) 1 (b) Frequency (Hz) Figure.. Spectrum of MA corrupted PPG in (a) and spectrum of of MA recovered PPG in (b) showing absence of MA in the spectrum Several pulse oximeter manufacturers are in fray for developing motion resistant pulse oximeters. Therefore, reduction of MA in PPG has been a challenging problem for researchers ever since the invention of pulse oximetry. Several signal processing methods were proposed in the literature to address this problem. Moving average method [3] is the most commonly employed method. It fails to eliminate the in-band noise effects. Different approaches using adaptive filtering were proposed [4]-[6]. Fir their operation, adaptive filters require a reference signal. Major limitation of adaptive filtering based methods for pulse oximeters is that they use 188

3 Journal of Engineering Technology (ISSN: ) Volume 6, Special Issue on Technology Applications and Innovation, PP additional hardware for reference signal generation. Use of a synthetic reference signal estimated from the artifact-free part of the PPG signal [7] was also reported for reducing motion artifact. In the Masimo SET [8], venous blood volume changes are identified as a significant contributor to noise during motion and hence a venous noise reference signal is extracted from the artifact-induced PPG signal itself, without extra hardware. The added artifact is then removed from the PPG signal using adaptive noise cancellation. A signal processing technique using multi-rate filter bank with a matched filter [9] performed better compared to the moving average approach. By exploiting the independence between PPG and MA signals, it was also demonstrated that MA were reduced by using independent component analysis (ICA) [1]-[1]. While the third order ICA of the time-derivative of PPG signals resulted in better artifact suppression, the ICA applied with a preprocessing called block-interleaving with low pass filtering performed better than ICA alone. Other approaches for reducing MA in PPG signals include adaptive filtering and improvements in mechanical design and sensor configuration [13]. Of which, motion tolerant wearable bio-sensors using MEMS accelerometer [14] based on adaptive noise cancellation utilizing accelerometer reference has motivated the researchers to think in the direction of improving sensor design. Recent processing methods like wavelet decomposition method [15], singular value decomposition (SVD) [16], cycle by cycle Fourier series method [17], higher order statistics (HoS) method [18], and ratio-metric method [19] extracted artifact reduced PPG signals restoring the essential PPG morphology. For pulse oximetry application, ensemble empirical mode decomposition (E MD) based method [1] has shown improved performance over simple empirical mode decomposition (EMD) based method [] in reducing MAs. An advanced time-frequency analysis method based on EMD using variance characterization of extrema was presented in [] and showed that the method was accurate in pulse rate estimation for heavily-corrupted PPG signals. Of all the methods available in the literature, only a few addressed the issue of MA reduction in PPG signals with reference to pulse oximeter applications. In this paper, we propose a simple yet efficient method based on complex empirical mode decomposition (CEMD) for MA reduction in PPG signals specially for pulse oximeter application. Motion Artifact Reduction using Complex EMD (CEMD) Method.1 Basic EMD Method Empirical mode decomposition (EMD) method proposed by N. E. Huang et al., is an adaptive time-frequency analysis method [4]. It decomposes a given signal f(n) into a set of AM FM components, called intrinsic mode functions (IMFs). The resulting M intrinsic modes Il ( n ) and a residual term can be represented as L f ( n) Il ( n) r( n) (1) l 1 where, n is sample number This basic form of EMD is also considered here in this paper for comparing the performance of CEMD over EMD. The EMD based MA reduction procedure is depicted in fig

4 Journal of Engineering Technology (ISSN: ) Volume 6, Special Issue on Technology Applications and Innovation, PP MA Corrupted PPG Signal f(n) Empirical Mode Decomposition Compute the spectrum of each IMF Eliminate IMF corresponding to MA and add remaining IMFs MA reduced PPG Signal f( n) Figure.3. Motion Artifact Reduction in Corrupted PPG signals using HHT based EMD method The various steps of EMD algorithm are summarized in the flowchart shown in Fig. 4. Figure. 4. Flow chart indicating various stages of EMD algorithm While processing using EMD algorithm, first the PPG signal is decomposed into several IMFs of different resolution scales, based on the intrinsic scale of signal, and then gets the characteristic information of signal quickly with the well local time-frequency performance. Then using the Hilbert transform, time-frequency characteristic can be described. Basically, implementation of EMD technique is well defined by a process called sifting. It decomposes a given signal f(n) in to IMFs as per (1). After having extracted all IMFs, they will be analyzed by Hilbert Transform (HT). The HT calculates the conjugate pair of the data set f(n), which can be thought of as convolution of f(n) with the function1/ ( n). 19

5 Journal of Engineering Technology (ISSN: ) Volume 6, Special Issue on Technology Applications and Innovation, PP f( k) fˆ( n ) () ( n k) k For each IMF component, Il ( n ), the analytical signal is to be obtained as given below z( n) I ( n) i fˆ ( n) (3) l In [1], it has been found that the main limitation of conventional EMD method is the length of the signal being processed. Especially the biomedical signals are often recorded over very long time spans. With simple and yet significant modifications, use of complex EMD (CEMD) for MA reduction is proposed in this paper.. MA Reduction using Complex EMD (CEMD) Method Complex EMD (CEMD) is an important extension that can be considered for improved MA reduction. In this method, the EMD results are analyzed in complex domain, separately for both positive and negative frequency parts of the data. The modified steps incorporated in the conventional EMD algorithm for PPG signal processing are described as follows: i. Recorded PPG data is framed into real and imaginary components, wherein the imaginary part is the Hilbert transform of the real one ii. Now, decompose the real and imaginary components of MA corrupted PPG data into IMFs iii. MA recovered PPG is obtained by identifying the range of MA noise frequency, eliminating IMF s (of both real and imaginary components) corresponding to MA and then adding the remaining IMF s falling in the particular range of desired portion. 3 Designed Analog Front-end for data Acquisition An analog front-end, shown in fig.5, was developed and used for simultaneous acquisition of red and IR PPGs for further processing. A clip-on type PPG sensor (Nellcor N1 make ), housing a red LED (66nm) and an IR LED (91nm) on one side and a photodiode detector on the other side was used to record PPG signals from different subjects and was interfaced to the signal conditioning circuitry. The red and IR PPG signals were acquired under LabVIEW environment at a sampling rate of 1 Hz using, NI-DAQPad-615, a data acquisition system manufactured by National Instruments. The LEDs and the photo diode are connected to an electronic signal conditioning circuit, shown in fig. 3, for obtaining two PPG signals. To avoid interference between the red and IR lights, the individual drives to the LEDs are time sliced (when red LED is ON, the IR is kept OFF and vice versa) at a rate of 1 khz. Due to the multiplexing of the red and IR LEDs, the output of the photodiode will contain both the red and IR PPG signals multiplexed at 1 khz, and hence the photodiode output should be de-multiplexed. The circuit shown in fig.5 accomplishes this task as well. 191

6 Journal of Engineering Technology (ISSN: ) Volume 6, Special Issue on Technology Applications and Innovation, PP Figure.5. Developed analog front-end used for PPG recording Experimental procedure adopted while recording the PPG data was approved by the ethics committee of the institute and the PPG data were collected after obtaining informed consent from all the volunteers. During recording the signals from volunteers, they were instructed to introduce the following most frequently encountered artifacts so that the recorded signals mimic the actual scenario. (i) vertical motion of finger (ii) horizontal motion of finger (iii) bending motion of finger 4 Results and Discussion The PPG signals were recorded with intentionally created artifacts to mimic real time situation. The recorded PPG data were processed and analysed using MATLAB R7b signal processing Toolbox. The frequency content of PPG include the pulsatile cardiac synchronoous AC portion (.5-4 Hz), respiratory activity (.-.35 Hz) and MA noise component (.1 Hz or more). 4.1 PPG Processing using Proposed CEMD Method As the SpO estimation needs clearly separable AC and DC components of PPG signal, first it is to be established that artifact-free portions of the PPG signal will not be disturbed by processing of the proposed CEMD method. The results depicted in Fig. 6 show restoration of PPG signal morphology. Here some clean sections of PPG recordings were processed by proposed CMED method. 19

7 Amplitude (V) Amplitude (v) Journal of Engineering Technology (ISSN: ) Volume 6, Special Issue on Technology Applications and Innovation, PP (a) (b) Time (s) Figure. 6. PPG before and after processing: (a) Clean section of red PPG; (b) signal in (a) after processing with CEMD Then proposed method is applied on the artifact corrupted signal. The considered signal is severely afflicted with MAs of vertical motion of finger. The CEMD processed signal is depicted in fig. 7. Visual inspection reveals a very good morphology for recovered signal. (a) (a) - 1 (b) Time (S) Figure. 7. PPG before and after processing: (a) PPG with MA; (b) signal in (a) after processing with CEMD To test the performance of the proposed method, the results are compared with othe variants of EMD i.e., basic EMD [] and E MD [1]. The results are portrayed in fig. 8 through fig. 1 for different MA cases namely vertical motion of finger, horizontal motion of finger and bending motion of finger. 193

8 Amplitude (V) Magnitude Amplitude (V) Magnitude Journal of Engineering Technology (ISSN: ) Volume 6, Special Issue on Technology Applications and Innovation, PP Time (S) Frequency (Hz) (a) (b) (c) (d) Figure. 8. Red PPG with Vertical motion of finger : (a) - MA corrupted PPG, (b) - MA recovered PPG using EMD, (c) - MA recovered PPG using EMD, (d) - MA recovered PPG using CEMD. Their spectra are also plotted. The recovered portions of MA reduced PPGs show the improved performance of the proposed CMED method in reducing MAs. The CMED processed signals show improved restoration of morphological features of the PPG. Visual inspection of their spectra reveals that a significant improvement is guaranteed in reducing MAs with the all the three methods. In particular, the CEMD technique out performed others making it an efficient signal processing technique for any type of MA reduction method best for pulse oximetry applications (a) (b) Time(S) Frequency (Hz) Figure. 9. Red PPG with Horizontal motion of finger : (a) - MA corrupted PPG, (b) - MA recovered PPG using EMD, (c) - MA recovered PPG using EMD, (d) - MA recovered PPG using CEMD. Their spectra are also plotted. (c) (d) 194

9 Amplitude (V) Magnitude Journal of Engineering Technology (ISSN: ) Volume 6, Special Issue on Technology Applications and Innovation, PP (a) -5 4 (b) - 3 (c) (d) Time (S) Frequency (Hz) Figure. 1. Red PPG with Bending motion of finger : (a) - MA corrupted PPG, (b) - MA recovered PPG using EMD (c) - MA recovered PPG using EMD, (d) - MA recovered PPG using CEMD. Their spectra are also plotted. In addition to visual inspection of processed outputs, to prove the efficacy of CEMD based method, the SNR and NRMSE are also calculated. 4. SNR and NRMSE Calculations (i) SNR: A figure of merit, SNR, defined as the ratio of signal power to the generated noise reference power was computed for the MA corrupted and MA recovered PPG signals. rms( signal) SNR ( db) log db rms( noise) Table 1 illustrates the SNR values for the PPGs inflicted with three different kinds of MA. The superiority of SEMD is clearly evident in representing high SNR values in all the cases. Table 1 COMPUTED VALUES OF SNR FOR MA CORRUPTED AND MA RECOVERED PPG FOR DIFFERENT MAs SNR of PPG Before EMD Processing Horizontal motion (4) Vertical motion Bending motion Using EMD Using E MD Using CEMD (ii) NRMSE: The normalized root mean squared error (NRMSE) quantifies deviation of the recovered PPG from the original PPG. NRMSE is defined as below and the values are presented in Table. 195

10 Journal of Engineering Technology (ISSN: ) Volume 6, Special Issue on Technology Applications and Innovation, PP N NRMSE ( db) log ( f ( n) f ( n)) ( f ( n)) n n N db (5) where, f( n) is the original signal, f( n ) is the recovered signal and N represents one period of the recovered PPG cycle. Table. COMPUTED VALUES OF NRMSE FOR CORRUPTED AND RECOVERED PPG FOR DIFFERENT MAs Horizontal Vertical Bending NRMSE of PPG motion motion motion Using EMD Using E MD Using CEMD Improved values of SNR and NRMSE in Both tables I and II, indicate that the morphological features of PPG can be better restored to large extent by CEMD based method compared to MED and EMED based methods for MA reduction. 4.3 Peak-to-Peak values of recovered cycles as an indicator of signal morphology Another figure of merit that describes restoration of signal morphology is calculation of peak-to-peak value of cycles. The statistical analysis is carried out in terms of mean + SD of peak-to-peak values of every five restored PPG cycles obtained after applying the EMD, E MD and CEMD methods on MA corrupted PPG signals. Results are presented in Table 3. The peak-to-peak values of recovered PPG cycles from using CEMD algorithm are very close to the clean sections of the PPG. Table 3. EFFECTIVENESS OF PROPOSED CEMD METHOD IN RESTORING PEAK-TO-PEAK VALUES OF PPG PPG Horizontal Vertical Bending motion motion motion Clean Section.371±.5.48± ±. Corrupted portion.46± ± ±.67 Recovered portion using EMD.391±.5.451± ±.11 Recovered portion using E MD.385± ± ±.138 Recovered portion using CEMD.38± ±.5.359± SpO Computation from Artifact Reduced PPG Signals Visual inspection and statistical measures like SNR, NRMSE and peak-to-peak values established the efficacy of the proposed CEMD method. The main objective of this paper is the present an efficient MA 196

11 Journal of Engineering Technology (ISSN: ) Volume 6, Special Issue on Technology Applications and Innovation, PP reduction algorithm for pulse oximeter applications. As already mentioned, the pulse oximeter non-invasively estimates pulse rate and arterial blood oxygen saturation(spo ) using PPGs obtained at R and IR wavelengths. For accurate estimation of SpO, artifact-free R and IR PPG signals with clearly separable AC and DC parts are required. Pulse oximeter manufacturers depend on calibration curves derived from the PPG data of young and healthy volunteers. One such empirical linear approximate relation for computation of the SpO is given by [3] SpO % ( 11 5 Q )% (6) where Q is known as normalised ratio of ratios, given by where R R ( AC / DC ) Q (7) ( AC IR / DC IR ) AC R = peak-to-peak values of the pulsatile components of the red PPG signal AC IR = peak-to-peak values of the pulsatile components of the IR PPG signal In general, the processor inside the pulse oximeter computes the equations (7) and then (6) from the acquired PPGs. To further establish the efficacy of the CEMD method, the CEMD processed PPGs are used for estimation of SpO. First, the normalized Q value is calculated from AC and DC component values of processed Red and IR PPG signals by measuring the peak to peak amplitudes consecutively for every five cycles in time domain. Computed Q values are used in equation (6) for estimation of SpO values. The table 4 presents the SpO estimations obtained from CEMD processed PPGs along with the ones obtained from EMD and E MD processed PPGs. It can be seen that, the %SpO estimations using CEMD processed PPGs are very close to that of SpO of clean sections in all cases of MAs. Table 4. COMPUTED R AND ESTIMATED SpO VALUES IN TIME DOMAIN PPG Horizontal motion Vertical motion Bending motion R value SpO R value SpO R value SpO Clean section.48±. 98%.56±.15 96%.48±.6 98% Recovered EMD Recovered E MD Recovered CEMD using using using.53± %.64±.1 94%.54± %.5±.6 97%.6± %.53± %.49± %.6±.17 95%.49± % Further, a 3 second duration R and IR PPG signal afflicted with MAs of vertical motion of finger, shown in fig.1, are processed on-line. The SpO estimations before and after processing the PPGs by various EMD algorithms are shown as SpO plot in Fig. 11. The SpO values estimated using raw PPGs during the period of motion artifacts (MA) fluctuate in a very random manner and are hence unreliable. On the other hand, the SpO values estimated using the CMED processed PPGs provided stable readings compared to other methods, 197

12 SpO (%) Journal of Engineering Technology (ISSN: ) Volume 6, Special Issue on Technology Applications and Innovation, PP thus establishing the efficacy of the CEMD method in reducing the MAs in PPGs and is the best suitable method for pulse oximeter applications. MA corrupted IR PPG 3 MA corrupted RED PPG - 1 MA reduced IR PPG using CEMD -3 MA reduced RED PPG using CEMD MA corrupted EMD E MD CEMD 6 1 Time (s) 3 Figure 11. SpO estimated for raw PPG corrupted with vertical motion of finger and processed PPGs using conventional EMD, EMD, and CEMD algorithms 4. Results and Discussion Pulse oximeters require artifact-free photoplethysmogram (PPG) signals for non-invasive estimation of arterial blood oxygen saturation (SpO ). Motion artifacts (MAs) disturb the morphology of PPGs and hence lead to erroneous SpO estimations. In this paper, an artifact-reduction algorithm based on complex empirical mode decomposition (CEMD) method has been presented especially for use in pulse oximeters. The method performed well and its efficacy was established by comparing its performance with basic EMD and ensemble EMD (E MD) based MA reduction methods. In addition, the PPGs processed by CEMD based method provided stable SpO reading during MA episodes. 198

13 Journal of Engineering Technology (ISSN: ) Volume 6, Special Issue on Technology Applications and Innovation, PP References [1] J. G. Webster, Design of Pulse Oximeters. New York: Taylor & Francis, [] K. A. Reddy, Novel methods for performance enhancement of pulse oximeters, Ph.D. dissertation, Dept. Elect. Eng., IIT Madras, India, 8. [3] T. L. Rusch, R. Sankar, and J. E. Scharf, Signal processing methods for pulse oximetry, Comput. Biol. Med., vol. 6, no., pp , Mar [4] A. B. Barreto, L. M. Vicente and I. K. Persad, Adaptive cancellation of motion artifact in photoplethysmographic blood volume pulse measurements for exercise evaluation, in Proc. IEEE- EMBC/ CMBEC, Sept. -3,1995, vol., pp [5] A. R. Relente and L. G. Sison, Characterization and adaptive filtering of motion artifacts in pulse oximetry using accelerometers, in Proc. Conf. EMBS/BMES, Houston, USA, Oct. 3-6,, pp [6] K. W. Chan and Y. T. Zhang, Adaptive reduction of motion artifact from photoplethysmographic recordings using a variable step-size LMS filter, in Proc. IEEE Sensors,, vol., pp [7] F. M. Coetzee and Z. Elghazzawi, Noise-resistant pulse oximetry using a synthetic reference signal, IEEE Trans. Biomed. Eng., vol. 47, no. 8, pp , Aug.. [8] J. M. Goldman, M. T. Petterson, R. J. Kopotic and S. J. Barker, Masimo signal extraction pulse oximetry, J. Clin. Monit. Comput., vol. 16, no. 7, pp ,. [9] J. Lee, W. Jung, I. Kang, Y. Kim and G. Lee, Design of filter to reject motion artifact of pulse oximetry, Comput. Stand. Interfaces, vol. 6, No. 3, pp , May 4. [1] P. F. Stetson, Independent component analysis of pulse oximetry signals, in Proc. 6 th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc., San Francisco, USA, Sept. 1-5, 4, pp [11] B. S. Kim and S. K. Yoo, Motion artifact reduction in photoplethysmography using independent component analysis, IEEE Trans. Biomed. Eng., vol. 53, no. 3, pp , Mar. 6. [1] J. Yao and S. Warren, A short study to assess the potential of independent component analysis for motion artifact separation in wearable pulse oximeter signals, in Proc. 7 th Annu. Conf. IEEE Eng. Med. Biol., Shanghai, China, Sept. 1-4, 5, pp [13] S. Rhee, B. H. Yang, and H. H. Asada, Artifact-resistant power-efficient design of finger-ring plethysmographic sensors, IEEE Trans. Biomed. Eng., vol. 48, no. 7, pp , July 1. [14] P. Gibbs and H. H. Asada, Reducing motion artifact in wearable bio-sensors using MEMS accelerometers for active noise cancellation, in Proc. 5 American Control Conf., Portland, OR, USA, June 8-1, 5, pp [15] J. Y. A. Foo, Comparison of wavelet transformation and adaptive filtering in restoring artifactinduced time-related measurement, Biomed. Signal Process. Control, vol. 1, no. 1, pp , 6. [16] K. A. Reddy and V. J. Kumar, Motion artifact reduction in photoplethysmographic signals using singular value decomposition, in Proc. 4 th IMTC, Warsaw, Poland, May 1-3, 7, pp

14 Journal of Engineering Technology (ISSN: ) Volume 6, Special Issue on Technology Applications and Innovation, PP [17] K. A. Reddy, B. George, and V. J. Kumar, Use of fourier series analysis for motion artifact reduction and data compression of photoplethysmographic signals, IEEE Trans. Instrum. Meas., vol. 58, no.5, pp , May 9. [18] R. Krishnan, B. Natarajan, and S. Warren, Two-stage approach for detection and reduction of motion artifacts in photoplethysmographic data, IEEE Trans. Biomed. Eng., vol. 57, no. 8, pp , Aug. 1. [19] James A. C. Patterson and G. Z. Yang, Ratiometric artifact reduction in low power reflective photoplethysmography, IEEE Trans. Biomed. Circuits Syst., vol. 5, no. 4, pp , Aug. 11. [] M. R. Ram, K. V. Madhav, E. H. Krishna, K. N. Reddy, K. Sivani and K. A. Reddy, HHT based signal decomposition for reduction of motion artifacts in Photoplethysmographic signals, in Proc. of 9 th IEEE International Instrumentation and Measurement Technology Conf., IMTC-1, Graz, Austria, May, 1, pp [1] M. R. Ram, K. Sivani and K. A. Reddy, E MD for Reduction of Motion Artifacts from Photoplethysmographic Signals, in Proc. of International Conference on Electronics and Communication System (ICECS 14), Karpagam college of Engineering, Coimbatore, India, Feb., 14, pp. 1-4, [] Bo Pang et al, Advanced EMD method using variance characterization for PPG with motion artifact, in proc. of IEEE Biomedical Circuits and Systems Conference (BioCAS), Shanghai, Chaina, 17-19, pp , Oct. 16 [3] K. A. Reddy, B. George, N. M. Mohan and V. J. Kumar, A novel calibration-free method for measurement of oxygen saturation in arterial blood, IEEE Trans. Instrum. Meas., vol. 58, no. 5, pp , May 9 [4] Huang, N. E., Z. Shen, and S. R. Long, M. C. Wu, E. H. Shih, Q. Zheng, C. C. Tung, and H. H. Liu, The empirical mode decomposition method and the Hilbert spectrum for non-stationary time series analysis, Proc. Roy. Soc. London, 454A, pp , 1998