Application Of DSP To Remove Noise From ECG Signal

Application Of DSP To Remove Noise From ECG Signal

Application Of DSP To Remove Noise From ECG Signal

Vol. 2 Issue 6, June – 2013

Rahul Dev, Prof. R. K. Mehrotra

1 M.Tech Student,Department Of Electronics and Communication , Ajay Kumar Garg Engg.College,Gzb(UP)

2 Professor, Department Of Electronics and Communication,Ajay Kumar Garg Engg. College.Gzb(UP)

Heart diseases, which are one of the death reasons, are among the several serious problems in this century and as per the latest survey, 60% of the patients die due to Heart problems. In 2011, 10 crores people are suffering from heart diseases in India. Many individuals remain unaware of the symptoms of heart attack or dismiss possible symptoms as being unrelated or not important enough to visit a doctor. Early diagnosis and medical treatment of heart diseases can prevent sudden death of the patient. One of the ways to diagnose heart diseases is to use Electrocardiogram (ECG) signals. ECG measures electrical potentials on the body surface via contact electrodes. However, timely and accurate detection of arrhythmias is a complex decision-making process for a cardiologist due to contamination of ECG signals with different frequencies of noise. For reliable interpretation of real-time ECGs, computer based techniques on digital signal processing (DSP) of ECG waveform have been reported. ECG signals are non-stationary signals including valuable clinical informations, but frequently these informations are corrupted by noise. Many tools, methods and algorithms based on signal processing theory have been proposed and implemented. In this paper, advanced digital signal processing is carried out in Matlab environment. And the present work basically focuses on implementation and evaluation of methods to remove noise from ECG signal.

1. The ECG is one of the oldest and the most popular instrument used in medical applications and has followed the progress of instrumentation technology. Its most recent evolutionary step, to the computer-based system, has allowed patients to wear their computer monitor or has provided an enhanced, high resolution ECG that has opened new scene of ECG analysis and interpretations [1-2].

   The ECG is a bioelectric signal, which records the hearts electrical activity time, therefore it is an important diagnostic tool for assessing heart function. The electrical current due to the depolarisation of the Sino-Atrial (SA) node stimulates the surrounding myocardium and spreads into the heart tissues. A small proportion of the electrical current flow to the body surface. By applying electrodes on the skin at the selected points, the electrical potential generated by this

   current can be recorded as an ECG signal. Early diagnosis and treatment of heart diseases can prevent sudden death of the patient. One of the ways to diagnose heart diseases is to use electrocardiogram (ECG) signals. ECG signals are formed of P wave, QRS complex, and T wave. They are designated by capital letters P, Q, R, S, and T. In the normal beat phase of a heart, the main parameters, inspected includes; the shape, the duration, and the relationship with each other of P wave, QRS complex, and T wave components and R-R interval [3]. During measurement of ECG noise(Anything other than muscular activity of heart) is superimposed on them, due to AC interference, loose electrode connection, malfunctioning of machine, patient movement like respiration etc all of them collectively called artifacts.

Vol. 2 Issue 6, June – 2013

Many times when ECG signal is recorded from surface electrode that are not tightly in contact with the skin as the patient breath the chest expand and contract producing a relative motion between skin and electrode. This results in shift in the baseline which is also known as low frequency baseline wander. Obviously the fundamental frequency of baseline wander is same as that of respiration frequency. It is required that this baseline wander is to remove from the ECG before extraction of any meaningful feature. Baseline wander makes manual and automatic analysis of ECG records difficult, especially in the detection of ST-segment deviations. This segment is very important and has the information related

In this section, various noise removal techniques are applied to MIT/BIH ECG database data sample, and the performances are studied on the basis of spectral density and average power of signal. In the first step, the simplest approach which is linear trend to remove baseline drift is applied after that various digital filters are applied to the noisy ECG data having baseline noise as shown in Fig 1. The adaptive filter is then applied on the sample ECG signal to remove power-line noise and finally the wavelet approach is used for overall de-noising of ECG signal.

to heart attack, this because the spectrum of baseline wander and low frequency component of ECG signal usually overlaps.

This paper work aims at removing this baseline wander signal while preserving the low frequency ECG clinical information. The main purpose of this study is to implement various digital filters and then wavelet for overall de-noising after that performance of each filter is also discussed. The present work is divided into following steps:

Step 1: selection of standard ECG data and extraction of ECG signal.

Step 2: Design and implementation of FIR and IIR filters for the removal of baseline noise from ECG signal.

1400

1300

1200

sample value

sample value

1100

1000

900

800

700

Noisy ECG data 207.dat

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

Fig. 1 The noisy ECG data having baseline noise

Step 3: Implementation of wavelet for overall de-noising. Step 4: Implementation of adaptive filters for the removal of power-line noise from ECG signal.

A major element of this stage was the extraction of ECG signals from the standard database that we have been chosen in this work. After extraction, the signals were subject to processing using several tools available by the MATLAB software.

Digital Filters: In signal processing, the function of a filter is to remove unwanted parts of a signal, such as random noise, or to specify useful parts of the signals, such as components lying in a certain important frequency range. There are two main kinds of filters; Analog and Digital filters. General digital filter design process can be divided into four main steps:

• Approximation.

• Synthesis and realization.

• Performance analysis.

• Implementation.

Both digital FIR and IIR filters are designed using this filter design procedures in order to approach the required filter specifications.

Implementation of the simplest approach Linear Trend or A Piecewise Linear Trend: The most simple approach is a linear trend or a piecewise linear trend. The MATLAB- function, de-trend performs piecewise linear de-trending.

Noisy ECG data 103.dat

1400

1300

sample value

sample value

1200

1100

1000

900

800

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

Fig. 2 Noisy ECG

signal 103.dat after detrending

350

1500

Noisy ECG data 114.dat

Vol. 2 Issue 6, June – 2013

300

250

200

sample value

sample value

150

100

50

1400

sample value

sample value

1300

1200

1100

0

-5

-100

-150

0 1000 2000 3000 4000 5000 6000 7000 8000

1000

900

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

sample number

Fig 3 Signal after de-trending

Fig 5 The noisy ECG having Baseline noise

Filtered by Rectangular Window

But de-trend fails in case of non-linear trends so by setting the option, BP yields a piecewise linear de-trending vector BP contains the indices of the breakpoints between adjacent linear segments. The breakpoint between two segments is defined as the data point that the two segments share.

Advanced DSP Technique to Remove Baseline Noise from ECG Signal

Design and implementation of FIR and IIR filters on ECG Data

FIR-design with MATLAB

500

400

sample value

sample value

300

200

100

0

-100

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

Fig 6 ECG signal 114.dat filtered by Rectangular filter

Function signal toolbox

calculation of coefficients

estimation of the order

filtering

type and frequncy of cutoff

frequency response

600

400

sample value

sample value

200

0

-200

-400

Filtered by Hanning Window

Implementations of FIR filter: In this section, FIR Equiripple filter, windowing FIR filters with Kaiser,

-600

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

Rectangular, Hanning and Blackman functions are performed.

Fig 7 ECG signal filtered by Hanning window

500

400

300

Filtered by FIR Equiripple Filter

Vol. 2 Issue 6, June – 2013

Implementations of IIR filter: In this case, four IIR filters

i.e. Butterworth filter, Chebyshev Type 1, Chebyshev Type II and Elliptic filter are implemented on ECG signal.

sample value

sample value

200

100

0

-100

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

Fig 8 ECG signal filtered by Equiripple Window

1200

1000

800

sample value

sample value

600

400

200

114.dat filtered by Butterworth

450

400

350

300

sample value

sample value

250

200

Filtered by Kaiser Window

0

-200

-400

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

Fig 12 ECG signal filtered by Butterworth filter

150

100

50

0

-50

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

1000

800

600

sample value

sample value

400

114.dat filtered by Chebyshev I filter

Fig 9 ECG signal filtered by Kaiser window

200

600

400

200

Filtered by Blackmann Window

0

-200

-400

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

sample value

sample value

0 Fig 13 ECG signal filtered by Chebyshev I filter

-200

-400

-600

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

1200

1000

800

sample value

sample value

600

114.dat filtered by Chebyshev Type II filter

Figure 10 ECG signal filtered by Blackmann window

400

200

Function signal toolbox

calculation of coefficients

estimation of the order

type and frequncy of cutoff

frequency response

0

-200

-400

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

filtering

Fig 14 ECG signal filtered by Chebyshev Type II filter

Fig 11 Block diagram of IIR-design

1000

800

600

sample value

sample value

400

200

0

-200

-400

114.dat filtered by Eliptic filter

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

Vol. 2 Issue 6, June – 2013

Advanced DSP Technique to Remove Baseline Noise from ECG Signal

Tables 1 show the comparison of different FIR filters. It has been observed that fall in value of spectral density is very large in case of FIR equiripple filter as compared to all windows. The trade-off between spectral density and average power is best among all the FIR filters. But it can also visualize that the waveform got distorted to some extend in case of FIR equiripple filter. The Kaiser Window and rectangular window is also showing better results at the expense of some more computational load as the order of the filter is large. But, in case of remaining windows i.e. Hanning

Fig 15 ECG signal filtered by Elliptic filter

Calculation of parameters:

The two important parameters to check the suppression of Baseline noises are spectral density and average power of signal. The cut-off frequency used for high pass filters is 0.5Hz, so the spectral density below 0.5 Hz is calculated before and after filtration of ECG signals. The signal should have good trade-off between spectral density and average power after filtration and also minimum order of filter as possible to avoid the computational complexity.

Table 1 Comparison of various filters for Removal of Baseline noise (ECG sample 103.Dat)

and Blackman windows, the order of filter easily grow very much high. It increases the number of filter coefficients which increases the large memory requirement and problems in hardware implementation. So, the FIR Kaiser Window filter can be best choice for the removal of Baseline wandering among FIR filters.

Tables show the comparison of different IIR filters. The best trade-off between spectral density and average power is shown by Chebyshev type II filter. It can also visualize from Fig that baseline noise have removed completely. So, the IIR Chebyshev type II filter can be best choice for the removal of baseline noise among IIR filters.

Implementation of wavelet approach on filtered data for overall de-noising: In this step de-noising oriented function is used. Default values for all the general procedures related to de noising using wavelets. The best combination based on the performance of digital filter is chosen and after applying wavelet on the filtered signal it is visualize that the overall noise have removed without effecting the parameters such as spectral density and average power.

Filter	Filter Order	Spectral density before filtration (dB/Hz)	Spectral density after filtration (dB/Hz)
IIR Butterworth	2	40.35	34.04
IIR Chebyshev type I	2	40.35	34.08
IIR Chebyshev type II	2	40.35	7.31
IIR Elliptic	2	40.35	34.08
FIR Equiripple	320	40.35	30.62
FIR Kaiser	450	40.35	29.74
FIR Rectangular	450	40.35	29.33
FIR Hanning	1200	40.35	30.83
FIR Blackmann	1500	40.35	30.98

Filter	Filter Order	Spectral density before filtration (dB/Hz)	Spectral density after filtration (dB/Hz)
IIR Butterworth	2	40.35	34.04
IIR Chebyshev type I	2	40.35	34.08
IIR Chebyshev type II	2	40.35	7.31
IIR Elliptic	2	40.35	34.08
FIR Equiripple	320	40.35	30.62
FIR Kaiser	450	40.35	29.74
FIR Rectangular	450	40.35	29.33
FIR Hanning	1200	40.35	30.83
FIR Blackmann	1500	40.35	30.98

1500

1400

sample value

sample value

1300

1200

1100

1000

900

114.dat

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

Fig 16 The noisy ECG having Baseline noise

450

400

350

300

sample value

sample value

250

200

150

100

50

filtered by FIR fillter

1600

1400

1200

sample value

sample value

1000

800

600

400

200

0

comparision of all signals

Vol. 2 Issue 6, June – 2013

0

-50

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

-200

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

Fig 21 waveform shows comparison of all signals

450

400

350

300

sample value

sample value

250

Fig 17 ECG signal filtered by FIR filter

filtered by FIR and wavelet combination

10

5

sample value

sample value

0

-5

-10

enlarged view

200

150

100

50

0

-15

1140 1160 1180 1200 1220 1240 1260

sample number

Fig 22 waveform shows the enlarged view of ECG signal

-50

500

400

sample value

sample value

300

200

100

0

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

Fig 18 ECG signal filtered by combination

filtered by IIR filter

Table 2 Comparison of parameters with and without wavelet (ECG sample 103.Dat)

Filter	Spectral density after filtration (dB/Hz)	Spectral density after filter and wavelet combination (dB/Hz)	Average power after filtration (dB)	Average power filter and wavelet combinat ion (dB)
IIR Butterworth	34.04	34.04	38.71	38.70
IIR Chebyshev type I	34.08	34.08	38.30	38.29
IIR Chebyshev type II	4.03	4.03	22.91	22.73
IIR Elliptic	34.08	34.08	38.30	38.29
FIR Equiripple	30.62	30.62	38.80	38.79
FIR Kaiser	29.74	29.74	37.93	37.92
FIR Rectangular	29.33	29.33	38.02	38.01
FIR Hanning	30.83	30.83	37.94	37.93
FIR Blackmann	30.98	30.98	38.08	38.08

Filter	Spectral density after filtration (dB/Hz)	Spectral density after filter and wavelet combination (dB/Hz)	Average power after filtration (dB)	Average power filter and wavelet combinat ion (dB)
IIR Butterworth	34.04	34.04	38.71	38.70
IIR Chebyshev type I	34.08	34.08	38.30	38.29
IIR Chebyshev type II	4.03	4.03	22.91	22.73
IIR Elliptic	34.08	34.08	38.30	38.29
FIR Equiripple	30.62	30.62	38.80	38.79
FIR Kaiser	29.74	29.74	37.93	37.92
FIR Rectangular	29.33	29.33	38.02	38.01
FIR Hanning	30.83	30.83	37.94	37.93
FIR Blackmann	30.98	30.98	38.08	38.08

-100

0 1000 2000 3000 4000 5000 6000 7000 8000

sample number

Fig 19 ECG signal filtered by IIR filter

The above table shows that the overall noise have removed without affecting the parameters such as spectral density and average power.

Removal of power line interference from ECG signal using Adaptive filtering.

• The adaptive LMS function to use.

• An unknown system or process to adapt.

• Appropriate input data to exercise the adaptation process. In terms of the generic LMS model, these are the desired signal and the input signal.

Vol. 2 Issue 6, June – 2013

1200

1000

800

600

400

200

0

original ecg signal

Fig 25 Signal after removal of noise

Table 3 Comparison of Spectral density with and without adaptive filtration at (60 Hz)

MIT/BIH DATABASE	Spectral density before filtration (dB/Hz)	Spectral density after filter (dB/Hz)
100.dat	1.234	-2.914
114.dat	7.081	2.197
123.dat	0.6943	-1.668
207.dat	5.402	-2.809

MIT/BIH DATABASE	Spectral density before filtration (dB/Hz)	Spectral density after filter (dB/Hz)
100.dat	1.234	-2.914
114.dat	7.081	2.197
123.dat	0.6943	-1.668
207.dat	5.402	-2.809

-200

-400

0 500 1000 1500 2000

Fig 23 ECG signal with noise

noise signal

4

3

2

sample value

sample value

1

0

-1

-2

-3

-4

0 200 400 600 800 1000

sample number

Fig 24 Noise Signal

Advanced DSP Technique to Remove Baseline Noise from ECG Signal

When the spectral density is calculated with and without adaptive filtration the table shows that the 60Hz noise is removed from the database.

Therefore it may be concluded that the choice of the cut-off frequency is very important, however frequency lower than required cut-off frequency, does not filter the actual ECG signal component, while some of the noise successfully, but the ECG signal is distorted in the process. Cut-off frequency varies corresponding to heart rate and baseline noise spectra. Thus, constant cut-off frequency is not always appropriate for baseline noise suppression; it should be selected after a careful examination of the signal spectrum. When FIR filter with wavelet is applied on signal it can be observe that the combination of Kaiser and wavelet yield the smallest phase delay among all the FIR filters combination. It can remove the Baseline noises without distorting the waveform. Furthermore, there is significant delay in the filter result, thus this combination can be applied to long data window. Therefore,

this combination is appropriate only for offline application, but for real time application, in which short intervals of data is filtered and fast implementation is important, FIR is not an appropriate filtering method. IIR and wavelet combination is more appropriate for real time filtering application due to its lower computational complexity, and its better trade-off between average power and spectral density. It completely eliminates the oscillations produced at the starting of the waveform called ringing effect. For performance analysis we use different baseline noise removal methods for the purpose of comparison. The results are presented in the tabulation form. From the table it can be concluded that it outperform the other method.

I would like to express my sincere thanks to my guide Professor R. K. Mehrotra, Department of Electronics and Communication Engineering, Ajay Kumar Garg Engineering College,Ghaziabad for their guidance and support throughout the work.

1. Carr, J. J. and Brown John M. (1998). Introduction to Biomedical Equipment Technology (3rd ed.): Prentice Hall, Inc.

2. Chavan M. S., Agarwala R., and Uplane M.D.,

   Suppression of Baseline Wander and power line interference in ECG using Digital IIR Filter International Journal Of Circuits, Systems And Signal Processing, issue 2,volume 2,2008.

3. Dhillon S. S. and Chakrabarti S., Power Line Interference removal From Electrocardiogram Using A Simplified Lattice Based Adaptive IIR Notch Filter, Proceedings of the 23rd Annual EMBS International conference, October 25- 28, Istanbul, Turkey, 2001, pp.3407-12.

4. Einthoven, W. (1906) Tele-cardiogramme. Arch Int de Physiol; vol 4 pp132-164.

5. Goldberger, et al. (2000) PhysioBank, PhysioToolkit, and PhysioNet: Components of a New Research Resource for Complex Physiologic Signals Circulation 101(23): pp e215- e220 Circulation Electronic Pages; http://circ.ahajournals.org/ cgi/content/ full/101/23/e215.

6. Jane, R. and Laguna, P., Adaptive Baseline Wander Removal in the ECG: Comparative Analysis with Cubic Spline Technique, IEEE, 0276-6547/92, 1992.

7. Jane, R., Laguna, P. Thakor, and Caminal, P. (1992) Adaptive Baseline Wander Removal in the ECG: Comparative Analysis with Cubic Spline Technique, IEEE proceeding Computers in Cardiology, pp 143 146.

8. Jardins, T. D. (2002). Cardiopulmonary Anatomy Physiology (4th ed.).

   Vol. 2 Issue 6, June – 2013

9. Javaid,R.; Besar, R. and Abas, F. S. (2006) Performance

   Evaluation of Percent Root Mean Square Difference for ECG Signals Compression, Signal Processing: An International Journal , vol. 2, issue 2, pp 1-9.

10. Kaur M., and Singh B. Power line Interference Reduction in ECG Using Combination of MA Method and IIR Notch, International Journal of Recent Trends in Engineering, Vol 2, No. 6, November 2009.

11. Norbert, H. (2006) A New Class of Digital Filters Designed for ECG Noise Reduction Medical Informatics & Technologies, Proceedings, pp 44-49.

12. Oppenheim, A. V., Schafer R. W., Discrete-Time Signal Processing, Prentice Hall Publishing Company, 1999.