Periodic Noise Suppression from ECG Signal using
Novel Adaptive Filtering Techniques
Yogesh Sharma 1, Anurag Shrivastava 2 1-Department of Electronics and Telecommunication,
Shri Shankaracharya College of Engineering & Technology, Bhilai, Chhattisgarh, India Email- yogeshetc@gmail.com
2- Department of Electronics and Instrumentation,
Shri Shankaracharya College of Engineering & Technology, Bhilai, Chhattisgarh, India Email- anurag_256@rediffmail.com
Abstract- Electrocardiogram signal most commonly known recognized and used biomedical signal for medical examination of heart. The ECG signal is very sensitive in nature, and even if small noise mixed with original signal, the various characteristics of the signal changes, Data corrupted with noise must either filtered or discarded, filtering is important issue for design consideration of real time heart monitoring systems.
Various filters used for removing the noise from ECG signals, most commonly used filters are Notch Filters, FIR filters, IIR filters, Wiener filter, Adaptive filters etc. Performance analysis shows that the best result is obtained by using Adaptive filter to remove various noises from ECG signal and get significant SNR and MSE results. In this paper a novel adaptive approach by using LMS algorithm and delay has shown which can be used for pre-processing of ECG signal and give appreciable result.
Keywords: Adaptive filtering, ECG, LMS Algorithm, Adaptive Noise Canceller (ANC).
1. Introduction
ECG is the graphical recording of the electrical activity of the heart, it is by far the most easily recognized biological signal, and it is also the one that is most commonly used for clinical diagnosis, and more important is the fact that the ECG wave shape is altered by cardiovascular diseases and abnormalities, each portion of the ECG waveform carries information that is relevant to the clinician in arriving at a proper diagnosis. The Electrocardiograph signal taken from a patient was corrupted by an external noise, so that we need a proper way to get a noise free ECG signal. Some of the characteristics of these signals are the frequency and morphology of their waves. These components are in the order of just a few up to 1mV and their frequency content within 0.5Hz and 100Hz depending on individual. The ECG signal is generally mixed with other biological signals like electroencephalogram, Electro-mayo gram, baseline Wander and power line interference. Due to the presence of artefacts, it is difficult to analyze the ECG, for they introduce spikes which can be confused with cardio logical rhythms. Thus, noise and undesirable signals must be eliminated or attenuated from the ECG to ensure correct analysis and diagnosis.
Adaptive filters are capable of separating interference components from the signal of interest, even under the case when interference components are overlapped to the signal in frequency. Furthermore, adaptive filters can process signals in real time and hence, provide good tools for time-critical applications. In this paper, the removal of random noises in ECG with a novel LMS is addressed.
IJECSE,Volume1,Number 2
Yogesh Sharma and Anurag Shrivastava
schemes, such as those based on empirical mode decomposition [9], ensemble empirical mode decomposition [10] and on the evaluation of higher-order statistics at different wavelet bands [11] can also be used for the same applications.
2. Conventional Method
Conventionally used ANCs basically require two inputs: primary input and reference input. For an ANC to be feasible, the reference input must be correlated with the noise part of the primary input in order to cancel the noise therein. However, there are cases in which commonly used ANCs are limited in use. For example, if off-line processing of ECG recordings is required, the recordings themselves are the only data available for processing. Moreover, in the case when reference input cannot be well-correlated with the noise part of the primary input, ANC cannot perform well as the case with a well correlated reference input. In such cases, alternative approaches other than an ANC must be pursued. Figure 1 shows the example of conventional model of adaptive noise canceller with reference input.
Figure 1. Conventional model of ANC [12].
The primary input is denoted as:
x(n) = S1(n) + n1(n) (1)
Where S1 is the signal and n1 is the noise. Under this condition the error signal becomes :
e(n) = x(n) – y(n) (2)
e(n) ≈ S1(n) (3)
Hence the error signal is the estimation of clean ECG signal.
3. Problem Identification
The conventional method works only when the reference input is well-correlated with the noise part of the primary input, i.e. n2. To achieve this goal, a lead need to be placed somewhere from which the correlation in between is maximum. This could be a problem because to have such a reference input may not be an easy task. The reason why this may happen is at least twofold. First, the component picked up by the lead for the reference input may correlate in some extent to the signal besides the noise. The path connecting the component picked up by the sensor to the reference input may cause distortion. As such, in reality, it could be difficult to have a reference input that is well-correlated with the noise part, while the signal part remains uncorrelated. This tells that the ANC may degrade its performance under limited availability of the reference input.
4. Proposed Method
x(n) = S1(n) + n1(n) x(n-d)
+
y(n) = s1’(n-d)
-
e(n)
Figure 2. Delay inserted in primary input.
x(n) = S1(n) + n1(n)
y(n) = s1’(n) +
x(n-d) -
e(n)
Figure 3. Delay inserted in reference input.
5. Learning Algorithm
Typical least mean square (LMS) algorithm is employed for updating the tap-weights of the adaptive filters. For example, in Figure 3, the adaptive filter output is of the form:
M-1
y(n) =
∑
wk(n) x(n-d-k) (4)k=0
where M is the number of taps of the adaptive filter. The error signal e(n) is:
e(n) = x(n) - y(n) (5)
The tap-weights of the adaptive filter is updated according to the rule:
wk (n +1) = wk (n) + µe(n)x(n – d - k) (6)
where k = 0,1, ...M-1 and µ is the step size controlling the speed of convergence. After completing the learning, the error signal becomes:
e(n) = x(n) - y(n) = s1(n) + n1(n) – s1’(n) ≈ n1(n) (7)
Hence the output of filter is the clean ECG signal. Delay
Adaptive Filter
IJECSE,Volume1,Number 2
Yogesh Sharma and Anurag Shrivastava
Figure 4 Simulation of ECG signnal for (a) Original ECG (b) 50 Hz noise signal (c) Noisy signal (d) Error signal (e) Filter output.
6. Simulation Results
Figure 4 shows the ECG signal and the 50 Hz noise signal simulated by using MATLAB software. Both the signals are mixed and passed through the filter. The mean-square-error (MSE) is calculated between the clean ECG signal and filtered ECG as follows:
MSE =
∑
{ 1
−
−
}
2 (8)With delays in primary input and
MSE =
∑
{ 1
−
}
2 (9)with delays in reference input.
Here ‘m’ is the number of samples within the segment where adaptive filter converged. Table 1 compares the result of the proposed filter with the output of adaptive filter with well correlated reference input for different filter lengths. It is seen that the output of the filter with delay in primary path performs closely to the ANC with well correlated reference input provided that the length M is large enough. However for the same result the filter with delays in reference input requires much larger value of M. It is suggested that the one using ANC with delays in the primary input performs better than that with delays in the reference input due to the fact that much shorter M is adequate for the former.
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 -1
0 1
Original ECG Signal (a)
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 -1
0 1
50 Hz Noise Signal (b)
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 -2
0 2
Mixed Signal (c)
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 -5
0 5
Error Signal (d)
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 -2
0 2
Table 1. Mean-Square-Error (MSE) comparison F ilter length (M) Μ Μ Μ ΜSE(dB)
ANC with well
correlated reference input
ANC with
delays in primary path
ANC with
delays in reference path
2 0
-29.78 -22.31 -14.57
3 0
-30.09 -27.53 -14.70
4 0
-30.56 -27.47 -15.94
5 0
-29.26 -27.38 -21.45
6 0
-30.44 -28.15 -27.01
7 0
-29.43 -27.59 -27.09
7. Conclusion
The issue of using ANC with limited availability on the reference input has discussed and the alternative of this is presented in this paper. Reference input not well correlated with the noise or signal part of the primary input may degrade the performance of the ANC. ECG signals can be treated as quasi-periodic signals. By shifting the noisy ECG signal in time, the noise component can become uncorrelated while the signal component still remains correlated. Thus the smooth ECG is obtained in the output which is comparable with the output of conventional ANC with well correlated reference input. Computer simulation results confirmed the effectiveness of proposed method.
References
[1] Widrow B, Glover JR, Jr., Mccool JM, Kaunitz J, Williams CS, Hean RH,Zeidler JR, Dong E, Jr., Goodlin RC (1975). Adaptive Noise
cancelling: principles and applications. Prodeed. IEEE. 63:1692- 1716
[2] Thakor NV, Zhu YS (1991). Applications of adaptive filtering to ECG analysis: noise cancellation and arrhythmia detection. IEEE Trans.
Biomed. Eng., 38: 785-794.
[3] Mehrkanoon S, Moghavvemi M, Fariborzi H (2007). Real time ocular and facial muscle artifacts removal from EEG signals using LMS
adaptive algorithm. International Conference on Intelligent and Advanced Systems. pp. 1245-1250.
[4] Wu Y, Rangayyan RM, Zhou Y, Ng SC (2009). Filtering electrocardiographic signals using an unbiased and normalized adaptive noise
reduction system. Med. Eng. Phys., 31: 17-26.
[5] Slim Y, Raoof K (2010). Removal of ECG interference from surface respiratory electromyography. IRBM, 31: 209–220.
[6] Rahman MZU, Shaik RA, Reddy DVRK (2011). Efficient sign based normalized adaptive filtering techniques for cancelation of artifacts in
ECG signals: application to wireless biotelemetry. Signal Process.
[7] Correa AG, Laciar E, Patino HD, Valentinuzzi ME (2007). Artifact removal from EEG signals using adaptive filters in cascade. Proc. Of
the 16th Argentine Bioengineering Congress and the 5th Conference of Clinical Engineering. pp. 1-10.
[8] Suresh HN, Puttamadappa C (2008). Removal OF EMG and ECG artifacts from EEG based on real time recurrent learning
algorithm.International J. Phys. Sci., 3: 120-125.
[9] Weng B, Blanco-Velasco M, Barner KE (2006). ECG denoising based on the empirical mode decomposition. Proc. of the 28th IEEE.
Annual International Conference on Engineering in Medicine and Biology Society (EMBS’06). pp.1-4.
[10] Chang KM (2010). Arrhythmia ECG noise reduction by ensemble empirical mode decomposition. Sensors, pp. 6063-6080.
[11] Sharma LN, Dandapat S, Mahanta A (2010). ECG signal denoising using higher order statistics in wavelet subbands. Biomed. Signal
Process. Control, 5: 214-222.
[12] DV Rama Koti Reddy, MZU Rahman, Y Sangeetha, N. S. Sudha, Baseline Wander and PLI elimination from cardiac signal, IJAEA , Jan
2011.
[13] Mbachu C.B , Onoh G. N , Idigo V.E , Ezeagwu C., Ezechukwu O.A , suppression of PLI using adaptive digital filter, IJEST, Vol.3, No. 8,
Aug. 2011.
