Modified filtering technique for ECG signal and power line

interference noise
Abstract
A modified multi-stage filtering technique was used in this work to separate the raw ECG signal from the noise that accompanied it. This is
done in an effort to prevent loss and damage to any of the signal's constituent parts. To accomplish this, three typ es of filters were applied
to distinct regions of the signal, as each part is subjected to different types of noise. Initially, the electromyography sig nal interference was
to be eliminated by passing the ECG signal through a modified low-pass filter with a cut-off frequency of 70 Hz. Subsequently, an additional
filter was employed to eliminate the signal's deviation from the baseline. Furthermore, a notch filter was applied to the signal in order to
further reduce power line interference noise. A window was used that had favorable features that could balance the amplitude of the side
lobe and the width of the main lobe, further reducing the harmonics caused by this wave of noise. By comparing the outcomes of this
research to the raw signal, noise on different regions of the signal was effectively decreased. Overall, in the filtered signal of one ECG
recording, the signal-to-noise ratio and mean square error reached 20.2521dB and 0.0021, respectively.
Keywords
ECG, Kaiser window, PLI noise, BW noise, EMG noise

1 Introduction service and such applications and other applications began

Electrocardiogram (ECG) signals have been used to be operated by this network [5, 6].
extensively to identify a variety of cardiac conditions ever In general, the ECG signal represents the electrical
since they were discovered towards the end of the activity of the heart throughout the process of pumping
eighteenth century. That has to do with how effective and blood into the body's arteries [7,8]. This signal can be read
accurate it is at spotting the majority of cardiac conditions, using an electrocardiograph by placing electrodes on the
such as palpitations, shortness of breath, arrhythmias, or patient's chest and limbs for the purpose of collecting ECG
coronary artery disorders. Despite this, heart disease is the records represented by the P wave, QRS complex, and T
primary reason of death in the world, owing to modern- wave. Starting with the atrium and moving toward the
day diseases as well as poor decision-making by ventricle, each segment of the ECG chart shows the
professionals based on incorrect diagnosis [1]. According activity of a particular place in the heart, either through
to estimates provided by the World Health Organization depolarization or repolarization [9].
(WHO), the death rates for these diseases are projected to Since each of these components propagates at a
be 31%, or over 17.9 million fatalities each year [2, 3]. particular speed and frequency, a time interval, such as the
These percentages can be reduced if medical treatment is ST interval and the PQ interval, is created between each of
provided and organized for the patient in a timely and them. The T wave typically has a frequency of 0 to 10 Hz,
effective manner. As a result, an effective technique must whereas the P wave and QRS complex have frequencies of
be provided to retain this signal in good quality, with high 5 to 30 Hz and 8 to 50 Hz, respectively [10]. Despite the
precision, and free of noise for the person diagnosed in fact that this signal provides a wealth of diagnostic
order to grasp the physiological status of the heart well and information, it is highly sensitive to loss or distortion. The
free of diagnostic errors. reason for this is due to the ECG signal's weakness in
In light of this, automated processing of biological data terms of voltage and frequency [11], which makes it more
has grown to be a crucial and critical issue in hospitals and sensitive to various sorts of noise, such as power line
intensive care units during the past few years [4]. This is interference, deviation from the baseline, electrode
due to a number of reasons, such as the need to exert less movement, muscle signal, or interference from other
effort, improve diagnosis accuracy, and complete a devices.
thorough examination of several clinical conditions as The interference of power lines in the ECG signal is
quickly as feasible. Furthermore, automation of these one of the most common types of noise caused by the
signals, particularly the ECG signal, has become a effect of alternating current on this signal [12]. There are
requirement, particularly since the 5G network went into more causes for this type of noise, such as the patient not
being grounded properly because the electrodes are
unclean or in an unfavorable position that prohibits this output instability. The least-mean-square filter is therefore
procedure from being carried out properly [13]. Depending regarded as the best since it has demonstrated its
on the sort of AC system being utilized, this form of noise excellence in eliminating power line interference at 50 Hz.
typically occurs at either 50 Hz or 60 Hz [14]. Saxena et al. opted to use several filtering techniques to
As a result of the aforementioned, it is required to filter remove PLI noise from the ECG data [17]. Five different
the ECG signal from the noise that is connected with it and ECG signal recordings (100, 102, 105, 111, and 115 m)
regenerate it in a way that can reliably read the wave were subjected to the FIR, IIR, and other filters in order to
parameters in order to protect the wave's constituent compare and determine the optimal filtering strategy.
components from loss and damage. This difficulty can be Furthermore, a number of features were examined, such as
overcome by employing a series of filters with various cut- SNR and MSE, in order to validate the effectiveness of the
off frequencies to accommodate various noise sources. filtration process's methods. As can be seen from the
In my earlier proposal [15], I implemented sequential evaluation findings, FIR and IIR perform poorly when
filters to get rid of the muscle signal noise that compared to the other methods utilized, as these two filter
accompanied the ECG signal as well as the wave's types yielded the lowest SNR value when applied to the
deviation from the baseline. This essay is a continuation of ECG data. When the order of the FIR and IIR filters is
that effort. To eliminate the interference from the power increased, however, they exceed the other techniques in
line and present it as an integrated filtering system for the terms of performance and provide the best results in terms
ECG signal, another filter was added to the previous filters of reducing PLI noise. There were certain drawbacks to
in this work. This system is based on the use of various increasing the filter order, such as phase delay and a more
types of filters appropriate for the types of noise to be complex design. As a result, the remaining approaches
filtered while ensuring that no parameter is lost from the were deemed the best in terms of performance since they
original signal. During the work, six recordings of the achieve a balance between good filtration and simplicity.
ECG signal were taken into consideration; each recording Prashar et al. constructed and tested many types of
expressed a certain pathological condition. Additionally, cascade filters to eliminate various sorts of noise [18]. A
the effectiveness of the suggested method was shown by high-pass filter was employed, which was created with
the use of a number of measures, including the Mean several windows such as Hamming, Hanning, Kaiser,
Squared Error (MSE) and the Signal-to-Noise Ratio Rectangular, and Blackman. Furthermore, two distinct
(SNR). At each filtering stage, the Power Spectral Density algorithms are used in the design of the adaptive filter, and
(PSD) was also examined. different approximation techniques are used in the
construction of the low-pass filter. The SNR and PSD
2 Related works
values were examined in order to assess the suggested
In the past few years, various signaling noises on ECG
approach. In comparison to other algorithms and windows,
signals have spurred some researchers to attempt to
the evaluation findings showed that the NLMS algorithm,
discover strategies to decrease this noise. As a result,
Blackman windows, and Elliptic were effective when
signal processing is critical for decreasing noise in relevant
utilized in the filters. Cascade filters were created by
ECG signals and improving their usefulness and quality.
joining these filters in series to filter the ECG signal,
Because of this, Gilani et al. investigated the use of several
resulting in an improvement in SNR of 7.75 dB.
filter types to eliminate PLI noise from the raw signal [16].
Malhotra et al. presented a new method for filtering the
Filters of three types employed to apply to the ECG signal
ECG signal and filtering it from artifacts accompanying
in the time and frequency domains: notch filters, band stop
the raw signal based on real-time wavelet filtering [19].
filters, and least-mean-square filters. The band-stop filter
Depending on this, the signal was divided into several sub-
is built with a cut-off frequency between 47 and 53 and an
signals using the discrete wavelet transform principle. The
order of 2 and 5. It was discovered during the comparison
wide-band signal was passed through several low-pass and
that, in contrast to the high-order band stop filter, the
high-pass filters to separate this band into several
conventional notch filter was unable to eliminate the PLI
frequency bands. In addition, IIR and FIR filters were
when the power line's frequency was changed. However,
implemented for the purpose of comparing them with
one of the most significant disadvantages of the band-stop
wavelet-based filtering approaches. The results indicate
filter is the loss of information at a particular range and the
that the proposed approach is more effective and more
stable under the same conditions compared to other filters the SNR and MSE values, it was concluded that the
in terms of removing deviations from the baseline. This is treatment algorithms based on Newton polynomials
because the approach based on wide band separation does perform well compared to other algorithms because they
not depend on the band transition but rather depends on have greatly improved the ECG signal.
the sampling frequency, which makes the filtering process Ulla et al. recommended employing multi-stage filters
more stable. and different algorithms to remove various types of noise
Manju et al. offered two proposals for filtering the ECG [23]. In light of this, the performance of various
signal from various sorts of noise using two types of algorithms was measured in terms of the percentage
filters: the Wiener filter and the Kalman filter [20]. The improvement in the SNR value of the ECG signal. Based
raw data has been subjected to these two filters in order to on the outcomes, it has been found that one of the
remove five types of noise: Gaussian power line algorithms performs well in terms of filtering the raw
interference, muscle artifacts, baseline wander, and signal from baseline drift and muscle artifacts. A higher
composite noise. A variety of characteristics, including SNR of 12.7319 dB was achieved by the filtered signal in
MSE, SNR, and PSD, were examined in order to confirm one of the techniques.
how well each filter proposal performed in terms of Samsudin et al. developed a low-pass filter with
eliminating artifacts. According to the results of the frequencies of 5 and 25 for the pass band and stop band
investigation, the Wiener filter outperforms the Kalman regions, respectively [24]. Throughout the course of the
filter in terms of performance. For example, the Wiener work, the direct form I structure and the direct form II
filter produced an SNR of 7.5505 dB for one type of noise, structure were the two distinct concepts that were put out.
while the Kalman filter produced an SNR of 5.5340 dB for The objective is to filter the ECG signal and assess how
the same noise. This is because there are two types of well these two designs use resources compared to one
systems that can affect the Kalman filter: linear and another. This was accomplished by analyzing a number of
nonlinear. Its use is more restricted in the latter. components, including flip-flops, digital signal processing,
Wisana et al. designed several IIR digital filters, such as and look-up tables. The Direct Form II results
Butterworth, Chebyshev, and Elliptic, to filter the ECG demonstrated that the increase in resource usage was
signal efficiently and at a low cost [21]. By analyzing the directly impacted by sharing the delay factors between
effectiveness of these filters separately, it was concluded input and feedback. Because of this, Direct Form II was
that using a Butterworth filter with order 8 and an able to employ Look Up Tables, Flip Flop, and Digital
appropriate cut-off frequency is the most effective in terms Signal Processing resources at a rate of 0.26%, 0.12%, and
of filtering because it has good bandwidth. Whereas using 2.50%, respectively, more efficiently than Direct Form I.
a Chebyshev filter with order 6 produced an emphasis of - Reddy et al. made a comparison between FIR and IIR
3.27104076 dB, increasing the filter order to 8 produced filters using different sampling frequencies to filter the
an emphasis of -5.08730424 dB. In the same context, ECG signal [25]. The aim is to evaluate the effectiveness
when using the elliptic filter with an order of 6, it of obtaining a linear phase response without phase
produced an emphasis of 1.55429354 dB. distortion. The results obtained showed that using a Kaiser
Altay et al. presented a proposal to improve the SNR window in the FIR filter gives more flexibility than the IIR
and MSE of the ECG signal based on Newton filters [22]. filter and improves the complexity of the design in
Different algorithms for filtering the raw signal were addition to improving the SNR.
adopted in this work for the purpose of removing high- Ardani et al. proposed designing a filter with a Kaiser
frequency and low-frequency interference. The effect of window to increase the speed of removing noise from the
the low-frequency component was reduced by developing ECG signal [26]. Several parameters were analyzed to
a high-frequency Newton filter to remove the deviation evaluate the working efficiency of the proposed
from the baseline resulting from this type of frequency. In technology, including power consumption and operating
addition, the effect of narrow-band electrical interference frequency. The results obtained showed a good increase in
was reduced through the use of cascade Newton's notch filtration speed and lower power consumption.
filters. Furthermore, cascades of Newton notch filters have The ECG data is treated in this study so as to suppress
been developed to reduce the effect of broadband current excessive noise from its components. This is accomplished
interference. Through the results obtained from analyzing by applying a multiple-phase filter to the noisy ECG
signal, eliminating a particular kind of noise at each level. 50 Hz, this filter is built with cut-off frequencies of 49 Hz
As a first stage in a feasibility investigation, the filtered and 51 Hz. Furthermore, the sample frequency of this filter
signal was assessed and contrasted with the noisy signal is set at 1500 Hz, with a filtering order of 100. In addition,
and many other methods to determine if this offered the window method was applied in this design with the
superior filtering of the ECG data. intention of perfectly filtering the signal and putting an
The rest of this essay is structured as follows: Section 3 end to the infinite harmonics that would otherwise come
presents the suggested method. In Section 4, the outcomes from this kind of noise.
of the suggested work are presented along with their When using the window approach, there is frequently a
analysis. The fifth section is where the conclusions are trade-off between the width of the main lobe and the
provided. amplitude of the side lobe, and this occasionally results in
the selection of a window with unfavorable traits. The
3 Proposed method
Kaiser window was selected after the aforementioned
This section demonstrates the recommended approach
issues were taken into consideration during the design
for filtering the ECG signal sequentially to remove various
process. This window was chosen for its adaptability
noise types. Three consecutive filters are used in three
since, if the window's length is set, certain of its
stages of this approach. Since the second step was
parameters can be adjusted to establish a balance between
extensively discussed in my previous work, this part
the amplitude of the side lobe and the width of the main
concentrated on the first and third stages. The proposed
lobe. According to the Eq. (1), the Kaiser window is as
approach for subsequent processing filters is shown in
follows [27]:
Fig.1. Each stage is actually a filter with unique design
(1)
requirements that are appropriate for its intended usage.
Power line interference noise at 50 Hz is one of the most Where,
common types of noise that interacts with the ECG signal N: Length of the window.
and causes signal constriction. 𝐼0: Modified Bessel function.
𝛽: Non-negative real number that determines the shape
of the window.

By adjusting the value of the parameter 𝛽, a suitable

trade-off between side lobe amplitude and main lobe width
can be achieved in design. Where the parameter β is
directly proportional to the width of the main lobe and
inversely proportional to the amplitude of the side lobe.
The amplitude and phase responses of the notch filter
utilized at this stage are shown in Fig. 2. The efficiency of
the performance in the filtration process is demonstrated
by the fact that the filter's amplitude is as high as possible
at 50 Hz and its energy is as low as possible at the same
Fig. 1 Block diagram of the proposed technique. frequency, as seen in the figure.

As a result, a Finite Impulse Response (FIR) notch

filter was utilized in this work, which attenuates
frequencies in a certain range while passing without
effecting the other frequencies. The reason for not
employing the Infinite Impulse Response (IIR) in this
design is owing to the FIR's capacity to filter the signal
without destroying it and to provide superior filtering
results, as opposed to the IIR, which destroys the signal's
phase characteristics. In order to eliminate interference at
 different medical condition. These recordings (100m,
101m, 103m, 104m, 105m, and 106m) were taken from
the laboratories of Beth Israel Hospital in Boston. The
purpose of reading only these recordings is to compare
them with my previous work proposed in [15]. In the
previous work, multiple filters were proposed for the
Fig. 2 Amplitude and frequency response of the proposed notch filter. purpose of filtering out the ECG signal from the EMG
signal and the deviation from the baseline only.
On the other hand, two other filters were used in this The previous work is considered incomplete in terms of
work, namely the low-pass filter and the zero-phase shift the filtering process when compared to this work. The
filter, in the first and second stages, respectively. A low- signal was partially filtered out of only two types of noise,
pass filter with the Butterworth approximation, which is and the power line interference noise was ignored without
characterized by having a flat frequency response without any action taken to filter out this type of noise. The
any changes in the passband, was employed in the first emphasis of this work was on introducing a new method
stage of the process to remove the noise in the EMG that complements previous work based on the entire
signal. In the second stage, when removing the deviation separation of the different types of noise to which the ECG
from the baseline, the zero phase shift filter was used, signal is exposed. A third filter was applied to this signal
which uses future information to remove the phase to filter it from power line interference noise, which is one
distortions of the signal. of the most common sources of noise in this sort of
To ensure the practicality of the suggested method, the transmission. The goal is to offer a comprehensive method
effectiveness of the used filters should be evaluated in for creating a collection of filters that rid the raw signal of
terms of their ability to eliminate noise from the ECG the majority of types of noise to which it is subjected.
signal. As a result, Eq. (5) was used to calculate the MSE In order to validate the suggested work's efficiency, it
between the original and filtered ECG data. In addition, was compared to the approach proposed by B. Liu et al.
The SNR was calculated using Eq. (3) [15]. [28] in 2021 and the method proposed by S. Basu et al.
[29] in 2020. A number of measurements, including PSD,
2 SNR, and MSE, were analyzed in order to ascertain the
∑𝑁
𝑖=1(𝑆(𝑖))
𝑆𝑁𝑅𝑜𝑢𝑡 = 2 (2) (1
∑𝑁
𝑖=1(𝑛𝑜𝑖𝑠𝑒(𝑖)) effectiveness of the suggested technique. The results are
detailed and thoroughly explained for each of the three
where, suggested filtering stages in the next two subsections.
SNRout : SNR of the de-noised ECG signal. 4.1 The first and second stage of filtration:
EMG noise is taken out of the raw signal at the first
N: size of the signal.
stage of filtering. The cut-off frequency in my previous
S(i) : Input ECG signal. research [15] was set to 100 Hz, but as expressed in the
𝑛𝑜𝑖𝑠𝑒(𝑖) = 𝑆(𝑖) − 𝑆̂(𝑖) (3) introduction section, the greatest frequency that the ECG
where, signal can reach is 70 Hz. In addition, as can be seen in
Fig. 3, while comparing the PSDs of the raw and filtered
Ŝ (i): Filtered ECG signal. signals, it was discovered that even after filtering the
𝑆𝑁𝑅𝑜𝑢𝑡 𝑑𝐵 = 10𝑙𝑜𝑔10 𝑆𝑁𝑅𝑜𝑢𝑡 (4) signal at 100 Hz, there was still a spectral energy
distribution present at frequencies higher than the cut-off
∑𝑁 ̂
𝑖=1(𝑆(𝑖)−𝑆 (𝑖))
2 frequency.
𝑀𝑆𝐸𝑜𝑢𝑡 = 𝑁
(5)

4 Result and discussion

The numerical outcomes of the suggested technique
mentioned in Section 3 are reported in this section. The
first stage of this effort involved reading multiple
recordings of the ECG signal, each signal representing a
 14.514 dB and 0.021, respectively, as shown in Table 1.
The zero phase shift filter (second stage) design
specifications have not been altered in this work because it
has demonstrated its effectiveness in removing the ECG
signal's deviation from its baseline. As a result, the
existing design specifications for the second stage of
filtering were used in this study.
4.2 The third stage of filtration:
Power line interference is one of the many
abnormalities that can taint an ECG recording and lessen
Fig. 3 Comparison of the PSDs of the raw signal and the filtered the accuracy of a diagnosis. As was noted in the
signal at fc = 100 Hz. introduction, this kind of noise is distinguished by its
The power spectral distribution is an expression of how interference with the ECG signal at a constant frequency
concentrated the signal is at certain frequencies. As a of 50 Hz. The P and T waves become indistinguishable
result, the distribution depicted in Fig. 3 was revisited in due to this type of noise's effect on the ECG signal's
this work with the goal of filtering the signal more breadth and narrowing. After the first and second stages of
correctly and clearly. As a consequence, frequencies filtering, power line interference is still present in the
greater than 70 Hz were regarded as noise frequencies that signal. In this work, a notch filter (consisting of a low-pass
needed to be reduced. Therefore, It was decided to remove filter and a high-pass filter) is built with a cut-off
the EMG signal at this frequency by redesigning the frequency of 51 Hz for the high-pass filter and a cut-off
Butterworth approximation low-pass filter with a cutoff frequency of 49 Hz for the low-pass filter. As seen in Fig.
frequency of 70 Hz. 5, the signal that this filter cleaned is shown. This step (the
A comparison of the raw signal and the filtered signal is third stage), as seen by the figure, saw the total elimination
shown in Fig. 4. As can be seen from the figure, the of this kind of noise.
filtered signal is clearer. This is due to the removal of
muscle signal noise from the raw signal after using a low-
pass filter.

Fig. 5 Filtered ECG signal after the last stage.

In addition, the values of SNR and MSE of the 104 m

record after this stage are 20.2521 and 0.0021,
respectively. Table 1 lists the SNR and MSE values for all
Fig. 4 Comparison of the raw signal and the filtered signal after the of the ECG signal recordings that were examined during
first stage. this study after each of the three filtering stages.
In order to verify the efficiency of the filtering process, Table 1 SNR and MSE of the proposed technique after each stage.
the SNR and MSE values of the filtered signal were
calculated and compared with the raw signal. The results In order to confirm the effectiveness of the suggested
showed a clear improvement in the signal quality, as the approach, the final outcomes of this work were also
values of SNR and MSE of the 104 m record became
contrasted with those obtained using other approaches. whether the suggested technique was effective in lowering
This comparison is described in Table 2. It is evident that the noise ratio in comparison to the raw signal, a number
the proposed method performs better than previous of parameters, including SNR, MSE, and PSD, were
methods in terms of the quality of the filtered signal. A examined throughout the evaluation.
schematic comparison of the suggested method with my The work's findings led to the conclusion that, in
previous work on cooperative filtering methods, and comparison to other approaches, this one works well since
methods without cooperative filtering proposed by B. Liu the signal purity was significantly increased. The updated
et al. [28] can be found further in Fig. 6. method has been found to have made a total contribution
of three stages toward decreasing the noise level in the
Table 2 A comparison between the proposed technique and other
various signal components. In the first stage of the work,
the SNR of one of the recordings was boosted to 14.514
techniques in terms of SNR.
dB, and in the second stage, it was increased to 17.914 dB.
In contrast, the MSE decreased 0.021 in the first stage and
to 0.0044 in the second. The PLI was removed at the third
step of the procedure, and its harmonics were removed as
well, keeping in mind the Kaiser window's ability to
balance the main lobe's width and the side lobe's
amplitude. After this stage, there was a 20.2521 dB rise in
SNR. As a result, if the modified method that has been
suggested is used, it will provide an entirely new approach
for analyzing the ECG signal because it has significantly
reduced noise on the majority of the signal.

Fig. 6 Comparison of SNR in the proposed method with multistage,

cooperative and without cooperative filtering.

5 Conclusion
In this article, a modified approach was used in order to
reduce the quantity of noise to which various sections of
the ECG signal are subjected. Since it serves as the
foundation for separating meaningful information from
noise in the signal, the spectral distribution of the signal
was thoroughly examined before any filtering was applied.
A modified filtration process with three successive phases
has been presented as a result of this. To determine