Cluster Computing
https://doi.org/10.1007/s10586-017-1602-0
FPGA implementation of modified error normalized LMS adaptive
filter for ECG noise removal
C. Venkatesan
1
· P. Karthigaikumar
2
· R. Varatharajan
3
Received: 20 October 2017 / Revised: 15 December 2017 / Accepted: 20 December 2017
© Springer Science+Business Media, LLC, part of Springer Nature 2018
Abstract
High frequency noise and channel noise are dominant in wireless ECG monitoring systems which can be modeled as white
Gaussian noise. Least mean square (LMS) algorithm based adaptive filters are the preferred choice for white Gaussian noise
removal, because they require fewer computations and less amount of power consumption. Though LMS algorithm is simple
to implement in real time systems, it is necessary to modify the LMS algorithm to reduce the mean square error for improved
filtering performance. In this paper, a delayed error normalized LMS (DENLMS) adaptive filter is studied with pipelined
architecture to remove the white Gaussian noise from ECG signal. The pipelined VLSI architecture is utilized to boost the
operational speed of adaptive filter by reducing the critical path using delay elements. The performance of pipelined DENLMS
algorithm is compared with ENLMS and DNLMS algorithms. The pipelined DENLMS filter increases the speed of operation
and reduces power consumption at the cost of increase in area due to the presence of latches. Virtex 5 FPGA XC5LVX330 Field
programmable gate array has been utilized as target chip to determine the speed, logic utilization and power consumption.
Keywords Adaptive algorithms · Gaussian noise · Field programmable gate arrays · Pipeline processing · Least mean square
methods
1 Introduction
Electrocardiogram (ECG) signal preprocessing is essen-
tial in biomedical applications to remove unwanted noisy
components and to enhance the desired feature extraction.
High-quality ECG data is required for improved clinical
diagnosis and better treatment of cardiac diseases. How-
ever, in real situations, the recorded ECG signals are affected
by baseline wander, motion artifacts and high frequency
noise. High-frequency noises are occurred in ECG signals
recordings due to power line interferences, electromyogram
(EMG) noise and electrodes. Baseline wander arises due to
the fact that there is a possible movement of either patients
or recording cables. Moreover, ECG signals are affected by
B C. Venkatesan
venkatintphd@gmail.com
1
Faculty of Information and Communication Engineering,
Anna University, Chennai, Tamilnadu, India
2
Department of Electronics and Communication Engineering,
Karpagam College of Engineering, Coimbatore, Tamilnadu,
India
3
Department of Electronics and Communication Engineering,
Sri Ramanujar Engineering College, Chennai, India
white noise due to poor channel conditions in biotelemetry
applications [1]. In view of these interruptions and noises dur-
ing acquisition, ECG preprocessing techniques are focused
mainly towards noise removal. Several ECG enhancement
methodologies have been discussed in the existing woks that
are mainly concentrated on suppression of noises such as
baseline wander, high frequency noise and white noise using
various signal processing tools and algorithms. Some of the
techniques are advanced averaging, wavelet transform, non-
linear filter banks and adaptive filters [2,3].
In telemedicine applications, transmission of ECG sig-
nals over a wireless channel is often affected by noises due to
improper channel. The noises are normally modeled as white
Gaussian noise [4]. In some applications, white Gaussian
noise is considered as a general frequency noise compo-
nent which is intentionally included to the uncontaminated
ECG signals. Chang and Liu [5] proposed a white Gaussian
noise removal technique using Wiener filtering approach [5].
Wiener filter is an optimal filter and is mainly used in place of
traditional low pass filters where the input signal power spec-
trum and noise spectrum cannot be separated [6]. Extended
Kalman filter (EKF) is an alternative filtering approach to
123