Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Nannan Zhang, Zedong Nie, Yu Luo, Leilei Du, Xiaohui Wang, and Lei Wang
Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences
Shenzhen Key Laboratory for Low-cost Healthcare
Shenzhen, China
wang.lei@siat.ac.cn
Abstract—Dynamic Electrocardiograph (ECG) monitoring by muscle tremor are another type of electrical noise in ECG
(known as Holter) plays an important role in the earlier systems. Besides, there is a wide-band noise generated in all
detection and diagnosis of various cardiovascular diseases. ECG stages of the ECG signal processing.
signals obtained from Holter systems normally contain a lot of
noises and artifacts. These noises degrade signal quality, which In Holter systems the noises mentioned above are very
may be critical for routine monitoring and diagnosis. To solve strong. This makes the ECG de-noising task notably
the problem, a reconfigurable overlapping fast Fourier challenging and important. Body sensor networks (BSN) are
transform/ inverse fast Fourier transform (FFT/IFFT) filter for emerging cyber-physical systems, which improve the quality
suppressing the power-line interference and the high-frequency of life. FFT hardware accelerates the processing units [3].
noise is presented in this paper. The filter is based on a 12-lead Processing on-node decreases the amount of communicated
Holter system with a high-performance analogue front-end and data [4]. Based on it, the ECG processing can be done in real
a field-programmable gate array (FPGA) for enhanced digital time and carried out by a low-voltage and low-power mobile
processing. This paper analyzes the performance of the healthcare device (such as a Holter).
reconfigurable overlapping FFT/IFFT filter in ECG de-noising
applications and validate it by real-world emulations. In the literature, several methods were proposed to remove
Furthermore, the de-noising performance of the reconfigurable a single noise component from ECG signals. References [5]-[7]
overlapping FFT filter was evaluated. suggest methods for removing the power-line interference.
References [8]-[9] suggest methods for suppressing the
Keywords—Filter; Reconfigurable; Overlapping FFT; ECG baseline wandering. Each of the algorithms suggested in the
de-noising; FPGA above-mentioned papers can be used to suppress only one
particular noise component. However, in the real world, the
I. INTRODUCTION ECG signals contain different types of noise. Furthermore,
the cut-off frequency of the employed filters is fixed and this
The demand for improving the quality of life and the rising
makes them unsuitable when variable-frequency noises are
healthcare costs lead to new requirements towards the
present.
healthcare services worldwide. There is increasing interest in
home healthcare that allows to shorten hospital stay and to In this study, we introduce a reconfigurable overlapping
reduce the associated costs. This means that there is a shift FFT/IFFT filter for de-noising of ECG signals on the nodes.
from hospital care to home care for some specific tasks, such This filter can suppress power-line interference and high-
as long-term continuous electrocardiograph (ECG) monitoring. frequency artifacts. An analysis of the filter is also presented.
Compared to traditional approaches, the processing of the
ECG signal analysis is the most widely used technique for
ECG signal on the nodes is much more convenient and
non-invasive diagnosis of various cardiovascular diseases.
efficient.
ECG signals are always mixed with noises whose sources are
not related to the heart and its activity. These noises include,
but are not limited to, power-line interference, baseline II. METHOD AND DESIGN
wandering, high-frequency artifacts, and other biomedical
signals produced by the human body. The most common type A. The hardware system
of disturbance is the power-line interference. The human body In the 12-lead Holter system, we employed the ADS1298
absorbs electromagnetic radiation from power lines and this is analogue front-end chip from TI, which is not yet commonly
associated with noise signals that contain a 50 Hz sinusoid (in used in commercial Holter products. A highly efficient data-
China and Europe) and its harmonics [1]. Human body management software module was designed to allow data
motion causes baseline wandering in the recorded ECG signal. exchange between the ADS1298 and the microprocessor
Its frequency band is much lower than the band of the ECG (STM32L151 from ST Microelectronics). Besides, an FPGA
signal which normally is in the range 0.5-40 Hz in monitoring (Spartan3E from Xilinx) module was utilized which allows to
systems [2]. The electromyography (EMG) signals produced
extend the functionality. The complete electronic system was time domain signal are removed and the final signal is passed
powered by a single 1.8 V supply. It is presented in Fig. 1. to the output.
The role of the overlapping parts is to decrease the pass filter. The output signal of the band-pass filter is shown
frequency aliasing that resulted from the lack of periodicity in in Fig. 5c.
the block data and the non-periodic signal in the FFT. Since
the overlapping parts that contain large frequency aliasing are
abandoned, the outcome is satisfying.
N-point FFT is performed for each series, including the
overlapping parts. The step of each series is L, and the length
of the overlapping part is M.
After the frequency conversion in the frequency domain,
N-point IFFT is performed to convert the signal from the
frequency domain to the time domain. Then, overlapping parts
of the signal in the time domain are abandoned, and the blocks
formed from the remaining signals are used to construct the
output signal of the filter. The percentage of overlapping parts
is shown in Table 1. The overlapping FFT/IFFT filter
processing is illustrated in Fig. 4.
Fig. 6. (a) Spectrum of the original ECG signal (b) Spectrum of the ECG
signal from the comb filter (c) Spectrum of the ECG signal from the
band-pass filter.
A. Comb filter
ACKNOWLEDGMENT
This study was financed partially by the Projects of
National Natural Science Foundation of China (Grant Nos.
60932001, 51105369 and 61072031), the National 863
Program of China (Grant No. 2010CB732606) and the
'Low-cost Healthcare' Programs of Chinese Academy of
Sciences.
REFERENCES
[1] G. Bonneau, M. Bertrand, R. Guardo, M. Lafortune, and A. R. Lerlanc,
“A preprocessor for on-line data reduction in accelerated playback of
ECG's, ” IEEE Transactions on biomedical engineering, vol. BME-30,
pp. 197 - 200, Mcrch 1983.
Fig. 8. Amplitude-frequency response curve of the band-pass filter.
[2] Vrudhula K. Murthy, Thomas M. Grove, George A. Harvey and L.
Figure 8 shows the amplitude-frequency characteristic of Julian Haywood, “Clinical Usefulness Of ECG Frequency Spectrum
the band-pass filter. The –3 dB bandwidth is 40 Hz. The Analysis, ” The Second Annual Syposposium on Computer Application
in Medical Care, pp. 610- 612, November 1978.
highest side lobe is –16 dB at 490 Hz, and the energy of the
[3] B. H. Calhoun, J. Lach, J. Stankovic, D. D. Wentzloff, K. Whitehouse,
ECG component contained in this lobe is too small. Therefore, A. T. Barth, J. K. Brown, Q. Li, S. Oh, N. E. Roberts, and Y. Zhang,
this side lobe can be ignored. In the range adjacent to the cut- “Body Sensor Networks: A Holistic Approach From Silicon to Users,”
off frequency, the amplitude of the highest side lobe is –32 Proceedings of the IEEE, vol. 100, pp. 91–106, January 2012.
dB at 60 Hz which is as good as the Hannning window at =1 [4] Proceedings of the IEEE, vol. 100, no. 1, pp. 91–106, 2012.Diana Albu,
Johan Lukkien and Richard Verhoeven, “On-node Processing of ECG
(Table 1). The rectangular window used in our filter is Signals,” 2010 7th IEEE Consumer Communications and Networking
evidently simpler to implement than the Hanning window, but Conference, pp. 1-5, January 2010.
has almost the same characteristics. [5] Mohammed Bahoura and Hassan Ezzaidi, “FPGA-Implementation of
Wavelet-based De-noising Technique to Remove Power-Line
C. Performance comparison Interference from ECG Signal, ” 10th IEEE International Conference
on Information Technology and Applications in Biomedicine, pp. 1-4,
In our method the most computationally intensive stages November. 2010.
are the FFT and IFFT. In our experiment, the de-noising of a [6] Yue-Der Lin and Yu Hen Hu, “Power-Line Interference Detection and
10-second ECG waveform, at 1 MHz clock of the FPGA unit, Suppression in ECG Signal Processing, ” IEEE Transactions on
took 0.17 s. The efficient recursive least-squares (RLS) biomedical engineering, vol. 55, pp. 354 - 357, January 2008.
adaptive notch filter suggested in the Yue-Der Lin's work [4] [7] Mohammed Bahoura and Hassan Ezzaidi, “FPGA-Implementation of a
showed a one-second delay for execution of the same task. Sequential Adaptive Noise Canceller using Xilinx System Generator,”
2009 International Conference on Microelectronics, pp. 213-216,
Mohammed Bahoura’s filter demonstrated a delay of 0.175 s Decmerber 2009.
between the output and the input [5]. The time needed for [8] Shubhajit Roy Chowdhury and Dipankar Chakrabarti, “Daubechies
processing by our suggested algorithm is much shorter than Wavelet Decomposition basedBaseline wander correction of Trans-
the length and the step of the input series. Therefore, the Abdominal Maternal ECG,” 6th International Conference on Electrical
suggested filter is able to provide real-time operation. and Computer Engineering, pp. 630 – 633, December 2010.
[9] Vinod K. Pandey, “Adaptive Filtering for Baseline Wander Removal in
ECG,” 10th IEEE International Conference on Information Technology
IV. CONCLUSION and Applications in Biomedicine, pp. 1 - 4, November 2010.
A reconfigurable overlapping FFT/IFFT filter for ECG [10] FREDRIC J. HARRIS, “On the use of windows for harmonic analysis
data obtained from a Holter monitoring system is presented. with the discrete Fourier transform,” Proceedings of the IEEE, vol. 66,
pp. 51-83, January 1978.
The Holter monitoring system used in this work incorporates