Surface EMG Signal Amplification and Filtering: Jingpeng Wang Liqiong Tang John E Bronlund
Surface EMG Signal Amplification and Filtering: Jingpeng Wang Liqiong Tang John E Bronlund
Surface EMG Signal Amplification and Filtering: Jingpeng Wang Liqiong Tang John E Bronlund
15
International Journal of Computer Applications (0975 – 8887)
Volume 82 – No1, November 2013
3. AMPLIFICATION AND FILTERING As showed in Figure 2, the EMG signal from the electrodes is
CIRCUITRY fed into the positive and negative input pins of INA128. R 1 and
The quality of an EMG signal from the electrodes is partially R2 comprise the resister RG. A signal from the mid-point of R1
dependent on the properties of the amplifiers. Because of the and R2 is fed into OPA2604 to provide a guarding potential and
weak amplitude of EMG signals typically in the order of tens to a reference potential.
thousands µV, it is necessary that the gain of the amplifiers
used in EMG applications is in the range from 1000 to 10000.
Consequently, the amplification process commonly
incorporates several stages. Figure 1 illustrates the block
diagram of the amplification and filtering stages implemented
in the circuit presented in this paper. The most important stage,
namely the first stage close to the electrodes, is conventionally
called pre-amplifier. The consideration to incorporate a pre-
amplifier is to have a high common mode rejection ratio
(CMRR), a high input impedance, a short distance to the signal
source, and a strong direct current (DC) signal suppression [3].
The outcome from the pre-amplifier is then processed by a
high-pass and a low-pass filter before entering into the second Fig.2: The schematic circuit of the pre-amplifier
amplification stage that amplifies signals again to attain an with reference and shielding-driven
expected gain. The second amplification stage is a simple
inverting amplifier and its gain can easily be adjusted by Because of the weakness of the EMG signal, the difference in
choosing different resistors. To further suppress high frequency the amplitude between some useful EMG signal components
noises, a low-pass filter follows the second amplification stage. and the noises can be very small. If the gain of the pre-amplifier
Finally, the output signal from the amplification and filtering is set too large, the noises will be simultaneously amplified
circuit is fed into an analog-digital converter (ADC). enormously, thus leading to the instability and saturation of the
subsequent amplifier. To avoid such a consequence, the gain of
the pre-amplifier is preferred to be set around 10. For the pre-
amplifier designed for this research, the gain was calculated
using equation (1) and set at 11.4.
Fig. 1: Block diagram of amplification
and filtering circuitry
50k 50k
G1 1 1 11.4 (2)
3.1 The Electrodes RG R1 R2
The design of the electrode is of paramount importance as the where G1 is the gain of the pre-amplifier; both R1 and R2 are
quality and the design of the electrodes directly affect the 2.4kΩ resisters.
quality of the EMG signals. There are mainly two types of
electrodes on the market, non-invasive surface electrode and 4. HIGH-PASS AND LOW-PASS FILTER
invasive inserted electrode (wire or needle). Both have been In the amplification and filtering circuitry, high-pass filters and
used for EMG signal detection. The surface electrode has two low pass-filters were used after the first and second
categories, passive and active. The passive electrode has a amplification stages. The reason is that the noises and the EMG
conductive (commonly metal) detection surface touching with signals are simultaneously amplified, which is not favourable
the skin, whereas the active electrode has a differential for the following process. To design a filter, the corner
amplifier with a passive electrode in order to reduce the effects frequency, the roll-off rate, and the circuit topology have to be
of capacitance coupling and provide very low output chosen. The order of a filter determines the roll-off rate of the
impedance, rather than transmitting the signal to the amplifier filter, namely the slope of its frequency response curve. A first-
through long lead wires. The advantage of active electrode is order filter has a roll-off rate of -6dB/Octave whereas a second-
the reduced capacitance coupling and the very low output order filter has -12dB/Octave, that is, the roll-off rate of the
impedance will not introduce significant noises from power line filter response is proportional to the order of a filter. Higher-
and cable movement artefact, but such electrodes are expensive order filters are usually constructed by cascading first- and
and complicated [1, 3]. Surface electrodes were selected for this second-order blocks [5].
research as they are non-invasive and economic.
There are several types of noises with different characteristics
within the low frequency components. As a result, the design a
16
International Journal of Computer Applications (0975 – 8887)
Volume 82 – No1, November 2013
high-pass filter is more complicated comparing with low-pass research, every amplification stage was followed by a second-
filter design. In recent years, more research has been seen in order Sallen-Key low-pass filter as shown in Figure 5. The
high-pass filter design and different corner frequencies corner frequency and the passband gain can also be calculated
including 5 Hz, 10 Hz, 10-20 Hz, 20 Hz, and 15-28 Hz, and using equation (3) and (4) the same way as for high-pass filter
different roll-off rates including -12dB/Octave, -18dB/Octave, design. In this research, the two second-order low-pass filters
and -24dB/Octave were investigated, employed and presented have the same properties and are placed before and after the
in literature [6-8]. second amplification stage as shown in Figure 1. The total
frequency response of these two second-order low-pass filters
In this research, a second-order Sallen-Key high-pass filters should be equivalent to a low-pass filter cascaded by two
(Figure 3) and a fourth-order Sallen-Key high-pass filter second-order low-pass filters, namely a fourth-order low-pass
(Figure 4) were studied and compared. The former has a slope filter. For this research, the corner frequency of the fourth-order
of -12dB/Octave and a 20 Hz corner frequency whereas the low-pass filter is 500 Hz and the roll-off rate is -24dB/Octave
latter has a slope of -24dB/Octave and the same corner [9].
frequency. Sallen-Key architecture is a widely used circuit
topology for building second-order filters, also known as a
voltage-controlled voltage source. The corner frequency and the
passband gain of the second-order Sellen-Key high-pass filter
are given by equation (3) and (4) respectively [9].
1
fc (3)
2 R1 R2C1C2
R4
G pass 1 (4)
R3
where fc is the corner frequency and G pass is the passband
gain.
17
International Journal of Computer Applications (0975 – 8887)
Volume 82 – No1, November 2013
than the second-order low-pass filter (Line D), by achieving the applications, the selection of sampling rate higher than 2000 Hz
same attenuation to spectral replications at 500 Hz or even may need to consider the hardware constrains.
higher attenuation within 0-500 Hz, without causing aliasing. In
other words, if the same sampling rate fs2 is used, the fourth- Unlike conventional Nyquist rate A/D converters, Sigma-Delta
order low-pass filter (Line E) can introduce lower amplitude of (Σ−∆) A/D converters use oversampling and decimation
spectral replications on the spectral baseband (0-500 Hz) than filtering technology. At the analog input end, a sampling rate
the second-order low-pass filter (Line D). Figure 6 also reveals that is much greater than the Nyquist rate is used. Through
that, for the signals with the maximum frequency less than 500 decimation and digital filtering, the sampling rate is reduced
Hz and using a Nyquist rate A/D converter, a sampling rate of down to the Nyquist rate at digital output end. Sigma-Delta
1000 Hz may not provide the necessary aliasing rejection for (Σ−∆) conversion technology minimises the size of the data but
the fourth-order low-pass filter (Line F). In this research, a reduces the requirements on the analog anti-aliasing filter
sampling rate of 2000 Hz was chosen. The experiment showed simultaneously. It means that using a Sigma-Delta (Σ−∆) A/D
the 2000 Hz sampling rate was higher enough to attenuate the converter, the EMG signals can be recorded at an output
frequencies above 500 Hz. However, as the sampling rate gets sampling rate of 1000 samples per second (SPS).
higher the size of the data is increased. Therefore, in real
0.9
F
0.8
C
Amplitude Response
0.7
0.6
0.5
0.4
D
0.3
E
0.2 A
0.1 B
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Frequency (Hz)
500 Hz f f
s1 s2
Fig. 6 The amplitude responses and spectral replications of Sellen-Key low-pass filters
Line A – the amplitude of the second-order low-pass filter
Line B – the amplitude of the fourth –order low-pass filter
Line C – the replication of the fourth-order low-pass filter at sampling rate of fs1
Line D – the replication of the second-order low-pass filter at sampling rate of fs2
Line E – the replication of the fourth-order low-pass filter at sampling rate of fs2
Line F – the replication of the fourth-order low-pass filter at sampling rate of 1 kHz
5. EXPERIMENT
To compare the effect of a fourth-order and a second-order In the experiment, a data acquisition card PCI-6229 from
high-pass filter on the low frequency components, the National Instrument and LabView were used to collect EMG
experiment circuit was configured as showed in Figure 7. After data. The amplified and filtered EMG signals were fed into a
the electrodes and the pre-amplifier, the signals is conditioned 68-pin shielded connector SCB-68 which is connected to the
using two different approaches before connecting to the SCB- 16-Bit PCI-6229 DAQ card in a computer. A programme was
68 connector from National Instrument as illustrated in the developed in LabView to log the EMG data. Matlab was
block diagram in Figure 7, a hardware interface box for employed to analyse the spectra of the EMG signals. The EMG
National Instrument PCI card. The difference between the two signals were recorded at a sampling rate of 2000 Hz.
approaches is in the high-pass filter section. The first approach
uses a second-order high-pass filter as showed in Figure 3 Three snap type pre-gelled surface EMG (sEMG) electrodes
whereas the other is a fourth-order high-pass filter shown in and a sEMG snap cable from Thought Technology Ltd were
Figure 4. The amplified and filtered EMG signals from the two used. Two electrodes with the inter-spacing of 2 cm were
approaches were simultaneously sampled and recorded using placed on the forearm flexor carpi radialis muscle and the
LabView. Hence, the EMG signals from the two approaches are reference electrode was located on the wrist. Figure 8 illustrates
actually from the same original source and in the same period of the experiment configuration used in this research.
time but through different high-pass filters.
18
International Journal of Computer Applications (0975 – 8887)
Volume 82 – No1, November 2013
The corner frequency of the high-pass and low-pass filter and transform was performed to the activation segments to analyse
the total gain of the amplification and filtering circuit shown in the spectrum.
Figure 7 were set as follows:
6.1 Ball Squeezing
The high-pass corner frequency is 20 Hz; The activation segments from the five data sets for ball
The low-pass corner frequency is 500 Hz; squeezing were analysed using Fourier transform. As shown in
The total gain is set at about 2900. Figure 7, each test had two sets of data. One is the result of
using a fourth-order high-pass filter while the other is the output
The human upper limb is able to perform sophisticated from the second-order high-pass filter. Figure 9 and 10 is the
movements as it has multiple degrees of freedom. However, up comparison of the amplitude response of an activation segment
until now, it is still a challenge for researchers to model and randomly selected from the experiment activation segments
control human hand movement. In most cases, only certain between the two different approaches discussed in Section 4.
hand gestures are studied. In this research, two upper limb
movements were used to analyse and test the proposed (a)
1
amplification and filtering design. One is ball squeezing, which
0.8
Amplitude Response
0.2
FLEXION
0 50 100 150 200 250 300 350 400 450 500
Frequency (Hz)
(b)
1
0.8
Amplitude Response
EXTENTION
0.6
0.2
Fig. 9 (a) Six types of hand grasping and
(b) Two forearm motions. 0 50 100 150 200 250 300 350 400 450 500
Frequency (Hz)
6. DATA ANALYSIS
Matlab was used to analyse experimental data and evaluate the
proposed EMG amplification and filtering circuit. The EMG
signal waveforms were displayed in Matlab. The activation
segments of the two upper-limb movements, that is, when the
target muscles were contracting, were clearly identified. Fourier
19
International Journal of Computer Applications (0975 – 8887)
Volume 82 – No1, November 2013
(a)
1
Amplitude Response
0.8
Fig. 12 The amplitude response of the EMG signal of
0.6
forearm flexion (zoomed in 0-50 Hz).
0.4
0.2
(a) Using the fourth-order high-pass filter
(b) Using the second-order high-pass filter
0 5 10 15 20 25 30 35 40 45 50
Frequency (Hz)
(b)
1 As the results shown in Figure 9, 10, 11 and 12, there is only
Amplitude Response
0.8
0.6
has some obvious noise signals. The frequency spectrum of the
0.4
noise signals is displayed in Figure 15, which clearly reveals
0.2
that the noise from the power line interferes with the 50 Hz
component. This noise and its harmonics is the primary noise
0 50 100 150 200 250 300 350 400 450 500 source. Conventionally, a notch filter at 50 Hz (or 60 Hz) is
Frequency (Hz)
employed to clean the power line noises. However, the majority
(b) energy of the EMG signal is within the 30-150 Hz range and
1
there is no notch filter can perform to the ideal level to only
Amplitude Response
0.8
0.8
0.6
EMG signal amplification and filtering circuits. The circuit
0.4
design proposed was tested stage by stage and then integrated
0.2
with a PC-based EMG data acquisition system through three
electrodes. The system is able to successfully acquire the EMG
0 5 10 15 20 25 30 35 40 45 50 signals and suppress the noises. The analysis of the experiment
Frequency (Hz)
results validates the effectiveness of the proposed design. The
(b) comparison between the second-and fourth-order high-pass
1
filter also confirms that a second-order high-pass filter has the
Amplitude Response
0.8
potential to replace the fourth-order high-pass filter.
0.6
Furthermore, the experiment also clearly shows that it is
0.4
necessary to remove the noises from power line to increase
0.2
signal–noise-ratio.
0 5 10 15 20 25 30 35 40 45 50
Frequency (Hz)
20
International Journal of Computer Applications (0975 – 8887)
Volume 82 – No1, November 2013
(a)
1
0.8
0.6
Amplitude (V)
0.4
0.2
-0.2
-0.4
-0.6
-0.8
-1
1 2 3 4 5 6 7 8 9 10
Time (S)
(b)
2
1.5
1
Amplitude (V)
0.5
-0.5
-1
-1.5
-2
1 2 3 4 5 6 7 8 9 10
Time (S)
10. REFERENCES
9. ACKNOWLEDGMENTS [1] De Luca, C. J. 2006. Electromyography. Encyclopedia of
This research is supported by the School of Engineering and Medical Devices and Instrumentation, pp. 98-109.
Advanced Technology (SEAT), Massey University, New
Zealand. A special thanks to the electronics technical team for [2] De Luca, C. J. 2002. Surface electromyography: Detection
the great assistance. and recording. DelSys Incorporated.
21
International Journal of Computer Applications (0975 – 8887)
Volume 82 – No1, November 2013
[3] Day,S. 2002. Important factors in surface EMG Control Applied to Upper Limb. Biomedical Engineering,
Measurement. Calgary: Bortech Biomedical Ltd. IEEE Transactions on, vol. 55, no. 8, pp. 1956-1965.
[4] INA128 Datasheet . 2005. Texas Instruments [14] Phinyomark, A., Nuidod, A., Phukpattaranont, P. and
Incorporated. Limsakul, C. 2012. Feature extraction and reduction of
wavelet transform coefficients for EMG pattern
[5] Zumbahlen, H. 2007. Phase Relations in Active Filters. classification. Elektronika ir Elektrotechnika, vol. 122, no.
Analog Dialogue, 41. 6, pp. 27-32.
[6] Cosmanescu, A., Miller, B., Magno, T., Ahmed, A. and [15] Jun-Uk, C., Inhyuk, M. and Mu-Seong, M. 2006. A Real-
Kremenic, I. Design and Implementation of a Wireless Time EMG Pattern Recognition System Based on Linear-
(Bluetooth®) Four Channel Bio-Instrumentation Nonlinear Feature Projection for a Multifunction
Amplifier and Digital Data Acquisition Device with User- Myoelectric Hand. Biomedical Engineering, IEEE
Selectable Gain, Frequency, and Driven Reference. pp. Transactions on, vol. 53, no. 11, pp. 2232-2239.
2053-2056.
[16] Zhang, X. and Zhou, P. 2012. Sample entropy analysis of
[7] De Luca, C. J., Donald Gilmore, L., Kuznetsov, M. and surface EMG for improved muscle activity onset detection
Roy, S. H. 2010. Filtering the surface EMG signal: against spurious background spikes. Journal of
Movement artifact and baseline noise contamination. Electromyography and Kinesiology, vol. 22, no. 6, pp.
Journal of Biomechanics, vol. 43, no. 8, pp. 1573-1579. 901-907.
[8] Smith, L. H., Hargrove, L. J., Lock, B. A. and Kuiken, T. [17] Khushaba, R. N., Kodagoda, S., Takruri, M. and
A. 2011. Determining the Optimal Window Length for Dissanayake, G. 2012. Toward improved control of
Pattern Recognition-Based Myoelectric Control: Balancing prosthetic fingers using surface electromyogram (EMG)
the Competing Effects of Classification Error and signals. Expert Systems with Applications, vol. 39, no. 12,
Controller Delay. Neural Systems and Rehabilitation pp. 10731-10738.
Engineering, IEEE Transactions on, vol. 19, no. 2, pp.
186-192. [18] Hosking, R. H. 2012. Critical Techniques for High-Speed
A/D Converters in Real-Time Systems.
[9] Analysis of the Sallen-Key Architecture. 1999. Texas
Instruments Incorporated. [19] Fukaya, N., Toyama, S., Asfour, T. and Dillmann, R.
Design of the TUAT/Karlsruhe humanoid hand. pp. 1754-
[10] Guanglin, L., Yaonan, L., Zhiyong, Z., Yanjuan, G. and 1759.
Rui, Z. Selection of sampling rate for EMG pattern
recognition based prosthesis control. pp. 5058-5061. [20] Yacoub, S., Raoof, K. and Eleuch, H. Filtering of Cardiac
and Power Line in Surface Respiratory EMG Signal.
[11] Cipriani, C., Antfolk, C., Controzzi, M., Lundborg, G.,
Rosen, B., Carrozza, M. C. and Sebelius, F. 2011. Online [21] Golabbakhsh, M., Masoumzadeh, M. and Sabahi, M. F.
Myoelectric Control of a Dexterous Hand Prosthesis by 2011. ECG and power line noise removal from respiratory
Transradial Amputees. Neural Systems and Rehabilitation EMG signal using adaptive filters. Majlesi Journal of
Engineering, IEEE Transactions on, vol. 19, no. 3, pp. Electrical Engineering, vol. 5, no. 4.
260-270.
[22] A. B. Sankar, D. Kumar, and K. Seethalakshmi,
[12] Hargrove, L. J., Scheme, E. J., Englehart, K. B. and “Performance study of various adaptive filter algorithms
Hudgins, B. S. 2010. Multiple Binary Classifications via for noise cancellation in respiratory signals,” Signal
Linear Discriminant Analysis for Improved Controllability Processing: An International Journal (SPIJ), vol. 4, no. 5,
of a Powered Prosthesis. Neural Systems and pp. 267, 2010.
Rehabilitation Engineering, IEEE Transactions on, vol. 18,
no. 1, pp. 49-57. [23] De Luca, C. J., Kuznetsov, M. L., Gilmore, D. and Roy, S.
H. 2012. Inter-electrode spacing of surface EMG sensors:
[13] Oskoei, M. A. and Huosheng, H. 2008. Support Vector Reduction of crosstalk contamination during voluntary
Machine-Based Classification Scheme for Myoelectric contractions,” Journal of Biomechanics, vol. 45, no. 3, pp.
555-561.
IJCATM : www.ijcaonline.org 22