Tweet
I agree strongly with Gianluca De Luca's and Rick Lambert's position on
anti-aliasing filters for EMG. There are many possibililities for
high-frequency contamination of EMG data. For example, our lab is not far from
a 50kW broadcast station, and we sometimes see 90MHz noise in raw data. Many
(if not most) computers introduce high-frequency noise into acquired signals,
especially if the acquisition card is located within the computer chassis. When
data contaminated with noise above half the sampling frequency has been
acquired, the noise is aliased down to lower frequencies and cannot be removed.
This noise often ends up looking much like low-level EMG in the acquired
signals, os you may have it and never know about it. There are only three ways
that I know of to avoid aliasing errors: collect perfect, noise-free data, use
unreasonably high sample rates (in the case of our radio signal, 180MHz), or use
front-end analog low-pass filters. The first two options are not realistic
under most circumstances.
As for high-pass filters for motion artifact, there is more room for
discussion. There is certainly no advantage to collecting EMG data down to DC
(0 Hz); the question is, what frequency range is appropriate? Large,
low-frequency artifacts often occur in surface and fine wire EMG. If amplifier
gains are set low enough to avoid saturation and the associated instrumentation
problems Gianluca described, then signal-to-noise ratio is compromised. But the
intended use of the EMG must also be considered. If the data are to be used for
identifying muscle timing and relative activity levels, loss of the small amount
of very low frequency EMG signal which may be occur under some circumstances
will not affect data analysis. For most EMG users, I believe that analog
high-pass filters (cutoff 30Hz or so) improves the reliability and quality of
the EMG recording, without loss of any useful information. This is particularly
true for movements involving impact (e.g. jumping), where motion artifacts are
greatest. However, if one is particularly interested in the frequency content
of the signal (e.g. fatigue studies), and/or looking at relatively slow
movements, then a valid argument could be made for avoiding analog high-pass
pre-filtering.
If anyone really wants to know about the frequency characteristics of a
particular muscle, they can acquire data during an isometric contraction (to
avoid motion artifact) using a high sampling rate and no high-pass filter, and
then use the FFT to look at the power spectrum of the signal. This analysis is
now relatively routine and can be done in most software being used for EMG, and
is a good exercise for anyone who works with EMG to increase understanding of
the nature of the signal. Then you can make your own decision about how much is
lost for a particular filter cutoff frequency.
The argument about storing "raw" vs. "processed" EMG has been around for quite a
while, but has usually been in reference to envelope-processing schemes which
greatly reduce information content and bandwidth. I am a strong advocate for
storing "raw" EMG, and performing any additional manipulation in the digital
domain. However, I think the use of analog pre-filtering is well accepted in
our field and is usually beneficial.
One final comment: filtering should never be used as a substitute for good lab
technique. No amount of filtering will overcome inadequate skin preparation,
poor attachment of electrodes and leads, use of long, unshielded cables, etc.
If your data is heavily contaminated with noise, the major sources of the noise
should be determined and eliminated if possible. Then, intelligent decisions
about filtering can be made.
__________________________________________________ ___________________
Scott Tashman, Ph.D.
Head, Motion Analysis Section Assistant Professor
Bone and Joint Center Department of Orthopaedics
Henry Ford Hospital School of Medicine
2799 W. Grand Blvd. Case Western Reserve University
Detroit, MI 48202
Voice: (313) 916-8680 or 916-7572
FAX: (313) 916-8812 or 916-8064
Internet: tashman@bjc.hfh.edu
__________________________________________________ ___________________
begin:vcard
n:Tashman;Scott
tel;fax:313-916-8812
tel;work:313-916-8680
x-mozilla-html:TRUE
org:Henry Ford Health Sciences Center;Bone and Joint Center
version:2.1
email;internet:tashman@bjc.hfh.edu
title:Head, Motion Analysis Section
adr;quoted-printable:;;Henry Ford Hospital=0D=0A2799 West Grand Blvd., ER2015;Detroit;MI;48202;USA
x-mozilla-cpt:;3
fn:Scott Tashman, Ph.D.
end:vcard
anti-aliasing filters for EMG. There are many possibililities for
high-frequency contamination of EMG data. For example, our lab is not far from
a 50kW broadcast station, and we sometimes see 90MHz noise in raw data. Many
(if not most) computers introduce high-frequency noise into acquired signals,
especially if the acquisition card is located within the computer chassis. When
data contaminated with noise above half the sampling frequency has been
acquired, the noise is aliased down to lower frequencies and cannot be removed.
This noise often ends up looking much like low-level EMG in the acquired
signals, os you may have it and never know about it. There are only three ways
that I know of to avoid aliasing errors: collect perfect, noise-free data, use
unreasonably high sample rates (in the case of our radio signal, 180MHz), or use
front-end analog low-pass filters. The first two options are not realistic
under most circumstances.
As for high-pass filters for motion artifact, there is more room for
discussion. There is certainly no advantage to collecting EMG data down to DC
(0 Hz); the question is, what frequency range is appropriate? Large,
low-frequency artifacts often occur in surface and fine wire EMG. If amplifier
gains are set low enough to avoid saturation and the associated instrumentation
problems Gianluca described, then signal-to-noise ratio is compromised. But the
intended use of the EMG must also be considered. If the data are to be used for
identifying muscle timing and relative activity levels, loss of the small amount
of very low frequency EMG signal which may be occur under some circumstances
will not affect data analysis. For most EMG users, I believe that analog
high-pass filters (cutoff 30Hz or so) improves the reliability and quality of
the EMG recording, without loss of any useful information. This is particularly
true for movements involving impact (e.g. jumping), where motion artifacts are
greatest. However, if one is particularly interested in the frequency content
of the signal (e.g. fatigue studies), and/or looking at relatively slow
movements, then a valid argument could be made for avoiding analog high-pass
pre-filtering.
If anyone really wants to know about the frequency characteristics of a
particular muscle, they can acquire data during an isometric contraction (to
avoid motion artifact) using a high sampling rate and no high-pass filter, and
then use the FFT to look at the power spectrum of the signal. This analysis is
now relatively routine and can be done in most software being used for EMG, and
is a good exercise for anyone who works with EMG to increase understanding of
the nature of the signal. Then you can make your own decision about how much is
lost for a particular filter cutoff frequency.
The argument about storing "raw" vs. "processed" EMG has been around for quite a
while, but has usually been in reference to envelope-processing schemes which
greatly reduce information content and bandwidth. I am a strong advocate for
storing "raw" EMG, and performing any additional manipulation in the digital
domain. However, I think the use of analog pre-filtering is well accepted in
our field and is usually beneficial.
One final comment: filtering should never be used as a substitute for good lab
technique. No amount of filtering will overcome inadequate skin preparation,
poor attachment of electrodes and leads, use of long, unshielded cables, etc.
If your data is heavily contaminated with noise, the major sources of the noise
should be determined and eliminated if possible. Then, intelligent decisions
about filtering can be made.
__________________________________________________ ___________________
Scott Tashman, Ph.D.
Head, Motion Analysis Section Assistant Professor
Bone and Joint Center Department of Orthopaedics
Henry Ford Hospital School of Medicine
2799 W. Grand Blvd. Case Western Reserve University
Detroit, MI 48202
Voice: (313) 916-8680 or 916-7572
FAX: (313) 916-8812 or 916-8064
Internet: tashman@bjc.hfh.edu
__________________________________________________ ___________________
begin:vcard
n:Tashman;Scott
tel;fax:313-916-8812
tel;work:313-916-8680
x-mozilla-html:TRUE
org:Henry Ford Health Sciences Center;Bone and Joint Center
version:2.1
email;internet:tashman@bjc.hfh.edu
title:Head, Motion Analysis Section
adr;quoted-printable:;;Henry Ford Hospital=0D=0A2799 West Grand Blvd., ER2015;Detroit;MI;48202;USA
x-mozilla-cpt:;3
fn:Scott Tashman, Ph.D.
end:vcard