Tweet
Dear Anthony,
You are doing a lot of things and it is important to recognize what each does.
First you are band passing the data. The low-pass part gets rid of
movement artifact. The low frequency cut-off gets rid of noise and if
you are going to sample the data, should be set at no more than half
of the sampling frequency to prevent aliasing. It is common to see
published papers with wide band-width EMG amplifiers, sampled at
500-1000/sec because we KNOW that EMG (surface anyway) has no energy
above 500 hz. Of course we do not know that there is no noise up
there or any other artifact and once the data is sampled, we will
never know so the high frequency cut off is important.
If you want to digitize the raw EMG, this is the time to do it. Other
wise, now you can rectify it as Chris explains (if he remembers
crystal sets, he is older than he looks). This nonlinear process
brings the modulation envelope down to low frequencies where it can
be extracted by low-pass filtering. THe process also increases the
high frequency content which is unwanted and removed by the same
filtering. You should not digitize the rectified EMG without further
filtering because you will bring back your aliasing problems.
When you do your moving average filter, you are doing two things at
once. If you replace point i with the average of points i to i+n, you
are performing a moving-average, low-pass filter on n+1 points. If
you use the process that you describe, you are also decimating the
data (reducing the number of samples) and the way you describe it is
not optimal. You can do both at once but not that way and you should
look in a good text book to see how.
There is no consensus as far as I am aware of what the best method of
low pass filtering your rectified signal should be. It depends on
what you want it for. Some want to estimate muscle force but many
want to look at muscle activation patterns and figure out what the
CNS is thinking about. Most people use Butterworth filters but I
have never been able to understand why. Butterworth (and others) can
be built with resistors and capacitors and so have a long history of
use in the analog world but that is no reason to impose them on the
digital world which has different constraints and possibilities. I
personally like moving average filters.
As for using the demodulated EMG to predict muscle force, this has
been going on forever with mixed success. Everything in the world is
a second order system, if you are not too fussy. And isn't if you
are. You can do your force estimation from the rectified signal, or
from the low-pass rectified signal, depending on how you want to do
it. I would advise against calling the phase shifting properties of a
low-pass filter, "electro-mechanical" delay but that is another story.
--
__________________________________________________ _________________
| Gerald Gottlieb (617) 358-0719
| NeuroMuscular Research Center 353-9757
| Boston University fax 353-5737
| 19 Deerfield St.
| Boston MA 02215
---------------------------------------------------------------
To unsubscribe send SIGNOFF BIOMCH-L to LISTSERV@nic.surfnet.nl
For information and archives: http://isb.ri.ccf.org/biomch-l
---------------------------------------------------------------
You are doing a lot of things and it is important to recognize what each does.
First you are band passing the data. The low-pass part gets rid of
movement artifact. The low frequency cut-off gets rid of noise and if
you are going to sample the data, should be set at no more than half
of the sampling frequency to prevent aliasing. It is common to see
published papers with wide band-width EMG amplifiers, sampled at
500-1000/sec because we KNOW that EMG (surface anyway) has no energy
above 500 hz. Of course we do not know that there is no noise up
there or any other artifact and once the data is sampled, we will
never know so the high frequency cut off is important.
If you want to digitize the raw EMG, this is the time to do it. Other
wise, now you can rectify it as Chris explains (if he remembers
crystal sets, he is older than he looks). This nonlinear process
brings the modulation envelope down to low frequencies where it can
be extracted by low-pass filtering. THe process also increases the
high frequency content which is unwanted and removed by the same
filtering. You should not digitize the rectified EMG without further
filtering because you will bring back your aliasing problems.
When you do your moving average filter, you are doing two things at
once. If you replace point i with the average of points i to i+n, you
are performing a moving-average, low-pass filter on n+1 points. If
you use the process that you describe, you are also decimating the
data (reducing the number of samples) and the way you describe it is
not optimal. You can do both at once but not that way and you should
look in a good text book to see how.
There is no consensus as far as I am aware of what the best method of
low pass filtering your rectified signal should be. It depends on
what you want it for. Some want to estimate muscle force but many
want to look at muscle activation patterns and figure out what the
CNS is thinking about. Most people use Butterworth filters but I
have never been able to understand why. Butterworth (and others) can
be built with resistors and capacitors and so have a long history of
use in the analog world but that is no reason to impose them on the
digital world which has different constraints and possibilities. I
personally like moving average filters.
As for using the demodulated EMG to predict muscle force, this has
been going on forever with mixed success. Everything in the world is
a second order system, if you are not too fussy. And isn't if you
are. You can do your force estimation from the rectified signal, or
from the low-pass rectified signal, depending on how you want to do
it. I would advise against calling the phase shifting properties of a
low-pass filter, "electro-mechanical" delay but that is another story.
--
__________________________________________________ _________________
| Gerald Gottlieb (617) 358-0719
| NeuroMuscular Research Center 353-9757
| Boston University fax 353-5737
| 19 Deerfield St.
| Boston MA 02215
---------------------------------------------------------------
To unsubscribe send SIGNOFF BIOMCH-L to LISTSERV@nic.surfnet.nl
For information and archives: http://isb.ri.ccf.org/biomch-l
---------------------------------------------------------------