Tweet
Hello All,
I would like to throw this question out concerning acceptable
methods of selecting cutoff frequencies when using a digital filter.
I am processing GRF data collected by a treadmill during walking and
running. Some of the trials were completed onboard an airplane, which
introduces additional noise to the recorded signal. I am smoothing the data
using a 4th order recursive digital filter.
As many of you know, the topic of selecting an appropriate cutoff
frequency is of interest, since the cutoff frequency must be one that
removes artifact but maintains the integrity of the measured signal. In the
literature, there have been a variety of approaches to this issue, including
selecting a single cutoff frequency at which all data sets are filtered, or
using some sort of methodology to select an individual cutoff frequency for
each data set.
I am using the algorithm described by Challis (1999) to determine
the optimal cutoff frequency for each individual trial. This procedure uses
an autocorrelation procedure to identify a suitable cutoff frequency for
each dataset. Of course, this means that many cutoff frequencies are used,
with ideally the most suitable cutoffs are chosen for each dataset.
The problem that I am having is that the procedure cannot
differentiate between noise and actual movement. In this specific case with
GRF data, one of the subjects is a heel-striker whose characteristic GRF
trajectory has an impact peak that is observable, but is subtle. The Challis
algorithm smoothes through the impact peak in these trials, making the
assumption that the impact peak is noise. This is obviously not the case,
and for these trials the Challis approach does not work. In these cases, a
cutoff frequency that is much greater than the Challis predicted cutoff is
more appropriate.
I am now concerned that I am treating data differently because for
some trials the Challis procedure works beautifully, and for others a
subjective decision based on my experience is a better solution. Does anyone
have any thoughts on this matter? Specifically, I would like to publish
these data and am concerned that reviewers may question why cutoff
frequencies were selected using different approaches in the same study. I
personally think that the approach that I am using (Challis for some,
subjective for others) is acceptable because essentially all I am doing is
selecting appropriate individual cutoff frequencies for each data set, just
with some I am using some help. I am thinking of the Challis algorithm as a
tool, and in some cases the tool is unusable. However there are other
approaches such as:
1) Use Challis (1999) on all data and accept the answers it
generates
2) Use a standard cutoff frequency for all data
3) Do not filter the data
Once again, these GRF data were collected onboard an airplane. I
would appreciate any feedback as my goal is to publish these data using the
processing technique that is most appropriate.
John DeWitt, M.S., C.S.C.S.
Biomechanist - Exercise Physiology Laboratory
Space Physiology & Countermeasures
NASA - Johnson Space Center
Houston, TX 77058
281-483-8939 / 281-483-4181 (fax)
John DeWitt, M.S., C.S.C.S.
Biomechanist - Exercise Physiology Laboratory
Space Physiology & Countermeasures
NASA - Johnson Space Center
Houston, TX 77058
281-483-8939 / 281-483-4181 (fax)
-----------------------------------------------------------------
To unsubscribe send SIGNOFF BIOMCH-L to LISTSERV@nic.surfnet.nl
For information and archives: http://isb.ri.ccf.org/biomch-l
Please consider posting your message to the Biomch-L Web-based
Discussion Forum: http://movement-analysis.com/biomch_l
-----------------------------------------------------------------
I would like to throw this question out concerning acceptable
methods of selecting cutoff frequencies when using a digital filter.
I am processing GRF data collected by a treadmill during walking and
running. Some of the trials were completed onboard an airplane, which
introduces additional noise to the recorded signal. I am smoothing the data
using a 4th order recursive digital filter.
As many of you know, the topic of selecting an appropriate cutoff
frequency is of interest, since the cutoff frequency must be one that
removes artifact but maintains the integrity of the measured signal. In the
literature, there have been a variety of approaches to this issue, including
selecting a single cutoff frequency at which all data sets are filtered, or
using some sort of methodology to select an individual cutoff frequency for
each data set.
I am using the algorithm described by Challis (1999) to determine
the optimal cutoff frequency for each individual trial. This procedure uses
an autocorrelation procedure to identify a suitable cutoff frequency for
each dataset. Of course, this means that many cutoff frequencies are used,
with ideally the most suitable cutoffs are chosen for each dataset.
The problem that I am having is that the procedure cannot
differentiate between noise and actual movement. In this specific case with
GRF data, one of the subjects is a heel-striker whose characteristic GRF
trajectory has an impact peak that is observable, but is subtle. The Challis
algorithm smoothes through the impact peak in these trials, making the
assumption that the impact peak is noise. This is obviously not the case,
and for these trials the Challis approach does not work. In these cases, a
cutoff frequency that is much greater than the Challis predicted cutoff is
more appropriate.
I am now concerned that I am treating data differently because for
some trials the Challis procedure works beautifully, and for others a
subjective decision based on my experience is a better solution. Does anyone
have any thoughts on this matter? Specifically, I would like to publish
these data and am concerned that reviewers may question why cutoff
frequencies were selected using different approaches in the same study. I
personally think that the approach that I am using (Challis for some,
subjective for others) is acceptable because essentially all I am doing is
selecting appropriate individual cutoff frequencies for each data set, just
with some I am using some help. I am thinking of the Challis algorithm as a
tool, and in some cases the tool is unusable. However there are other
approaches such as:
1) Use Challis (1999) on all data and accept the answers it
generates
2) Use a standard cutoff frequency for all data
3) Do not filter the data
Once again, these GRF data were collected onboard an airplane. I
would appreciate any feedback as my goal is to publish these data using the
processing technique that is most appropriate.
John DeWitt, M.S., C.S.C.S.
Biomechanist - Exercise Physiology Laboratory
Space Physiology & Countermeasures
NASA - Johnson Space Center
Houston, TX 77058
281-483-8939 / 281-483-4181 (fax)
John DeWitt, M.S., C.S.C.S.
Biomechanist - Exercise Physiology Laboratory
Space Physiology & Countermeasures
NASA - Johnson Space Center
Houston, TX 77058
281-483-8939 / 281-483-4181 (fax)
-----------------------------------------------------------------
To unsubscribe send SIGNOFF BIOMCH-L to LISTSERV@nic.surfnet.nl
For information and archives: http://isb.ri.ccf.org/biomch-l
Please consider posting your message to the Biomch-L Web-based
Discussion Forum: http://movement-analysis.com/biomch_l
-----------------------------------------------------------------