glgottlieb58

12-18-1995, 12:27 AM

>Dear Biomch-L,

>

>Are there any studies which compare displacement/deformation data derived

>directly (eg. displacement transducer) with data derived indirectly via

>double integration of acceleration data (eg. with accelerometer) ?

>

I no of no formal studies but we have look at this over the years in terms

of our single joint studies. The short answer is that you cant get there

from here. You cant get a good estimate of position from acceleration and

you cant even get all that good an estimate of velocity. This is based on

the assumption that you use the usual 12 bit A/D converters and instruments

good enough but not a lot better.

The reason, as best as I can figure it out, is that you only have +/- 1/2

bit accuracy in your definition of zero and this is a bias, not a random

error. Integrating it out gives you significant errors over several

hundred integration steps which is usually a minimum number that one uses.

Accelerometer signals look great and are much better than second derivates

of angle but if you want to see what happens when you crank up the

resolution from 12 to 16 bits, take a look at our paper "Electromechanical

delay: an experimental artifact.", Corcos et al, J. Electromyography and

Kinesiology, 2:59-68, 1992.

I played with the idea of asking to get the accelerometers off of the

ballistic missiles they were destroying for START a few years ago figuring

if they can use them to compute trajectories over a few thousand miles,

they might work for a 60 degree elbow movement but decided it was probably

more trouble than it was worth.

__________________________________________________ _________________

| Gerry Gottlieb (617) 353-8984 beauty |

| NeuroMuscular Research Center 353-9757 is |

| Boston University fax 353-5737 truth |

| 44 Cummington St. etc, etc. |

| Boston MA 02215 /\ http://nmrc.bu.edu/MCL/glg.html |

|_______________________/\ / \ /\________________________________|

| \/ \/

>

>Are there any studies which compare displacement/deformation data derived

>directly (eg. displacement transducer) with data derived indirectly via

>double integration of acceleration data (eg. with accelerometer) ?

>

I no of no formal studies but we have look at this over the years in terms

of our single joint studies. The short answer is that you cant get there

from here. You cant get a good estimate of position from acceleration and

you cant even get all that good an estimate of velocity. This is based on

the assumption that you use the usual 12 bit A/D converters and instruments

good enough but not a lot better.

The reason, as best as I can figure it out, is that you only have +/- 1/2

bit accuracy in your definition of zero and this is a bias, not a random

error. Integrating it out gives you significant errors over several

hundred integration steps which is usually a minimum number that one uses.

Accelerometer signals look great and are much better than second derivates

of angle but if you want to see what happens when you crank up the

resolution from 12 to 16 bits, take a look at our paper "Electromechanical

delay: an experimental artifact.", Corcos et al, J. Electromyography and

Kinesiology, 2:59-68, 1992.

I played with the idea of asking to get the accelerometers off of the

ballistic missiles they were destroying for START a few years ago figuring

if they can use them to compute trajectories over a few thousand miles,

they might work for a 60 degree elbow movement but decided it was probably

more trouble than it was worth.

__________________________________________________ _________________

| Gerry Gottlieb (617) 353-8984 beauty |

| NeuroMuscular Research Center 353-9757 is |

| Boston University fax 353-5737 truth |

| 44 Cummington St. etc, etc. |

| Boston MA 02215 /\ http://nmrc.bu.edu/MCL/glg.html |

|_______________________/\ / \ /\________________________________|

| \/ \/