GGPSA Biomechanical Data filtering
by  Giannis Giakas PhD
1999



Parameters file structure
-------------------------
The parameters file must have 7 numbers in one collumn.
There should be nothing else but numbers.

There are in the following sequence:
Sampling frequency
Number of Poles
Percentage for noise estimation
Signal to noise ratio
Number of extrapolated data points
Order of the butterworth digital filter
Undersampling coefficient

--------------------------------------------
Explanation  and recomendations
--------------------------------------------
Sampling frequency
Self explained


Number of poles  (recomended 20)
These are used for the modelling of the signal in the frequency
domain. The poles calculated are used for the calculation of the
PSD of the signal and for data extrapolation.
The higher the number of poles the better the modelling of the 
signal. If you include the maximum number of poles it will model
exactly the signal. A value of 20 will do the job in most cases.
If you have a "strange" signal, especially at the edges, you may 
want to increase this a little bit and vice versa if you have
a projectile movement (e.g. shot put).
This must be an INTEGER value.



Percentage for noise estimation (recomended 80)
This will range from 1-99 (expressed as a percentage). It will
estimate the mean PSD of the signal from YOUR_VALUE up to 100%.
As noise is usually in the higher region of the power spectrum
a value between 60-80 will give a good estimate.

Signal to noise ratio (recomended 50 but...)
The better the quality of the signal the lower the lower the 
value and vice versa. For optoelectronic systems with good 
calibration an approximate value of 50 will be OK. For manual
digitising, poor resolution, poor calibration etc, you will
need to dramatically increase it (100, 150,...).


Number of extrapolated data points (recomended 15 but...)
The signal will be extrapolated (both sides) using linear prediction.
Remember that a high number of extrapolation is not always good 
especially if you are interested for the edge of the signal. A number 
higher than a threshold (this is related to the number of poles) will
not make any great difference.
This must be an INTEGER value.


Order of the butterworth digital filter (recomended 2)
The higher the order the sharper the filter at the cut-off region. Some 
investigators have used a forth (4) order as well. 
This must be an INTEGER value.


Undersampling coefficient (1 for no undersampling)
This is if you have oversampled your signal, you have the "comfort" of
undersampling it. Undersampling will speed up filtering and in some 
cases it will do better filtering. For example, if you collect walking 
video data with a sampling frequency of 1000 Hz while you could do as good
job as with 200 Hz. Then undersampling is recomended (by me !!!).

For half data (half sampling frequency) then the value is 2, For the 
above example to go from 1000 Hz to 200 Hz  the value should be 5.
This must be an INTEGER value.