Hi David,

are you familiar with Matlab? If yes, I just found a toolkit that could be of some help (http://www.mathworks.com/matlabcentral/fileexchange/27561-measures-of-analysis-of-time-series-toolkit--mats-).

Good luck on your studies.

Hi, David,

If you have in your data less then 100 points, but wish to have at least 100, it is not the interpolation, but the extrapolation procedure. Usually, if you don't know much about your extrapolated signal, the only way to fill out the missing data is to create 3d order polynom using the available points near the extrapolated region and extend the polynom data to the extrapolated region. (use Matlab toolbox). It makes sense thogh when the number of missing points is about 1-5% of the the common length of the signal. All statistics measures make sense only when averaging over the set of the trials (not inside the particular signal).

Hi David

If you use Matlab, maybe you can use the "spline" function. This function is very easy to normalize your data length.

Code:

x = linspace(0,100,length(Data));

y = 0:100;

NormalizeData = spline(x, Data, y);

Then, you will get 101 points data (motion cycle 0%~100%).

Hope these can help you.

Best Regards

Kuang-Wei 

Interpolating (or extrapolating) missing data points is always serious business in inverse dynamics and has been addressed quite intensively. Its been too many years since I played around with it, but Vicon integrates a quintic spline in thier software - or else there is Matlab of course. Richard Baker at Salford is a good person to ask for the latest info. He has a biomechanics blog called Walking with Richard. You might post your query there: http://wwrichard.net/blog/

Interpolation is easily done in Excel also. I haven't seen anyone mention but if you interpolate or extrapolate the whole motion/s to 100 data points you can then express events as a % of the throwing motion. (ex. ball release occurred at 72% or max elbow ext velocity occurred at 45%). 

See this:

http://g-biomech.blogspot.com.br/2014/07/signal-interpolation.html

Hi David,

You got lot good answers already. However, here is an additional way to solve your problem: Use the software StatFree (http://www.uni-konstanz.de/FuF/SportWiss/vieten/Software/). It has a function called “tF interpolation”, which you use just by clicking the right buttons (no programming necessary). The function does an interpolation using a cubic spline and let you display your data with higher or lower “sampling” frequency. Of cause it does not give you more information but the number of data in your data file will change. Some more on this is in the help file.

Now the specific procedure: You sampled your data at the frequency f for a time t. If your first data point is at t=0s you will have f*t = n-1. Probably most of your trials have different ns. If you now want to organize all your data to have N data points you calculate a new frequency for each trial to use within the “tF interpolation”. The new frequency is f’ = (N-1)/t. E.g. originally you have sampled your data set at f = 100 Hz for t = 0.74 s. This results in n = 75. If you want to have N = 100 it follows f’ = 133.78 Hz.

Maybe this will help.

Manfred

David,

The question might need a bit of clarification.  Do you have complete movement trials that are either less than 100 data points or more than 100 data points?  If this is the case, then you can easily resample the data (interpolate) to force all the trials to be of length 100.  The effect this will have will depend on how few points you have to begin with.  For example, resampling 10 data points to 100 data points will likely introduce some artifacts, whereas resampling 90 data points to 100 data points is probably somewhat less of a problem.  That being said, the nature of your signal will also influence this process. 

I would consider the frequency of the primary movement in the context of your sampling and resampling frequencies (as mentioned by Manfred).  If the frequency of the movement is much lower than either your sampling or resampling frequencies, then you are probably okay.  Otherwise, you might need to apply your resampling cautiously.

A suitable SNR really depends on the question you are trying to answer.  Keep in mind that if you have multiple trials and multiple subjects, combining these data will have the effect of increasing your SNR, however you might lose some of the subtle variations in the signals.

Hope that helps.

Sean