Filtering Data

Applying a filter to a data stream will remove noise and make a graph of the data easier to read and interpret. Most software packages provide a filter for each channel, and the user can select how much to filter, or smooth, each trace. The sample rate at which a particular channel is set can also be used to filter the data. For example, a channel that does not change quickly, such as speed, can be sampled at a slow rate that will naturally smooth out most noise. Other channels may need to be sampled at a higher rate and then filtered to avoid losing data.

Acceleration unfiltered (red) and properly filtered (blue)

This graph shows an acceleration trace in its unfiltered state (red) and properly filtered (blue). The red trace would be very difficult to deal with as there is too much noise to properly evaluate the value at any given point, and this would also affect any maths channels created using this channel. The filtered trace retains the shape of the original graph but without the noise and jagged edges; no data is lost.

Excessive filtering can cause a loss of data

This graph illustrates the effect of filtering the same channel too much. The blue trace is the same acceleration graph as above, properly filtered, and the red displays the same channel with more filtering. The data has been smoothed far too much, with important transitions and levels completely lost.

In general, set the sampling rate for a given channel no higher than necessary. Not only will this avoid unwanted noise on the channel, it will also reduce the memory requirements and file sizes you’ll have to deal with. Look at each channel unfiltered, and add only enough filter or smoothing so that noise on the trace is smoothed out but the data trace itself is unaffected. The attached examples clearly show the effects of not enough and too much smoothing.