Smoothing signals is really important to remove unwanted artifacts. Be sure to understand the effect of your filter, test it yourself, and make sure to consult literature.
Anyone who has performed signal analysis has some understanding of filters, and biomechanics is no different. The underlying assumption is that some component of the signals is not related to the movement, but noise due to the measurement system.
Fair enough, nothing is perfect! Now, how do we measure what this noise is? It’s a pretty complex process, and is dependent on a few factors. Namely, there is intrinsic measurement error with the sensor itself, but also how much that sensor actually represents the movement of the underlying skeleton (soft tissue artifact). The sensor’s ability to represent the underlying bone is by far the biggest contributor to noise (and error) within signals. So, with the sources identified, what’s next?
In the infancy of biomechanics, an experiment would ensue, but now, we can look at previous studies, cite them, and go with that filter frequency and type. Yet, there is a little bit more to this. Consider an experiment with markers placed on the body, and we report joint angles. Now, do we filter only the raw data (marker trajectories) or just the joint angles, or both? It really depends on who you ask, but both methodologies are widely present in literature. In my experience, it varies from lab to lab.
Now it gets even more complicated when filtering inverse kinematic solutions. If you only filter the input, there will be higher frequency noise in the joint angles produced by the IK (try it out!). A friend and guru of IKs told me you must filter the IK output, which I buy given his experience. But again, the devil is in the details; filtering the output joint angles will make the IK violate constraints, which is also not ideal (good luck writing that into your methods). One way around this is to filter the relative quaternions in the pose matrix, and this preserves constraints. Yikes! It’s getting a little bit complicated and seems to be moving away from the biomechanics question we were examining.
Time to take a step back because a lot of people smarter than me have looked at this a lot, which is why I actually feel pretty good citing previous studies. But be careful, just because it has been done in the past, doesn’t mean it’s ideal. Now that we have skeletons that we can overlay on videos, it’s pretty easy to see the effect of a filter, and I recommend doing this a few times to understand the consequences of these choices. See picture below. Identical trial, going from raw data, 24 Hz filter, and 16 Hz filter. For pitching, 16 Hz is pretty standard, and as you can see, that hand is attenuated.
So, what to do? Know the effect of your filter! If you can back up your decision, go for it! Speaking generally, literature is an asset when it comes to selecting the filter type and cutoff frequency. Having reviewed many articles, if someone submitted an article with a higher cutoff, as long as some good supplemental material was provided, I wouldn’t be too bothered.