The most useful thing I ever learned about interpolation was how it relates to resampling to different rates and the specific techniques used to perform resampling when dealing with the Nyquist limits.<p>If you're removing samples, the biggest concern is whether the original signal has been bandlimited to the target rate, which can be achieved in the interpolator or by preparing the signal elsewhere.<p>If you're adding samples, then the interpolation function is burdened with reconstruction and the avoidance of artifacts in the process. This tends to inevitably lead towards the use of windowed sinc.<p>For this reason, scaling down images and raising the pitch of sound samples work out to be easier tasks than scaling up or lowering the pitch.<p>Of course, for the application shown in the article, the use of interpolation is in a different realm entirely.