>>3792999(writing this shit for the 3rd time, because I'm retarded...)
a good question as well. sharpening can be a science. imo, it depends on the material and the purpose. you most likely are familiar with the difference between sharpness and sharpness impression. so, for me it is not only the question how to sharp but also when. as I see it, sharpening should be the very last step before transcoding. and if you have the right original material, which means sharp but with no sharpness impression (so: soft but from a good sensor with a good lens ... avoid in-camera sharpening where possible), it can be sufficient to simply use the right resampling algorithm to get the desired sharpness impression.
pic related shows an overview over some resampling algos. when you compare them to the wavelet analysis below, you can see that resampling mostly affects the highest frequencies (at least the ones which tend to ringing and blocking). for a high resoluted output device this can be satisfactory. and this way you render the least amount of artefacts and loss due to impression-sharpening. (at the bottom of pic related you see a result which is only done with resampling to hd.)
but if you render for sd, or projection devices e.g., then you want to enhance the sharpness impression. I personally more or less exclusively use unsharp masks on the luma channel. for the radius I tend to use something near to the frequency like in the 3rd from top pic of the wavelet analysis. maybe the 4th. sometimes I use several soft masks on different frequencies.
it's a matter of taste again. imo, there should be a gentle fall-off of sharpness to higher frequencies. I dislike this harsh hd-look you see often today. I guess this will look cringy in 20 years. film is, like any other art, the process of omitting information. there is so much detail in reality I really don't want to see when watching a movie.