A low pass filter, or filtering function, is required for both decimation and interpolation. In the case of decimation it is used to eliminate the high-frequency components in the spectrum to prevent this noise from aliasing when the signal is downsampled for conversion. In the case of interpolation, the low pass filter provides a “smoothing” function for the taps between actual samples. In either case this digital filtering perform a roll-off type function in order to achieve greater linearity and accuracy.
My personal observation is that the difference between the “analog” and “digital” sound reside in their high-frequency contours, where digital playback is able to preserve and reproduce more high frequency content and details than analog, where these fine extended high frequency content is often ”consumed” by the stamping process of the medium, the mechanical to electrical and magnetic transducers‘ conversions that take place before the analog signal gets to the amplification stages. Ironically, the same degradation or “consumption” of these high frequency low level details get ”consumed” in real life during live unamplified performances through distance from the source by natural absorption, diffusion and native instrument directivity profiles. In other words, analog may mimic the “real-time” live listening experience more accurately of sound through distance from the source signal degradations, where as digital is more accurate and preserves the original source signal as captured by the local or localized acquisition microphones.
There is much misinformation on the internet and most of what we hear can easily be explained by science and technology. As I have said numerous times, don’t discount distortion or degradation as it sometimes is required to make playback sound more “natural”.