Do you think Digital Products have too many filters? Or not enough? How many do you use?

caesar said:

...Any one here play around with different filters?

Click to expand...

Yes, I played with four PCM filters. I've listened to 1) NOS (no filter), 2) soft linear-phase half-band FIR, 3) brickwall linear-phase half-band FIR and 4) sub-Nyquist frequency stop-band brickwall linear-phase half-band FIR. They all sounded very different from each other. I've not yet had opportunity to play with any of the minimum-phase digital filters, but would like to.

Your notion of the one "best" filter for each digital format is problematic, for it depends on what one objectively defines as, or more problematically, subjectively perceives to be the one best in each case. Of the four filter types I've extensively auditioned, each has it's own subjective benefits. In theory, the objectively correct filter anti-image function is any that fully removes the ultrasonic image band - which extends all the way down to 22kHz for CD- while maintining response flatness fully up to 20kHz. Hence, the ubiquitous SINC function (brickwall) FIR filter. Practical implementation, as well as human perception, unfortunately complicate the matter.

For example, half-band type filters are by far the most common implementation of brickwall filter because they utilize the least processing resources, however, half-band filters are also notorious for allowing some violation of Nyquist at the highest frequencies, allowing aliasing products to then fold in-band should the original recorded event have contained any information above 22kHz. A definite no-no that the audio industry has long pretended not to see. Consequently, one of the original objectives of the apodising filter concept was to chop off any aliasing products folded back in-band should a half-band A-to-D anti-aliasing filter have been utilized at recording. Of course, along with removing any aliasing products, such filters also chop off some legitimate high frequency information. So, which filter function is best can depend on the A-to-D implementation as well as the frequency spectrum of the recorded music.