If anybody wants some foundational knowledge that helps to understand interpolation filters like Schiit's Megacomboburrito filter, I found this guy's videos really helpful. You just need a bit of maths background, but the nice thing about these videos is you don't need to have learned differential equations (I never got that far in calc), because with FIR filters you're dealing w difference equations, which are easier to understand.
Based on Mike & Jason's posts, I believe Megacomboburrito uses a FIR filter, with 10's of thousands of coefficients, or "taps". As somebody mentioned (can't recall who that was. Oh well...), a typical way to go about making an interpolation filter is to add zeros between the original samples as the first step. Then a FIR filter can essentially transform them into non-zero values, through successive delays and multiply-accumulate operations using the chosen coefficients. I wouldn't be surprised if Schiit was using a technique other than this "zero stuffing", but there's no way to know for sure about that at this point, unless I've missed some of their posts. This guy's videos step through the process of implementing a FIR lowpass filter:
It seems that part of, if not most of the secret sauce in the MCB filter is in how the FIR filter coefficients were chosen. You can use something like matlab to generate filter coefficients for you but you'll end up with a filter that probably sounds nowhere near as good as MCB. Mike et al came up with their own way to generate the coefficients they needed for bit perfect interpolation, optimized in both the time and frequency domains. And they've programmed sharc DSP chips, presumably mostly to run the multiply-accumulate operations to implement the FIR filter with its many taps, though there may he more to it than that when it comes to what the sharc chip is doing.
at an animal auction
