Does a 4 million taps FIR filter sound better than a 16K one? Let's find out.

[Davide]

I take advantage of this thread to ask a simple question, I believe, but which I cannot find easy answer on the web.
In a FIR-based digital DRC system where there are also crossovers, how do I determine the number of taps I actually need?
In the AudioLense manual it says not to use too many taps than the necessary because otherwise they will be a series of useless zeros.
But alas there is nothing else written ...
Can someone help me?

 

[amirm]

Ideally you go down to 1 Hz resolution in bass. That means you should use 64K taps for typical 44.1/48Khz content. Having too many taps increases latency and computational power required. If neither is an issue, it is not critical to pick too high a value.

 

[mdsimon2]

@Davide The best way I've found is to try to create the filters you want in rePhase. You can see how well the filters fit your intended target for a given number of taps (and windowing and optimization).

Important to note that more taps do NOT necessarily mean more delay, it only means more delay if you are implementing a filter with taps before the impulse peak. You only need taps before the impulse peak if you are implementing linear phase filters.

IMO, the only reason to implement linear phase filters is to implement linear phase crossovers, all other correction should be minimum phase. You can have minimum phase filters with 100K+ taps with essentially no delay if you use 0.1% centering in rePhase (as opposed to middle centering).

Michael