Originally Posted by 3ll3d00d
Whether any individual beq could use fewer filters or not is probably a question of time (spent reviewing and/or testing) more than anything. I imagine
prefers to get through more films to that though.
Originally Posted by Stereodude
Overshoot of filters isn't necessarily a float point vs. fixed point precision issue. It's just how these sorts of filters work.
Originally Posted by 3ll3d00d
To recap, he wants a filter with a given slope (S) and wants to minimise overshoot (Q) so he stacks multiple filters (reduces gain per filter) to get there. That's all there is to it.
d00d pretty much hit the nail on the head. It's really about achieving a steeper slope with less overshoot to make the filter design quicker. It can certainly be done multiple ways, but as overshoot from multiple filters or multiple sets of filters interacts, it becomes more difficult to counter, possibly requiring multiple overshoot compensation filters, or at least more time spent dealing with it, which I prefer not to spend. I came to the conclusion that keeping every BEQ to 6 filters or less wasn't going to be possible without compromising on certain mixes. I'm not really willing to compromise a BEQ for the sake of filter count, personally. The vast majority of us either have a 2x4 HD or have a workaround to use 7+ filters if necessary, or they combine filters in extreme cases. So this is not something I worry about and I've chosen to maximize BEQ count since there's really nothing to be gained in my mind by spending any additional time on this sort of thing, and that's a major reason that we're at 730+ BEQs today.
Secondarily, I did some testing with low shelf filters at different frequencies, Q, and gain, and definitely found the 2x4 HD to have what I interpret as biquad precision issues down low. I am not referring to the overshoot native to a filter with Q>.707, but an inconsistency in the filter results, a deviation from the expected response. I spent some time experimenting, but questioned spending a bunch of time trying to figure out exactly which combinations of Freq/Q/Gain would have these precision issues, when there wasn't really much I could do about it anyway. One thing is for sure, they occur much more frequently as we approach 10Hz, and this makes sense, as the biquad coefficients start approaching their limits and precision becomes more important. Increasing Q and gain also moves these coefficients closer to their limits, so I thought minimizing them for other reasons above would have a positive side effect on this issue as well. But as I got deep into this rabbit hole and realizing I was spending way too much time on this sort of thing, when I wanted to use multiple smaller filters for the other reason anyway, I abandoned testing any further and moved forward.
I appreciate you bringing these questions to light and wanting to discuss them, but I don't see a compelling reason to re-visit testing biquad precision or adjust my approach to filter design, personally.