Befaco modules development blog

[DaveVenom (Dave Benham)]

Continued from Announcements thread

@hemmer

@DaveVenom

Octaves has a wicked amount of aliasing. Is that intentional? Looking on a scope it looks like the waveforms have the telltale spikes indicating antialiasing has been applied. I can also see the band limiting on an analyzer. But it doesn’t seem to be controlling the aliasing.

Hi Dave, it’s certainly not intentional! There is oversampling in place but on reflection I think the default setting (x2) is not sufficient. I had experimented with ADAA for the implementation (effectively each square sums with a hard clipper between each sum), but didn’t get satisfactory results.

@DaveVenom

Does the hardware sound similar when isolating the highest frequency output?

The hardware is analog (triangle core into comparator) so will not alias but it does sound incredible harsh (especially when base frequency is high). But there is (tonal) harsh and (atonal) aliasing and clearly the first is what is desired. I’ll certainly revisit the default settings, but also look at if the antialiasing can be improved otherwise. At x8 oversampling the CPU is 6% on my machine which feels high, so maybe x4 is a compromise.

Green naive, pink x4 oversampling. (currently default is x2 oversampling, not shown)

x4 would certainly be better than x2. My Venom Benjolin Oscillator relies on oversampling, and I am not satisfied until x8 is used.

I am getting ready to begin development of a new Venom oscillator, and plan to rely on oversampling as well so it can anti-alias all manner of manipulation, including FM, PM, sync, etc. With everything I have planned, I can only hope to keep the CPU usage reasonable.

Perhaps you could add a context menu option to Octaves for oversampling amount.

It looks like you are using simd to handle polyphony. But I imagine mono is much more common than poly, and all six Octaves oscillators are active at all times. You might consider using simd across the oscillators instead of polyphony channels so that you always get simd benefit. Or perhaps apply one set of simd to the first four oscillators, and use two sets of simd applied to polyphony for the last two oscillators. I’m sure that will complicate the code, but it might be worth it.