I have proposed a 3rd order Sigma/Delta Audio DAC somewhere on this forum.
Yup - here http://minimig.net/viewtopic.php?p=5136#p5136
You mentioned the downside being the need for a 128 sample buffer for oversampling. Am I right in thinking that the window of the running average filter used for oversampling is actually narrower than a single sample at the speeds proposed? If so, am I right in thinking the buffer isn't actually needed, and all you need to do is keep track of when each sample edge enters and leaves the window?
I thought about that too : IIRC, smallest period on an ECS amiga is 124. It is a good way to save some RAM blocks.
After some fiddling, the window is now 256-sample wide (actually, 248 samples because I need a small attenuation to not saturate the output driver).
248 gives : 3546895 / (2 * 248) = 7151 Hz cutting frequency. This is close enough to the real HW.
Without the filter, the sound is too metallic.
Moreover, by doubling the window one more time, we can easily recreate the LED filter at 3.5 KHz.
Maybe, someone can design a dual-mode filter that does the two different filtering based on the LED state.