Hi Tinker
Would it be possible to get a rundown of the oversampling digital filter the DAKSA will use? Hugh mentioned the analogue stage had a filter at 350kz. This would indicate 16x oversampling. Is that correct? Do you envisage your approach to improve on the ringing performance
There is still tweaking and fiddling going on, but right now we are probably going to settle on 8X oversampling. Which means a speed of 352.8kHz.
Oversampling filters are pretty standard devices, few are badly implemented, although some are better than others. Typically accumulator DSPs are used. I'll see if I can find good paper on these on-line and how oversampling filters are implemented. If a 16 bit signal is going in, at least an extra bit is required at the output to prevent loss of resolution, even if integer math is used (this is a different problem from digital gain, which also needs extra bits unless the gain is a multiple of 6dB). The number of extra bits required depends on the efficiency and quality of the implementation. 20 bits ought to be enough for a gorilla who failed calculus, although when I was learning the ropes I designed pretty bad filters that could have been designed by lower primates. [Ben, I doubt that! HD] Of course filter designs can be optimised by computer and so there is no need to rely on my somewhat rusty maths skills. We're still auditioning some different filters, but the technical parameters read very similarly even if they don't sound the same.
I suppose that this is a fancy way of saying, R+D is still in progress and the issues involved get a little... involved.
An analogue anti-aliasing filter is always needed IMHO (and we've talked about this before), but I don't feel it has to be a "brick wall" with 120dB attenuation in the stop band. To do
that means a really steep filter which will sound very nasty: phase shift, ripple, ringing etc. As per our preceding discussion we know that we can't hear the images, but they can cause other problems with some amplifiers. However, all that is required is to attenuate the possible images enough so that they don't discombobulate the amplifier. In most cases, about 20dB ought to do the trick.
A Sallen-Key filter achieves a 12dB/octave slope, so based on just 8X oversampling that means at least a factor of 60 attenuation (35.5dB)before we get to the first group of images. This is a good engineering optimisation: enough to get rid of the problem without an iron-clad overkill that sounds like chalk pushed backwards across the blackboard.
Incidentally, the standard S-K topology can be made to be critically damped with judiciously chosen parts.
So, no ringing there! Of course there are other circuits with this property, but there are always peculiar (and often sonic) reasons for particular choices.
Anyway. I still have a little work to do on this circuit, quite a bit of it listening and tweaking so particulars will have to wait till it's all nailed down.
