[sdiy] Novation peak NCOs

[rsdio at audiobanshee.com]

On Apr 21, 2017, at 4:01 PM, Tom Wiltshire <tom at electricdruid.net> wrote:
> On 21 Apr 2017, at 23:15, James J. Clark <clark at cim.mcgill.ca> wrote:
>> Regarding the 24MHz issue, the documents that I have read on the Peak (https://uk.novationmusic.com/peak-explained) only say that the DAC is oversampled to 24MHz, which is pretty typical for sigma-delta style audio DACs these days. Its a way to get high bit widths - typically 16 or 24 bits. It does NOT mean that the signal processing is done at 24MHz. In fact the "peak-explained" document specifically says that the wavetable waveforms are band-limited before being stored in the table, to avoid aliasing. This wouldn't be necessary if the signal path was running at 24MHz.
>> 
>> I would guess that the signal path is running at whatever the DAC sample rate is (which is not 24MHz - that is just the master 1-bit clock rate).
> 
> Yes, I wondered about that. Thanks for clarifying.
> 
> So we're thinking this is an FPGA synth design with some high-but-not-extraordinary sample rate, that then uses a 1-bit DAC run-in at 24MHz for output. x64 oversampling is common for this type of thing, so let's assume that for a moment - that gives us a 375KHz sample rate, roughly twice the 192KHz rate we're more used to.
> 
> Is that a fair summary? Is that what we think they're up to?

It’s probably a fair guess as to what they’re doing, approximately.

However, I caution against assuming that a 375 kHz sample rate produces the same results from a 1-bit core as 24-bit or even multi-bit sigma-delta core can achieve. Search for any of a number of papers from Stanley P. Lipshitz and John Vanderkooy which explain why 1-bit is a bad idea for professional audio.

Basically, the problem with 1-bit is that there is no “zero” - a signal must always be rising or falling, and can never be silent. It’s pretty much always full of ultrasonic noise. 1-bit sigma-delta has serious issues with distortion, limit cycles, instability and noise modulation. It’s not that sigma-delta itself is a problem, because 5-bit or greater sigma-delta can be perfected to remove all of the problems of digital quantization. Another issue, if I understand their papers, is that proper dither, which is necessary to remove digital artifacts, requires a dither amplitude of two LSBs, and that’s simply not possible with 1-bit.

If you’re keen on this sort of stuff - and even if you’re not going to implement all of the math yourself - then join the AES for access to their library, and search for Vanderkooy and other authors who really know their stuff when it comes to digital audio.

Brian Willoughby
Sound Consulting