This is where delta sigma (aka PCM1794 / ESS Sabre etc) DAC's have dirty secret.
just 1 slight correction, the PCM1794 is a dual format chip, it has different inputs for PCM and DeltaSigma.
The dirty secret you speak of, is the delta sigma noise inherent with 1bit data.
I don't think this dddac is utilizing the DeltaSigma capability, I believe they are using PCM only ?
Oz,
No it is a sigma delta DAC architecture. It -has- to use sigma delta capability.
Lets go back to the start.
- 1 bit DACs:
This is equivalent of 1 resistor being switched from gnd to vcc (+3.3V).
The OP signal 'energy' is like a light dimmer, modulated time wise to have the appropriate signal level, then filtered.
- True multibit DAC (R2R):
R2R ladder network of resistors that allow (theoretically) 2^24 discrete OP current levels. There is no 'on time'
modulation.
- Multi bit sigma delta (Sabre and 1794 etc):
For simplicities sake, lets say 5 bit unity weighted. That would be 32 individual resistors of the same value
switched between gnd and vcc. The individual step level is 1/32 x true 1 bit, but they are still time modulated
to make the effective 'bits in between. They also use other tricks like DEM (dynamic element matching) to
randomly select those 32 resistors, turning any resistor mismatch / errors into noise. There's a lot more
to it but that's a very basic conceptual overview.
I believe Sabre has the highest number of discrete OP levels for a SD DAC, with something like 256 unity
weighted bits in 'stereo mode'.
So in the 1794's case, even when run in non oversampling mode without a digital filter there are
only so many OP bits that are switched on and off at very high speed to make up the true 24 bit
OP.
In fact, disabling the digital filter may well actually spread the HF, normally out of band noise closer
to the audio band. I'm not totally clear on this - hence my question WRT showing the extended
frequency plots - especially with 44.1 RB CD
Which isn't a bad thing, it means if we feed it a clean 24bit music file it shouldnt actually need any further filtering, and in NOS mode the dac itself won't mess with it any further, just simply convert it. I like the idea of that.
Yes - that would be nice but it doesn't happen. Any SD DAC needs to do 'time splicing' to get the effective bit resolution,
there's no free lunch.
It's one of the reasons I was waiting for the next generation R2R DAC's - which got dumped. There's a lot to be said for
just a very good R2R DAC.
There is a fork in the road, and there is a choice to make.
Do we pursue a DSD Dac, do it better than the other limited offerings available? With this option, filtering is needed to clean up the dirty secret.
Or ,,,,,, do we pursue a 24bit or 32bit PCM hi-res solution? With this option NOS mode is the way to go, I do agree with the way the DDD chaps have chosen to do it.
I think trying to do both formats, may not be ideal for either??
Trying to do both formats -right- certainly makes things more complex, no doubt.
The clocking and interface requirements are quite different.
In either case, DSD or PCM, Reinhard is right, half the puzzle is sorting out the "feed" of the data.
That's no problem. I don't want to give too much away but the 'feed' is only part of the issue.
The end game is a) Complete isolation from source (PC) and b) absolute lowest jitter at the DAC.
DDD has made a very good effort but it is still some distance from absolute cutting edge and ticking all the
boxes.
I was at Stevens a week ago. Doug bought around his laptop, configured for music playback with Jplay, out to Offramp, into Killerdac. It just didn't have the liquidity and clarity of the Wadia spinner into Killerdac. Zen, I'm hoping your solution will get us there, as we're all sick of burning.
Yeah, same, I'm mostly using PC sound card OP (emu1212) these days purely for flexibility - even though the quality is average.
OK - there's work to do and these posts don't get it done any quicker
cheers
T