High end all digital dsp crossover ?

[twest820]

Originally Posted by abraxalito

If you can choose external DACs its very easy to screw up the time alignment between channels by using different ones.

It's also tricky to distribute the MCLK. DACs usually cost less than LVDS transcievers (or whatever other impedance controlled interconnect one may be using for clock distribution) and the jitter cleaner needed to get the clock back to the quality level it had coming out of the XO. If one's working with high performance DACs I'd be careful about noise coupled in from sending I2S over an interconnect as well---doesn't take much to blow 120+dB DNR.

This is one of those things where boards need to be built both ways and measured to determine how well splitting out the DAC actually works. That takes more time than most DIYers have and measurement gear well beyond what most folks would consider a hobby price point, though. So, unless one is planning on auditioning many different DACs, it's arguably lowest cost (in both time and money) to integrate the DAC onto one's board. This doesn't work so well if one's wanting to buy preassembled boards and connect them together but, in the absence of an interconnect or impedance standard to enable such integration, such integration isn't exactly set up for success anyway.

If one's working with lower quality clocks---WM880x outputs, for example---and midrange DACs---say 100-110dB DNR and -90 to -100dB THD---board to board integration's less challenging as 10 or 20 picoseconds of additive white baseband jitter becomes tolerable for audio data sampled at 44.1. But that's a different direction than the DIY focus on flagship DACs and high sample rates.

[qusp]

nope, I have all of that sorted, ackos master clock module and dacs, as well as ians fifo all have multiple and buffered impedance controlled i2s and mclk outputs on u.fl/w.fl uBNC and my dacs have impedance controlled i2s and mclk inputs. this is all fine, I just need a dsp! preferably one that also plays by the book. the fifo in its current form is obviously no good for multichannel, but all else above is

besides if you use software that has phase adjustments and the same system is used to measure and setup the speakers, wouldnt any existing phase difference between the channels be neutralized? i'm not looking to have that problem, just sayin

[twest820]

Great, can you share your phase noise measurements of the clock distribution?

[qusp]

Originally Posted by twest820

Great, can you share your phase noise measurements of the clock distribution?

not mine, but Ian should have something related thats not much more than the buffer phase noise plus clock. the master clock module + buffer of Ackos is only brand new, its a bit pricey so havent sprung for it yet, not till i'm finishing up the chassis. the layout is certainly fine and all boxes ticked

you'll still need a buffer on a PCB if you want to drive multiple dacs (particularly sabre dacs as they prefer all dac cells driven), so isnt the only comparison that of u.fl interconnect vs trace?

the point is moot though, the whole point of this thread is because we dont want an onboard dac of someone else choosing and we are sick of only being offered options that have one. so remind me what you hope to achieve?

[qusp]

perhaps you are offering to layout and pay for individually tailored dacs for everyone?

[twest820]

Originally Posted by qusp

remind me what you hope to achieve?

What I was saying in post 11 was it's unclear how the jitter of on board and cross board clock distribution compares and that DIYers typically lack the measurement gear needed to find out. Unfortunately that lack seems to apply here---sorry, not sure who Ian is or where I'd search to find his measurements---so the question remains unresolved.

This isn't to say one can't hold femtosecond timing tolerances through connectors and across multiple boards, just that it's hard to do at DIY friendly price points when assembling boards from different vendors. This is essentially the problem that killed Rambus and it's one that the engineering team I was on in the early 2000s solved routinely. However, the hardware assemblies started at USD 15,000 and it took multiple years to get suppliers lined up that could reliably hit the design requirements. There's been ~10 years of improvements in board fab since then but I would still expect picoseconds of degradation. One can likely get away with not applying a jitter cleaner to that with ESS DACs as ESS does some rather clever things in computing input referred clock errors. But other DACs are not so tolerant and, even for ESS DACs, jitter rejection depends on the incoming phase noise spectra.

If one wants to take it on faith a standalone DSP is integrating as expected with a third party DAC that's certainly a choice an individual can make. Just don't expect others to be willing to make the same assumption.

[dahlberg]

Ok, so there is difference of opinion if this is even possible. Are there any way to work around problems that might occur ? Separate clocks for instance ? Is spdif a better/safer way to get the expected results ? I also have some difficulty to see how a few centimeters of u,fl connected cable could mess up that much compared to the same length of trace. But as I mentioned earlier I have no real clue Another way to approach this would be to actually decide on onboard dac's as well. Ess seems to be something we all could agree on, then there is just analog output to fiddle around with.

[abraxalito]

ESS wouldn't be my choice - but as I'm already (slowly) working on my own design for a sound-quality-first digital XO there's no way I'd want to discourage anyone else from using ESS. The more available options, the merrier

[twest820]

Originally Posted by dahlberg

Ok, so there is difference of opinion if this is even possible.

It's very possible. All I was saying is it's unfortunate the measurement part of "trust but verify" requires a larger budget than most DIYers have. qusp's point that since that leaves one with "trust but verify by design" one might as well verify the design and get on with life is a fair one.

Originally Posted by abraxalito

ESS wouldn't be my choice

Yeah, mine either; dahlberg, you can find more about my reasoning in the ES9016 thread if you want.

[qusp]

no, no question its possible, twest is just arguing over minutea ( thats what we do here isnt it?) because thats all hes got left to argue about. IE fem-to second timing on individual dacs that will be used to drive individual amps with individual drivers and played in asymmetrical rooms, through speakers with -60-70dB distortion.

I didnt say there was no difference between trace and u.fl, I simply said that was ALL we were talking about and that an all in one unit is out of the question, unless each person is designing their own illustrated just between 4 people on one page having a difference of opinion…

would I prefer an all in one PCB if it did everything I want? well perhaps at this stage no, because I already have the dacs, but otherwise yes, I agree thats ideal; will that ever happen? unlikely.

getting everyone to agree on ESS? lol

as for verifying, yes I have to trust, I trust that the designers use the same tech for much more systems critical outcomes than audio; Ian for medical systems, Acko for military and all PCBs go through rigorous probe testing etc. after that I just have to swallow and move forward if I want the object and sound I want.