Asynchronous I2S FIFO project, an ultimate weapon to fight the jitter

DSD over PCM is just DSD converted to PCM in the player afaik, in which case the fifo should do just fine. it pretty much has to be converted, there is no other place for it to be converted before the dac. if it passes through as PCM then yes it will I would think, if it doesnt, then no it wont. pretty simple equation. fifo is built to reclock i2s, that is all…
 
Hello Ian,

first, thank you for sharing your fabulous project and for the great job to organize the GB.


In the search for alternatives to Crystek's XO, I would like to know if NDK's oscillators, 90.3168 MHz & 98,304 Mhz, are "compatible" with Dual XO clock board.
Those XOs would prevent the double speed mode.
If your answer is affermative, please let us know the right settings.

I attach the datasheet.

Greetings

hirez69

Hi hirez69,

You may try 90.xxx and 98.xxx XO for Fs=88,96,174.2,192,384 Khz . Theoritially it could but not guaranteed because it almost reach the limitation of internal timing. But for sure those frequency will not support 44.1 and 48KHz,for the FIFO do not have a 2048*Fs mode.

Good luck

Ian
 
Morning! (at least for me :) Yea, I know FIFO doesn't do DSD, I was more curious as to whether it might 'pass through' DSD. I'm also curious about about how the DSD over PCM 1.0 spec. will 'play' with FIFO. I'm using an eXD PCM/DSD board in one of my builds that's using this spec.

Hi MisterRogers,

Sorry for the late reply. The current version of FIFO support I2S( or left,right justified now) only. They will not accept DSD stream, nor let it pass through. Actually DSD is totally the different story than the I2S. It need more memory space even running at same Fs. It comes with different timing as well. The multi-channel FIFO may support both of them at same time. But at least I need 64Mb on board memory.

Have a good night.

Ian
 
Thanks for taking the time Ian. I figured as much. Not a show stopper; the eXD board I'm using for USB->I2S has both I2S and DSD headers. I'll just need to build in a DSD mode selector and keeping the I2S/DSD streams separate (there's a firmware option to multiplex I2S/DSD I was considering). I'm very much looking forward to building my DAC using your FIFO kit. Thanks for all of your time and effort Ian!
 
Disabled Account
Joined 2008
Hi hirez69,

You may try 90.xxx and 98.xxx XO for Fs=88,96,174.2,192,384 Khz . Theoritially it could but not guaranteed because it almost reach the limitation of internal timing. But for sure those frequency will not support 44.1 and 48KHz,for the FIFO do not have a 2048*Fs mode.

Good luck

Ian

Ian

thanks for the answer. NDK 90.x and 98.x XOs are very expensive and before buying I want to be sure. I think that the limitation for 44.1 and 48Khz closes any possibility, unless I accept to do "XO-Rolling".:)
Any chance to add 2048xFs mode?

hirez69
 
ok i've moved the convo over here away from the GB thread, sorry about that Ian, i'm trying to find a solution to the issue under discussion above, others just want to poo poo any suggestions, rather than working to solve the issue. how about using a very low phase noise clock plus something like Si5315?

then of course we have options to make a quad clock board, up the memory to allow 2048, or upsample in software like puremusic on the fly and just use the NDK. they will always be upping the speed of dacs though, whether we need it or not …

the DSPLL type, whether SiLABs or vectron or some other manufacturer, would allow us to simply keep pace as far as the dacs can match. the ESS for example will run much faster again with OSF disabled, but I dont think it will allow a doubling of speed of MCLK again from the 90.x/98.x without overheating and burning. dunno maybe some sort of phase change heatsinking solution on the dac would allow it hehe
 
Last edited:
Would you please specify where is used? In which project here at diyaudio or elsewhere?



My point is clear: si570/571 proposed by you isn't suitable for audio. It is just a generic XO with a variable output frequencies.

Ian has trialled this clock himself subjectively and commented on it on diyaudio.


This doesn't means that this project isn't valuable. It's beautifully conceived but not complete (IMHO)

I am interested to hear what features you find lacking / missing from this project?
 
his experience of the part quite clearly extends no further than the reading of datasheets. or rather one specific number which he has heard is important. it is, but its far from the end of the story when it comes to practical application. still waiting on an alternative that I didnt mention already before he started arguing; but maybe we just let him go design something better? i'll keep an eye out …

the project is about as good as it gets and the collection of experienced members, hobbyists and professionals alike that are gathering around or have already bought it is a pretty good indication.

the points under discussion are a matter of ease of use under fairly extreme variation, from the slowest to the highest speeds available. upsampling x 2 in software is not going to do any damage, so its a viable solution.

at least one other project, the widget used the Si parts as one option, though I think they used one of the lower spec parts.
 
Last edited:
Disabled Account
Joined 2008
Ian, could the board or firmware be rejigged to allow the dual clock board to contain say a 45.1584 plus 98.304 XO's? then the only speeds not covered would be 48kHz and 354.x kHz which in my world are pretty rare

great idea, qusp.
do you want to say 352.8Khz? This frequency is not so rare,
is the one used for the DXD recording format:

HIGHRESAUDIO | Home of audiophile music


Fidelio Audio

hirez69
 
yeah thats what I meant correct. too many numbers bouncing around my head.

yeah but thats just a conversion to PCM from DSD, not a recording format, there are no ADCs other than the ESS one that will do such speeds afaik. most high rez recordings use the 24x multiples for 96kHz and 192kHz. You could use whatever was more important to you, you could have both XOs as 22.1 multiples ie 45.1584 and 90.3168. but for me I really only want 96/192/384 for hirez and the 45.1584 will cover everything up to 176.x for the 22.1.

edit: the page you linked shows only studio masters in wav at 352, or DSD, no DXD.

4 files on that page does not make it common ;p
 
Last edited:
Member
Joined 2004
Paid Member
Wave at 352 seems to be what DXD actually is. Digital eXtreme Definition - Wikipedia, the free encyclopedia

Hires files really exist at 96 (common) 176.4 (RR HRX and a few others meant for mastering for CD), 192 (nice big number so it must be good), DSD (mostly rips from SACD but a handful of actual released files exist). There are a handful of 2X or 5.6 MHz DSD files but they are not a real distribution format and make for gigantic files.

Supporting DSD would require a seperate path through the fifo for the extracted DSD data. if Ian could recognize the DSD flag (DoP) and pull the data inside the FPGA and route it to a separate set of outputs for DSD inputs on the DAC chips that support DSD this would be a leading edge device. That may be a pretty involved project. The clock rates are compatible with 176.4 KHz.

This is like a 78 RPM microgroove vinyl disk. I have one. Getting together a turntable to play it was a task that took some time. It did sound good, had about 8 minutes of music on each side and was clearly a dead end. If you want to make something that can play an odd format because it can you will have a feeling of accomplishment afterwards. But if you want music its a lot of struggle with no real reward at the end.
 
+1 to a FIFO based I2S/DSD solution (bypass, DSD FIFO, etc.). Along with multi-channel support, we can keep Ian busy for the foreseeable future :) It's funny how we all are; give us something that a year ago simply wasn't available, and even before using/deploying - we want more. I know I'm that way. So for all of us, I'll say it again Ian - THANK YOU! If/when you get to these other things, we'll be right there. But rest, pace yourself, enjoy life.

Re: XO group buy - if we can come to a consensus, I'm more than happy to organize and drive it. My needs are pretty simple: A quality pair of clocks that'll handle all frequencies up to and including 384mhz
 
+1 to a FIFO based I2S/DSD solution (bypass, DSD FIFO, etc.). Along with multi-channel support, we can keep Ian busy for the foreseeable future :) It's funny how we all are; give us something that a year ago simply wasn't available, and even before using/deploying - we want more. I know I'm that way. So for all of us, I'll say it again Ian - THANK YOU! If/when you get to these other things, we'll be right there. But rest, pace yourself, enjoy life.

Re: XO group buy - if we can come to a consensus, I'm more than happy to organize and drive it. My needs are pretty simple: A quality pair of clocks that'll handle all frequencies up to and including 384mhz

weve just spent the last couple pages here and the GB thread discussing that this is impossible with current hardware no matter what clock you use unless its able to change frequency on the fly, so I would say your needs are not simple at all :p. funny thing is with the right output filter you kinda dont need the dac at all with DSD

thanks for the info on DXD Demian, still doesnt make it any more important to me as recordings in that format natively are VERY rare and I care not for nordic folk music so i've been watching from afar mostly, I dont care to pay more for upsampled or 'remastered' music.

I mainly use 96/192/384 for video and my own production needs, so I can playback tracks i'm working on for kicks and render analogue modeled synthesis more realistically
 
Last edited:
Yea, my tongue was was firmly planted in my cheek with that comment :) On point and of general interest in this topic; I posted a similar query (switching I2S/DSD) in the TP forum, and Russ responded with a top view of a mux/demux board designed for just that. He indicates that it'll be available very soon (emphasis mine).

Less on point, I've amassed quite a cache of dff/dsf SACD ISO rips of music I enjoy (Dire Straights, Steely Dan, Diana Krall, etc.) that sounds fantastic through my I2S/DSD multiplexed eXD board/BIII. Eliminating the PCM conversion is quite an improvement (IMO, subjectively, blah blah). Though...I'm always up for some stellar nordic folk music :)
 
Member
Joined 2004
Paid Member
Probably the best value and almost the best you can get for less than $100 each are the Crystek CCH. A group buy won't save much over the single unit price at Digikey or Mouser. They do benefit from a really good power supply (see my circuit earlier in this thread that was triggered by remarks from Gordon Rankin about those oscillators). The synthesized clocks are better than I first thought but still not in the same class. I know of a better solution at $100 ea (min quantity of 10) and $500 plus per oscillator will get a 20 dB improvement but would require 2U of rack space to support. And probably not be audible.
 
Demian, you wouldnt happen to know at what point crystek will sell custom speed parts would you?

touche =) sorry misterogers: i'm used to that strange textual blindness here.
none of the artists you have named have released anything that could have been recorded in any of these formats. the best you could hope for is a reel to reel digital rip


edit re crystek: i'm going to email them about MOQ

the new CCHD-575 looks promising too. of course its got 'only' -90dB at 10hz which they term 'ultra-low phase noise' as well, but apparently this isnt good enough for feeding my chickens accurately each day … (Demian comment not aimed at you, just someone slagging the 570)

the 575 is marginally better than the 950 which I believe still is or at least was, the standard part on the buff II.

the 957 doesnt even seem to be available above 50MHz, rather than just not being a standard part like with the 950 and 575 but i've sent them an email anyway
 
Last edited: