@JohnW:
Thanks for sharing about the AK4498/AK4491 & AK4499 listening impressions.
Very interesting information that the combo could sound better than the 4499 ... And to me it is no less interesting that the combo may actually accept 1.536 MHz & DSD1024.
If only I knew of a USB/ethernet/other interface that could get thus high frequencies out of the computer ... Any chance you know of such an interface?
Cheers,
Jesper
Thanks for sharing about the AK4498/AK4491 & AK4499 listening impressions.
Very interesting information that the combo could sound better than the 4499 ... And to me it is no less interesting that the combo may actually accept 1.536 MHz & DSD1024.
If only I knew of a USB/ethernet/other interface that could get thus high frequencies out of the computer ... Any chance you know of such an interface?
Cheers,
Jesper
I'm currently working on a small discrete DAC / XMOS solution that will support DSD1024 - its complicated for us that we need to use the MQA libary and the XMOS struggles with 768KHz PCM (Due to the MIPS required for the MQA library) - but if we removed MQA (atleast for the time being) then we should have the XMOS MIPS for DSD1024.
I need to understand how AKM support DSD1024 / 50MHz SDM when the SCF in the AK4498 operates at 256fs - I suspect that they could parallel load the DSD1024 into the 6 bit SCF somehow... I'd need to sit down and figure if this is even possible... what does it mean for the original DSD1024 data...
This is all on the presumption that AKM dont decimate the Native DSD to some form of PCM DXD... its my gut feeling that the DSD Data is not processed native (atleast with higher rate DSD).. but I really could be wrong.. one could possibly pull some funky tricks with the 6 bit SCF "array"... Maybe.....
I'm currently trying to hand build some PCB's ATM and my solder paste has dried... I wish I had more hours in the day (I tend to only have a clear head to think about such matters when I'm in the bath)!!!...
Last edited:
We would appreciate to hear from you if and when you know the answer. Always interesting to hear about such things.
Also, given the very high sample rate DSD AKM has decided to support, one wonders if they have any plans to do an update of something like AK4137 to bring it more in line with the newest dac capabilities
So after giving it some thought, I cannot see how AKM can support anything above DSD256/DSD512 without decimating the SDM data.
As the SCF are clocked at 256fs, you could atleast theoretically toggle the SCF MSB and input data to DSD256. I was wondering if you could do some funky moving average filter with the other SCF's, but I dont see it... (thats not to say they dont).
AKM are using DDR type double edged clocking (in stereo mode) - so MAYBE any (if it exists) Native DSD data could be pushed to DSD512 if they where being clever...
Without getting really really funky, I cannot see how they can support NATIVE DSD above DSD512, I suspect DSD1024 might be a decimated DSD mode...
The Datasheet indicates a couple of DSD filter paths - I took the second path to be some form of a moving average filter so could be considered Native DSD... maybe this path is limited to DSD256/512 (SDM 512 If they used double edge clocking)...
Last edited:
Given the very high sample rate DSD AKM has decided to support, one wonders if they have any plans to do an update of something like AK4137 to bring it more in line with the newest dac capabilities
At higher DSD rates, such as SDM512 and above, it gets significantly harder to keep the "pulse per pulse energy" constant, a greater portion of the pulse energy contribution is within the rising and falling edge transistion periods - and any "Tri-state" essentially undermined logic state that occurs during these transitions (I'm trying to say you are spending a great portion of your pulse cycle period in the Logic's "Linear" mode (including ringing / under / over shoot), as apposed to a steady "Saturated" = Clean High / Low state)..
I see no reason to push beyond SDM512... Clean 25MHz pulses (50MHZ Clocks) are still within the realm of Fast discrete CMOS logic without getting into silly power dissipation levels...
SDM1024 is just one step too far IMO. If one wanted to retain a 75% RTZ Pulse period for SDM1024 one would require a 200MHz Clock - at 200MHz the larger geometry discrete CMOS Logic start to experience silly power levels...
On the little Discrete DAC's I'm currently working on, we can support SDM1024 (with our ~100MHz Audio Master clocks), but we reduce the RTZ Pulse period to 50%, this results in worst Dynamic range due to reduced VOUT from the DAC array at FS (0dB).
Back to the AK4137, We really like this part - BUT its so full of bugs (we have to pull many work arounds with the aid of an external FPGA), if you know what your looking for, you can read between the lines of the datasheet and recognize where various modes have been dropped due to design bugs... it be good to get an updated AK4137 where SDM256 worked correctly (in upsample AND downsample mode) - maybe even SDM512... but atleast correct the bugs with the intended modes!
Last edited:
Hi Mr Westlake
Nice to see you here
An thanks for sharing all this, both your valuable technical knowledge and also your subjective findings as I don't mind them at all.
In fact, I am rather a bit upset by the arguments that followed, but then I just skeep them.
I would be interested to know how you rate the potential of this combo vs R2R DACs, including sound potential, be it here (with or without flames) or per PM time permitting.
Thanks a lot again
Claude
Nice to see you here
An thanks for sharing all this, both your valuable technical knowledge and also your subjective findings as I don't mind them at all.
In fact, I am rather a bit upset by the arguments that followed, but then I just skeep them.
I would be interested to know how you rate the potential of this combo vs R2R DACs, including sound potential, be it here (with or without flames) or per PM time permitting.
Thanks a lot again
Claude
Claude,
Thank you
Not easy to answer as there are many poor R2R DAC's - but what I can say is that in my experience there is a huge sonic difference between a discrete DAC's and "intergrated IC" DAC's - the same can be said for Discrete opamps over IC opamps etc...
There are so many advantages to separating the Analogue domain of the DAC design (by "Analogue domain" I referee to the actual DAC array / SCF / Clock distribution to the DAC Array, internal power / ground distribution etc.) - the "digital" sections that are really operating in the Analogue domain.
So with AKM splitting the digital / analogue sections they have gone a decent way to avoiding a lot of compromises an IC designer has to make - it be a dream for any higher performance DAC designer to be able to have the luxury of separated Digital and Analogue die, no longer having to “shoe horn” an ultra high performance “analogue” circuity onto a digital process. So much work is done trying to fairy dust an analogue circuit out of “digital” building blocks – I know this as I have been part of many Audio IC designs – and its always the same limitations everytime!!!
I really big advantage is being able to isolate what effectively becomes common-mode digital substrate noise from the analogue section – such is an IC designers dream!
AKM have taken a page from much earlier DAC designs where IC designers struggled with the debilitating limitations of Digital IC processes – the Crystal CS4328 was one example of a High performance DAC for its day containing two isolated (and different IC processes) silicon die on a single IC leadframe – so you had in effect two IC's in one physical package.
Philips also expanded on the performance of the SAA7350 by using its undocumented publicly at the time SDM modulator output test pins and designed the BiCMOS SCF DAC – the TDA1547 (DAC7) resulting a two chip high performance DAC solution where the SAA7350 served as the digital section for the "Analogue" DAC7.
Back in the early 90's, there was some form of collaboration between AKM and Crystal semiconductors so I wonder if the spirit of the CS4328 remains in the heart of the original AKM design team (if they are still even around today)
In answer to your question, I suspect a WELL design R2R DAC will still sonically outperform the AK4498 combination.. however the point is that the AK4498 sets a new standard in (atleast one that's in production “today”) “off the shelf” DAC IC solution / chipset...
Thank you
Not easy to answer as there are many poor R2R DAC's - but what I can say is that in my experience there is a huge sonic difference between a discrete DAC's and "intergrated IC" DAC's - the same can be said for Discrete opamps over IC opamps etc...
There are so many advantages to separating the Analogue domain of the DAC design (by "Analogue domain" I referee to the actual DAC array / SCF / Clock distribution to the DAC Array, internal power / ground distribution etc.) - the "digital" sections that are really operating in the Analogue domain.
So with AKM splitting the digital / analogue sections they have gone a decent way to avoiding a lot of compromises an IC designer has to make - it be a dream for any higher performance DAC designer to be able to have the luxury of separated Digital and Analogue die, no longer having to “shoe horn” an ultra high performance “analogue” circuity onto a digital process. So much work is done trying to fairy dust an analogue circuit out of “digital” building blocks – I know this as I have been part of many Audio IC designs – and its always the same limitations everytime!!!
I really big advantage is being able to isolate what effectively becomes common-mode digital substrate noise from the analogue section – such is an IC designers dream!
AKM have taken a page from much earlier DAC designs where IC designers struggled with the debilitating limitations of Digital IC processes – the Crystal CS4328 was one example of a High performance DAC for its day containing two isolated (and different IC processes) silicon die on a single IC leadframe – so you had in effect two IC's in one physical package.
Philips also expanded on the performance of the SAA7350 by using its undocumented publicly at the time SDM modulator output test pins and designed the BiCMOS SCF DAC – the TDA1547 (DAC7) resulting a two chip high performance DAC solution where the SAA7350 served as the digital section for the "Analogue" DAC7.
Back in the early 90's, there was some form of collaboration between AKM and Crystal semiconductors so I wonder if the spirit of the CS4328 remains in the heart of the original AKM design team (if they are still even around today)
In answer to your question, I suspect a WELL design R2R DAC will still sonically outperform the AK4498 combination.. however the point is that the AK4498 sets a new standard in (atleast one that's in production “today”) “off the shelf” DAC IC solution / chipset...
Last edited:
Not forgetting that a little later they took this a stage further with the introduction of TDA1307 (mid 1990s if I recall). Did that ever make it into any commercial DACs? I don't recall ever seeing one.
Yes, I used the TDA1307 in a few designs such as one of the later digital filter / modulator options for the Pink triangle Dacapo / Ordinal etc.
The SAA7350 sounded better then the TDA1307 as the TDA1307 was sonically limited by its internal non bypassable 8x digital OS filter - I could use the PMD100 / YSF210 etc. combined with the SAA7350 and these are better sounding filters IME (I've still yet to hear a better sounding digital filter (non FPGA based) then the PMD100 (I've not heard the PMD200 yet)).
Last edited:
More than a ilttle ironic given they write in the TDA1307 DS :
The oversampling digital filter in the digital audio
reconstruction system is of paramount influence to the
fidelity of signal reproduction. Not only must the filter
deliver a desired stop-band suppression while sustaining a
certain tolerated pass-band ripple, but it must also be
capable of faithfully reproducing signals of high energy
content, such as signals of high level and frequency,
square wave-type signals and impulse-like signals (all of
these examples have their counterparts in actual music
program material). Filters optimized only towards
pass-band ripple and stop-band suppression are capable
of entering states of overload because of the clustered
energy content of these signals, thus introducing audible
degradations in processing the mentioned types of
excitations. To dimension a high-fidelity digital filter, a
balance must be established between filter steepness and
overload susceptibility.
The oversampling digital filter function in the TDA1307 is
designed, in combination with the noise shaper, to deliver
the highest fidelity in signal reproduction possible.
Yep - indeed!
Philips with Burr brown / NPC (DF1700 etc.) had a particular sonic signature with there digital filter designs - a more Bass heavy sound but lacking L/R sound stage / fine detail - while Sony, Yamaha, PMS PMD100 have a more open / natural sound.
The odd thing is that you would find very diametrically opposed preferences, never a twine would meet - it was either the BB /NPC /Philips sonic signature or the sound of the Yamaha / Sony / PMS PMD100 filters...
Last edited:
Claude,
I've been a member of DiyAudio since 2004, I've just not posted much these recent years as the SNR has degraded so much!!!
John, I was wondering your assessment of the audible affect, if any, of the tiny repeating passband frequency response ripples which are evident in the zoomed response graphs of every IC DAC digital image-rejection filter I can remember seeing. Indicating, time-domain signal echoes being produced within filters designed via the REMEZ algorithm, per Lagadec. Obviously, these ripples are too tiny to be audible as such in amplitude terms. Lagadec's concern was, of course, the potential audible time-domain effects.
Since this is an long ago identified technical issue, and these ripples remain apparent in newly designed IC DAC filters, I assume the industry does not view them with any concern. Or, at least, not with enough concern to incur the complexity/cost of a more sophisticated filter design that does not produce them. Do you hold a particular position/opinion regarding the audible significance of this mechanism?
Last edited:
John, thanks!
Very interesting to read for those of us interested in the subject matter.
thanks...
Thanks for that post John. That clearly explains the limitations to bigger and bigger numbers! I note the data sheet for the AKM 4499 shows that the DAC performs best with DSD 256, and performance actually starts to degrade with DSD 512. I also remember Andreas Koch suggesting that the sweet spot for DSD conversion may really be 256, perhaps he was also referring to the speed of the "gates" or "switches" used in the conversion.
As you are working with a discrete converter, and the AKM 4499 appears to be used as a relatively simple (in direct DSD mode) IC version of a discrete converter, what are your thoughts of discrete vs. IC implementation given your response above. The numbers for the 4499 might suggest there are advantages to doing an IC?
Thanks for that post John. That clearly explains the limitations to bigger and bigger numbers! I note the data sheet for the AKM 4499 shows that the DAC performs best with DSD 256, and performance actually starts to degrade with DSD 512. I also remember Andreas Koch suggesting that the sweet spot for DSD conversion may really be 256, perhaps he was also referring to the speed of the "gates" or "switches" used in the conversion.
As you are working with a discrete converter, and the AKM 4499 appears to be used as a relatively simple (in direct DSD mode) IC version of a discrete converter, what are your thoughts of discrete vs. IC implementation given your response above. The numbers for the 4499 might suggest there are advantages to doing an IC?