When I acquired the AK4137 I was interested in comparing it with SRC4392. For the sake of completeness, I listened to each of the upsampled PCM modes, each one with each of the 4 interpolation filters. I also tried the DSD modes. I wasn't expecting it, but the DSD256 sounded subjectively better to me than any of the other modes, either with filter #1 or #2, but I tend to favor #2 the most, although it still doesn't sound quite right to me. My guess on this issue, and you probably know by now I have been doing a lot of guessing, is related to PCM vs DSD filtering requirements. DSD uses fairly basic filters vs PCM filters which have to transition to stop band much more steeply as compared to more gentle DSD filter roll-off. Near as I can tell, DSD sounds subjectively better to me because the filter sounds better. The dac DSD filter doesn't sound better than DAC-3's external PCM interpolation filter that runs in an FPGA, but the DSD filter in the dac chip sounds better to me than the PCM filters in the dac chip. Until I have an external PCM interpolation filter, DSD256 is the best sounding option I have, at least to my preference and to the preference of other people that have listened to my modded dac. That's all there is to it.
Based on the article I linked to re DAC-3 and the master tapes use to make CDs, it appears that the tails are supposed to be there. Also, the older dacs have more distortion and fewer bits, but they appear to be by nature of the technology less senstive to jitter effects than S-D dacs are. Therefore, I doubt that jitter is doing nearly as much to them and their reproduction of reverb tails as jitter is doing to S-D reverb tails. Another factor is that I can't see any rationale for mixing and mastering a record to have a sub-audible vocal reverb bed. Mix engineers of pop records often go to great lengths to create reverbs that are audible. When I do hear the reverb tails they sound good, and they sound carefully crafted to enhance the music, and not overbearing at all. In the case of the article I linked to the complaint was about missing room reverb in some cathedral or similar room (IIRC). Those places typically have a whole lot of natural reverb. Putting up a pair of stereo mics to capture a live event accurately should capture a fair amount of the room sound present. Normally, nobody would intentionally remove all of it to leave the recording with no natural vocal ambience at all. Anyway, as you can see, I am skeptical that that the reverb tails are supposed to be missing, but before I read the article I linked to I did keep the idea in mind as a possibility.
In addition, the experience with Katana reproducing reverb tails pretty well would seem to correlate with its very low measured jitter, lower than DAC-3.
The adjustments I make to DPLL bandwidth on my modded Q2M dac also correlate in the same direction as Katana. ESS says the lowest stable DPLL bandwidth should be used. Lowering DPLL bandwidth reduces jitter by LP filtering it away (attenuating it actually). The only reason for not always using the lowest DPLL bandwith is because the bandwidth can be too low to be able to track higher levels of jitter without causing tracking to unlock and unpleasant sound artifacts being produced. So, to recap in slightly different language, lower dac chip DPLL bandwith = less jitter and more reverb tails. However, if incoming I2S jitter is too high, DPLL cannot cope with it at low DPLL bandwidth settings, and the jitter will cause loss of lock, a condition that produces unacceptable noise artifacts for listeners. Currently, my modded dac is capable of arbitrarily low DPLL settings without loss of lock, so I am free to experiment and explore the effects of different DPLL bandwidth settings.
Regarding voltage controlled oscillators, most of them that can be swept very much already have a lot more jitter than a crystal, here I am thinking of the Si-549 programmable clocks. Of course, there are a few low phase noise crystal clocks that can be pulled a little using a control voltage. Don't know if one would fit on my dac board. It could take a fair amount of time and effort to rig up a test as well. Sometimes I get tired of working on dacs, but interest eventually has always rekindled at some point. The idea of making an external interpolation filter, and maybe a minimal delay reclocker to be located between dac input and AK4237 output both seem worthwhile. Those are things I would like to make some progress on. For the filter and for some other questions, I have a lot of reading to do that will take some time, and probably a fair amount of time to spend with a computer running Vivado, and maybe some other applications. For the reclocker idea, I need to take some measurements. Other things besides those priorities may have to wait.
Last edited:
Is that correct in general?
It was my understanding of the asrc-process that any jitter effect introduced by the process/IC will lead to different data, hence the recommendation to avoid using several asrc-processes in a row/chain because the effect accumulates.
Of course we are talking about really low levels if everything works as intended.
Btw, afair Putzeys/Grimm were quite sure that even quite low level, low frequency jitter (even in the wander region) led to audible differences. Iirc it was based on anecdotal evidence though.
Pretty sure Bruno considers AD1896 and later parts to be mostly audibly transparent, however I can’t find the relevant text so I’d let him speak for himself.
Rest to Mark:
I don’t know why there is so much talk about ASRC and PLL filter bandwidths when all of this only applies to the Sabre parts.
ASRC isn’t something you should sprinkle everywhere. In an ideal USB DAC you should probably have zero ASRCs. One for an SPDIF input is good too.
It’s a problem solver to bridge clock domains with, not something you sprinkle in to taste.
Rest to Mark:
I don’t know why there is so much talk about ASRC and PLL filter bandwidths when all of this only applies to the Sabre parts.
ASRC isn’t something you should sprinkle everywhere. In an ideal USB DAC you should probably have zero ASRCs. One for an SPDIF input is good too.
It’s a problem solver to bridge clock domains with, not something you sprinkle in to taste.
I wonder if the big reduction in pre and post ringing you have on modern DAC filters compared to jelly bean converters from a few years back might not be the reason.
That said, it really needs to be measured to get a handle on it - if it’s reverb tails I expect that it’s still a lot higher than -90 dBFS.
Sony said for archival purposes, it was because it would be future proof. meaning, any kind of coding could be applied.
-RNM
It's still unclear what you mean by reverb tails, but as I suggested, if it's the tail end of a reverberation, RT60 is easy to measure. So why don't you try?
Can you suggest a piece of "pop" music that has these complex reverb effects?
Apologies - your answer is helpful, but I meant the timing measurements - peak, rms values and what sort of distribution, gaussian, etc.
TDA1387 ?
More on Benchmark/ Border Patrol merrits (I’ll never want to be in a Hi-End constant brain dichotomy state)
BorderPatrol Digital to Analogue Converter SE Page 2 | Stereophile.com
George
Stereophile 2019 recommended components list was published today. I do notice that the Benchmark does sit in a lonely position in terms of price/perceived performance ratio sitting amongst other units at 5x its price or more. It's still more than I am likely to be able to afford or justify for the next decade (I need 2) but you have to take you hat off to them for not entering the blinged up fleecing game.
I'm far to be an expert in digital, but, for what I know, F.M. distortion appears with periodical jitter. And it seems logical.
It's annoying, intellectually. How to remedy something if we can not explain the cause of a phenomenon that we feel ?
Well, personally, I have not been bothered by this phenomena. I use a Berhinger DCX2496 which I modded clocks, power supplies and analog stages.
The result is good enough for me and I should have other parts to improve first in an utopian but fun quest of perfection ;-)
As habit, I found that the mods made on analog stages were the most decisive.
I would add that my attention is more focused on the upper part of the dynamics (the attacks) than the bottom, and that, rarely listening to classical music, I usually listen to quite dense musical messages where the tails of mice are usually crushed by the elephant which follows them very closely ;-)
A remark in the form of a question, for people, like J.C., who, accustomed to analog sources and its weak signal/noise ratio, seems to be spoiled by the low resolution of the digital systems on very low levels, it might be good to keep a little analog noise in its amps and keep the analog threshold higher than the digital noise level ? A pink noise generator that we could adjust in level ?
Sometimes the best is the enemy of the good ;-)
Last edited:
This is from Benchmark web page, sound very strange for me that they said it, specially first one about Feedback.
#
Feedback systems are an important part of most audio devices. Feedback can be used to reduce the distortion produced by an amplifier. Feedback is a recursive process and this makes it subject to stability issues. Feedback loops have a delayed response; some error must occur before the correction process gets started.
Feed-forward correction can replace or supplement a feedback system. Feed-forward correction is non-recursive and this makes it inherently stable. Feed-forward correction remains stable and effective when the output loading changes.
Benchmark amplifiers use a combination of feedback and feed-forward correction. The feed-forward correction amplifier adds a correction signal at the final output of the amplifier.
#
#
Feedback systems are an important part of most audio devices. Feedback can be used to reduce the distortion produced by an amplifier. Feedback is a recursive process and this makes it subject to stability issues. Feedback loops have a delayed response; some error must occur before the correction process gets started.
Feed-forward correction can replace or supplement a feedback system. Feed-forward correction is non-recursive and this makes it inherently stable. Feed-forward correction remains stable and effective when the output loading changes.
Benchmark amplifiers use a combination of feedback and feed-forward correction. The feed-forward correction amplifier adds a correction signal at the final output of the amplifier.
#
I see from your editing you are approaching the vinyl fans club
https://www.diyaudio.com/forums/att...tridge-dynamic-behaviour-45-surface-noise-jpg
(Courtesy of Hans Polak)
George
I see from your editing you are approaching the vinyl fans club
God, after his day of rest, created the digital and saw that it was good.
He appeared on the mountain and invested professor Tournesol with a mission: Convert the faithful of the old religion to the digital.
(I hope that I will not end up nailed on the door of a barn ;-)
Not what I meant an algorithm like the FFT does not lose any accuracy by eliminating computations it is not an "approximation" in that sense. Arbitrary ASRC in closed form is probably computationally too intensive.
It is good to remember the error in a sampled data point is not a scalar but a vector, and an answer that is rounded off could be the wrong amplitude at exactly the correct time or exactly the right answer at slightly the wrong time. Not sure what you mean by different data, a fixed set of data fed through the same algorithm on a processor will give the same answer every time. When you run an FFT on a data set the answer is always the same to all 64 bits. In fact I have done the experiment, most major software uses the same IEEE math library and the answers are all the same to 64 bit on all that I tried (UNIX, PC, Mathematica, Python, any GCC program, etc.).
A fancy way to say feedback goes round and round? Maybe they can explain how a transient analysis on a simulator gives the right answer going forward only once?
"Feedback loops have a delayed response; some error must occur before the correction process gets started."
Absolute nonsense.
People confuse phase difference (ALL reactive circuits - and that includes every amplifier under the sun - have some form of phase shift between input and output - basic law of physics) and delay which is an entirely different thing. I think the 'delay' around an amp is about 15 nano seconds or something.
Such a nice amplifier and they've danced around to make the subjectivists happy. Pity.
Absolute nonsense.
People confuse phase difference (ALL reactive circuits - and that includes every amplifier under the sun - have some form of phase shift between input and output - basic law of physics) and delay which is an entirely different thing. I think the 'delay' around an amp is about 15 nano seconds or something.
Such a nice amplifier and they've danced around to make the subjectivists happy. Pity.
Last edited:
Exactly
The problem with feed-forward and output stages is that you don't know the load or its phase. We built a little feed-forward buffer as a test IC that was for only back-terminated 50 Ohm loads (100 Ohms). It was quite remarkable at about -96dB for an open-loop diamond output.
- Status
- Not open for further replies.