We already agree the time window can violate Nyquist unless anti-aliasing is used and a true sine wave at 22050 Hz obviously does not need anti-aliasing.
The ideal anti-aliasing is convolution with the sinc function and current processing power can get very close. The ideal being an infinitely small transition region in both amplitude and phase.
Why not propose a test relevant to these so called audio transients, like say a triangle wack in the middle of 10 sec of silence. BTW many here would be disappointed at how little real high frequency energy is in some of these events. Please make a hypothesis of what we should look for i.e. how do we identify these timing or phase issues. I'm willing to help out but you need to specify the problem exactly. Someone needs to provide the data.
I can do millions of points, etc. easily.
Yes this is pathological behavior and I would look (not listen) and fix it. BTW someone else on ASR has improved some EES DAC card to almost eliminate the hump. The problem does appear to be the processing of differential signals as suspected by some of us.
First of all it helps to remember that the premises of the sampling theorem can't be met in reality, as a signal can't be both, band limited and time limited at the same time.
Further we know, that every processing step introduces some error, some more severe than others, some more pronounced at the beginning of the CD era.
Is it about what can be done with "millions of points" or is it more about what is done (was done the generations before the actual one) in usual devices when replaying music in real time?
And the ESS Sabre hump looks like the green line in the attached picture
Credits for the picture goes to: Article: Understanding Digital Audio Measurements | Audio Science Review (ASR) Forum
Ultima Thule,
Thank you very much for the link, really good geek reading..
I am unhappy that he did not give an explanation as to the hump, but such is life. edit: scott did, thanks.
On one of his graphs, 1Khz test at 48K, he mentions the two spikes between 45 and 50, I would have thought he would say the spikes are 1Khz showing up on both sides of the 48k.
nitty details of no issue of course..
Again, thank you.
Scott,
It's tough to propose the range of tests I would do if I had all my own equipment, I really hate pulling other's strings. the results of tests may inform the next thing to do, and I am not comfortable with that.
The first thing I would do is a 20Khz sine at 44.1. Feed the data into an NRZ system clocked at 44.1 with brickwall, then 88.2 into a 4 pole as Lavry mentioned.
For output check, I would run FFT (which should show nothing, as this test is too simple). Then, a subtractive harmonic analysis (I'm sure everybody was wondering why I even showed that all those bazillion posts ago.
The second wave I would go with is a 5/10/15/20 mix ala Lavry, into a pure NRZ at 44.1, but the subtraction analysis on that would be to very closely examine the phase relationships between the 4 sines. That would be far more consistent w/r to my concerns.
Running higher, say 22K at 44.1, but then oversampling 2x at the dac to get the images out of the way would be useful as the beat occurs every 440 samples,, so if there were a modulation of the output level, it would be obvious. that high a freq would be ok with 2x and images out to 88, but I worry that tossing that into a 44.1 converter without a brickwall in front might introduce unwanted things.
Again, my apologies, I do not have any system that could do this, otherwise I would have..
jn
Last edited:
One of the many links of the last few days showed that it seemed to be the quality of the D/A filtering that was a potential problem, so using a higher rate external converter punts the problem down the line, if an external (higher rate) dac shows there are noticable differences, and levry? found that a well engineered filter made 44.1/16 show differences, then we can start working on what is required of the filter, if a higher rate dac still gets mostly no difference votes we may consider it is not so much of a problem.
Theoretically H2 can be cancelled, at least in part. H3 only in a few specific cases.
I suppose this is based on your listening test. During that test, how high of frequency were you able to hear well enough to evaluate the accuracy?
As you've been called out many times, you cannot avoid the room issues enough to call such experiment an objective comparison.
It may satisfy you personally but such comparison doesn't come close the objectivity needed to produce meaningful results.
So, to resume, Let-us record in 24 bits with the highest frequency as possible, for no need of anti aliasing filters for anything else than electronic noise (passive filters with no phase turns in the audio band), then produce a 24/96 file.
And same thing for the DACs: Over sampling as much as possible, filtering the sampling frequency passively to avoid IM in the analog electronic that follows the DACs.
And same thing for the DACs: Over sampling as much as possible, filtering the sampling frequency passively to avoid IM in the analog electronic that follows the DACs.
OK, then give us the details on how the comparison is set up so that the readers can try it. Thanks in advance.
I don't think anyone is exactly sure what causes the hump, except the manufacturers who have fixed it, eg benchmark, and they are not saying...
As ever, there's a lot of speculation! 🙂
AKG have high performance DACs that don't exhibit this issue though.
My intent was to imply we can now stretch the time limit a lot. At 44.1 I can, for instance, load an entire 2 min. piece of music with ample zero padding at both ends and FFT process it as an entirety. The .0067Hz bin resolution reduces some of the confounding artifacts.
Okay, let me redact and say that we might have gotten to the eventual point about 150 posts earlier if we came in with where we stood. And with a whole lot less frustration involved. I know personally (and am trying to be proactive) that I react very differently, generally more politely and more direct to the thrust of the discussion when someone says, "hey, I'm coming from here".
I see this "purely intellectual and a bit of devil's advocate but I'm not going to out and say it" thing a lot in academic circles, so I wonder if it's more a cultural thing. Goodness knows I've been guilty of it in the past, quite possibly still so.
If anything I'm making a plea for people (and using you as an example) of giving more context, which tends to smooth out the bumps in the road a bit.
As I said some reference designs use 4 resistor plus op-amp differential amplifiers. Someone on ASR fiddled with tweeking feedback capacitors and some resistor values and seriously reduced the hump.
Apologies - missed that. The curious thing is why ESS have not published a fix... Last I heard their own eval board showed the issue. Or perhaps they consider it unimportant.
I read some of the Lavry stuff but could you point me to the one you are talking about now. I had mixed feelings about what I did read.
BTW please remember all the time I spent on the lumped vs. T-line speaker cable stuff. As I remember it was us against everyone else, which reminds me I miss DF96 another who has abandoned us?
- Status
- Not open for further replies.