If the latency is say 1 second, meaning that the music starts to play 1 sec after you press 'play', would that be a problem?
Jan
In fact, the latency is just 40mS, from the horse's mouth : http://www.diyaudio.com/forums/digital-line-level/259397-new-chord-hugo-dac.html#post3998035
For a radio or TV studio that work live, yes, can be a problem.
But for home use, I do not see any problem.....
Oh dear! Something like those plots get trotted out every few months by NOS fans. What they clearly show is that the person putting up those plots does not understand sampling or filtering. The 'sharp' NOS square wave is full of ultrasonic image energy, which ought to be filtered away. The 'ringing' OS square wave has no such problems, even though by eye it looks much worse to the uninitiated. I can only repeat the advice you have been given: go and read a good book.Kastor L said:
I think he was referring to a mathematical Dirac pulse (infinite height, infinitesimal width), not a brand name.
If you want a good filter then you must have high latency - causality requires it. Nobody has yet managed to make an acausal CD player, although I am sure there are 'high-end' people working on it. High latency does not harm: it just means you have to start your CD a few 10's of ms earlier than you want to hear the music.
I found a DAC which would answer most of my unresolved questions, so I'm sharing it here.
I realise some people here think they are "beyond this", which is fine, but please accept that a lot of people need to listen for themselves before they can believe, that's just the nature of belief.
Scientific papers don't have any kind of total insight, imho, for instance casual looking null results at times, such as no one could differentiate MP3 versus Flac.
Wavedream DAC | Rockna Audio
It's R2R and has selectable Nos mode, IIR mode and FIR mode.
It uses a "2000 tap" filter length, so that should, if the extra length is valid, sound much more accurate or realistic than Nos in theory.
A few hours of blind listening should reveal a lot.
No idea what the price is, but it's the only product I found so far with R2R, Nos and IIR all in one.
As for the latency in a long FIR, I think the Hugo is actually around 80ms, after USB protocol running in ASIO that should be around 100ms.
100ms doesn't have any effect on music in solitary, but as soon as we introduce audio/visual sync / interaction, audio/audio interaction or physical interaction, like an instrument connected to a computer, then 100ms is by far and wide useless.
It may perform fine as a music DAC but in any other application it's unacceptable, that was the context of my comment.
"I think he was referring to a mathematical Dirac pulse (infinite height, infinitesimal width), not a brand name."
Dirac HD heavily revolves around impulse response, it's related to the Dirac function, it seems like a theoretical solution for time error in transducers.
I realise some people here think they are "beyond this", which is fine, but please accept that a lot of people need to listen for themselves before they can believe, that's just the nature of belief.
Scientific papers don't have any kind of total insight, imho, for instance casual looking null results at times, such as no one could differentiate MP3 versus Flac.
Wavedream DAC | Rockna Audio
It's R2R and has selectable Nos mode, IIR mode and FIR mode.
It uses a "2000 tap" filter length, so that should, if the extra length is valid, sound much more accurate or realistic than Nos in theory.
A few hours of blind listening should reveal a lot.
No idea what the price is, but it's the only product I found so far with R2R, Nos and IIR all in one.
As for the latency in a long FIR, I think the Hugo is actually around 80ms, after USB protocol running in ASIO that should be around 100ms.
100ms doesn't have any effect on music in solitary, but as soon as we introduce audio/visual sync / interaction, audio/audio interaction or physical interaction, like an instrument connected to a computer, then 100ms is by far and wide useless.
It may perform fine as a music DAC but in any other application it's unacceptable, that was the context of my comment.
"I think he was referring to a mathematical Dirac pulse (infinite height, infinitesimal width), not a brand name."
Dirac HD heavily revolves around impulse response, it's related to the Dirac function, it seems like a theoretical solution for time error in transducers.
Last edited:
That DAC looks to use OEM modules from MSB to implement the R2R DAC. So they did go 'off the shelf' in choice of DAC - rather negating their marketing puff no?
It is a piece of marketing.
A thermometer, string or R-2R DAC with Nos / IIR / FIR selectable filters, preferably with a super long tap length, then lastly a vanishingly low THD / IMD line-out is all that is needed.
It's not fancy or exotic, it just doesn't seem to exist.
A thermometer, string or R-2R DAC with Nos / IIR / FIR selectable filters, preferably with a super long tap length, then lastly a vanishingly low THD / IMD line-out is all that is needed.
It's not fancy or exotic, it just doesn't seem to exist.
Hidden data can be retrieved to a certain degree :Spline interpolation - Wikipedia, the free encyclopedia
This guy stated it reasoning about the game of roulette :Blaise Pascal - Wikipedia, the free encyclopedia
He also invented the syringe and studied air preasure.
He also invented the syringe and studied air preasure.
Another frenshman did contributions :Bézier curve - Wikipedia, the free encyclopedia
My wife has a curve lineal with Be´zier curves to cut clothes.
She is a fassion designer.
Reminds me on High End Audio somewhat.
We are easily mesmerized by illusions.
My wife has a curve lineal with Be´zier curves to cut clothes.
She is a fassion designer.
Reminds me on High End Audio somewhat.
We are easily mesmerized by illusions.
No. Splines don't add data, they just please the eye. There is no reason to assume that a spline curve interpolation between sample points can somehow reproduce 'missing' data which was removed by the anti-aliasing filter, however nice it may look.Joachim Gerhard said:
The thing about splines is that each set of points needs a new spline. If you have a large set of points then you need lots of splines; it is equivalent to having a filter whose (digital) coefficients or (analogue) parameters are frequently (and abruptly) changed every few sample points. I would expect this to be audible, and horrible.
No idea about splines.
Secret rabbit code has a lower noise floor than SoX for some reason in at least one measurement I've seen somewhere.
I did inadvertently debunk the Metrum and Phasure measurements via the discussion here, but "super long tap length" is still on the table, not sure how it translates into reality.
In fairness the Metrum is apparently using a string DAC, which apparently has less or zero glitch error.
This glitch error isn't analyzed very much as far as I know.
It seems to correlate with the design of the string / thermometer / R-2R 6-bit section in the PCM1792, PCM1795, PCM1794.
They seem to have lower distortion than similar designs at least and have a linear sound.
ES9018 has proprietary noise shaping to "fix" the error of a normal Sigma-Delta.
Considering PCM1794 and ES9018 it does appear like these designers don't accept thr normal practice Sigma-Delta architecture.
WM8741 and AK4399 have minimum-phase settings.
No idea what this long tap length is though and I haven't seen any measurements, just random comments like "we need 1 million taps".
If true then Nos should sound pretty weak, without any taps, just the raw 44,100 samples per second.
A chip like CS4398 sounds "perfectly fine", this is splitting hairs, just useful to know how to achieve the "most ideal" sound in all respects.
Just my views.
Secret rabbit code has a lower noise floor than SoX for some reason in at least one measurement I've seen somewhere.
I did inadvertently debunk the Metrum and Phasure measurements via the discussion here, but "super long tap length" is still on the table, not sure how it translates into reality.
In fairness the Metrum is apparently using a string DAC, which apparently has less or zero glitch error.
This glitch error isn't analyzed very much as far as I know.
It seems to correlate with the design of the string / thermometer / R-2R 6-bit section in the PCM1792, PCM1795, PCM1794.
They seem to have lower distortion than similar designs at least and have a linear sound.
ES9018 has proprietary noise shaping to "fix" the error of a normal Sigma-Delta.
Considering PCM1794 and ES9018 it does appear like these designers don't accept thr normal practice Sigma-Delta architecture.
WM8741 and AK4399 have minimum-phase settings.
No idea what this long tap length is though and I haven't seen any measurements, just random comments like "we need 1 million taps".
If true then Nos should sound pretty weak, without any taps, just the raw 44,100 samples per second.
A chip like CS4398 sounds "perfectly fine", this is splitting hairs, just useful to know how to achieve the "most ideal" sound in all respects.
Just my views.
This I understand...
... this also..
The last statement I do not understand, so please help me.
I'm under the impression that with moving one (original) sample (x ) to the next (x+1), the calculation of the Spline (which can be calculated over an interval of more than 3 points, ranging from eg. x-2 to x+2 ) follows gradually, not affecting the already calculated samples in the interval (x-1) to (x).
If the Spline is calculated over an interval of 5 or more (original) samples, the deviations from an "Ideal" curve should not be that significant to be audible... I guess. Changes will be gradual and within the spectrum of the upsampled signal, so above the audible limits.
If I'm wrong, please tell me.. (sorry for the crappy English)
Every time you recalculate the spline you, in effect, redesign the reconstruction filter. This could be every few samples. You can't do this 'smoothly', as you only have new data at each sample. In any case, why use a varying spline (designed to look good to the eye) when the correct method is to use a fixed brick-wall filter (designed to recontruct the original signal)?
Some asynchronous sample rate converters use curve fitting to suppress the jitter like this one :http://www.abc-pcb.com/abc_docs/Q5M-DS-110A.pdf
Since it only mathematically reconstructs the signal without pre-echo if the tap length is somewhere in the order of 20,000 to 1,000,000 for 16-bit audio, which induces a lot of latency.
IIR is the correct way to reconstruct the signal.
That's what it says at least.
http://www.cirrus.com/en/pubs/whitePaper/DS668WP1.pdf
Modern TV and computer screens already have high latency, it's horrible, we don't need it in audio equipment as well.
USB has some latency as well, you need to increase the polling rate to 1000 Hz just to get a mouse to move correctly, which means you need to hack Windows somehow, plenty of this discussion on the internet.
I recall the screen latency on recent Sony Vaio notenooks is really out of sync.
Some Sharp televisions try to correct it.
Last edited:
Not sure what you're getting at here. Have another look at the Wavedream link you posted earlier. The linear phase filter has pre-ringing, but the minimum phase filter does not. Both use 2000 taps. That suggests it is the type of filter that determines whether there is pre-ringing, not the number of taps.
- Status
- Not open for further replies.