Good
Ok sample-and-hold DAC is the right English name , sorry for my bad name...
The point is the good sound of certain Nos and what is need on oversampling technique
to do the same musicality 🙂 .
Otherwise others digital or analog elaborations
Ok sample-and-hold DAC is the right English name , sorry for my bad name...
The point is the good sound of certain Nos and what is need on oversampling technique
to do the same musicality 🙂 .
Otherwise others digital or analog elaborations
Its my view that no particular 'oversampling technique' can give the same musicality as NOS because oversampling results in poorer performance from the DAC due to an increased proportion of time spent settling. In other words, the faster the DAC is clocked the greater the proportion of its time is going to be spent at the wrong value.
Open Brain
I agree to your view , but way don't consider new solutions ?
Dither or sine to create the right median value on nos or others things , is a good or bad new way 🙂
I agree to your view , but way don't consider new solutions ?
Dither or sine to create the right median value on nos or others things , is a good or bad new way 🙂
I'm not sure I see the need for any new solutions. If NOS does the job best (subjectively) then there's no need for oversampling.
Of course if my hypothesis is correct then the way to improve on NOS is to get a DAC with the fastest possible settling time. In an active system where the (say) bass/mid drive unit needs to handle only up to 3kHz it might be possible to run a NOS DAC feeding this channel with undersampling (relative to 44k1). These are a couple of experiments I've had in mind for a while.
Of course if my hypothesis is correct then the way to improve on NOS is to get a DAC with the fastest possible settling time. In an active system where the (say) bass/mid drive unit needs to handle only up to 3kHz it might be possible to run a NOS DAC feeding this channel with undersampling (relative to 44k1). These are a couple of experiments I've had in mind for a while.
Settling time change the right median value of the sample ..
NOS Dac have the best fit on this ....🙂
NOS Dac have the best fit on this ....🙂
so made a very quick study of the material. As i though the book camp has VASTLY oversimplified this subject into " this is the way it is "
1) yes if a signal is effectively band limited , then the Nyquist criteria CAN apply, unfortunately
"One last consideration is our assumption of band-limited signals. Mathematically, a signal can never be truly band-limited. A law of Fourier transformations says that if a signal is finite in time, its spectrum extends to infinite frequency, and if its bandwidth is finite, its duration is infinite in time. Clearly we cannot have a time-domain signal of infinite duration, so we can never have a truly band-limited signal. Most practical signals, however, concentrate most of their energy in a definite portion of the spectrum. The analysis above is effective for such signals."
So in most practical applications one is only bandlimiting within bandpass set of frequencies , NOT baseband , far from ideal when reconstructing, with a filter. Suddenly your maths not nearly so deterministic since you are choosing which part of the spectrum to concentrate on.
2) Analog audio signals are VERY rich in terms of spectral density, in cymble sounds ( and others ) there is a considerable amount of discontinuous signal going on ( non integrateable ) . At which point Fourier and Laplace simply fall over , so whats happening to bits of discontinuous signal propagated through an anti-alias filter ? Its not simple maths, and there is no "text book" to refer to since its still being researched on the maths level, let alone when it come to hardware.
3) Ok showing a "trace" is misleading, and am sorry, ( unless the fundamental frequency is so very clear that there are few if any harmonics )
4) Analog propagation of sound from source through air to our ears works. We "hear " stuff, no digital filters anywhere. Since all propagation is in fact discrete ( through air or anything else ) why do we hear correctly , why are our brains not confused by masses of spectral images ? Think about that one for a bit, its proof that nyquist criteria is only accurate in theory. The real world demands MUCH higher sample rates, and the structure of our brains reflects this.
I can and will go on.
As for my idea, its just an idea, and i have not presented it correctly, pretty much none of you understand what i mean, and thats down to me and not presenting my thought correctly.
Finally , yes i was insulted, i do not " chip away at the legs of giants " , that statement and others in the similar vein are designed to make me feel small, designed to grind a opinion into the dust. Not once have i heard "in my humble opinion" , seems to be very little humble here.
If one can not even admit the possibility of being incorrect ( like i have done countless times in this post ) then there is something wrong with that person.
1) yes if a signal is effectively band limited , then the Nyquist criteria CAN apply, unfortunately
"One last consideration is our assumption of band-limited signals. Mathematically, a signal can never be truly band-limited. A law of Fourier transformations says that if a signal is finite in time, its spectrum extends to infinite frequency, and if its bandwidth is finite, its duration is infinite in time. Clearly we cannot have a time-domain signal of infinite duration, so we can never have a truly band-limited signal. Most practical signals, however, concentrate most of their energy in a definite portion of the spectrum. The analysis above is effective for such signals."
So in most practical applications one is only bandlimiting within bandpass set of frequencies , NOT baseband , far from ideal when reconstructing, with a filter. Suddenly your maths not nearly so deterministic since you are choosing which part of the spectrum to concentrate on.
2) Analog audio signals are VERY rich in terms of spectral density, in cymble sounds ( and others ) there is a considerable amount of discontinuous signal going on ( non integrateable ) . At which point Fourier and Laplace simply fall over , so whats happening to bits of discontinuous signal propagated through an anti-alias filter ? Its not simple maths, and there is no "text book" to refer to since its still being researched on the maths level, let alone when it come to hardware.
3) Ok showing a "trace" is misleading, and am sorry, ( unless the fundamental frequency is so very clear that there are few if any harmonics )
4) Analog propagation of sound from source through air to our ears works. We "hear " stuff, no digital filters anywhere. Since all propagation is in fact discrete ( through air or anything else ) why do we hear correctly , why are our brains not confused by masses of spectral images ? Think about that one for a bit, its proof that nyquist criteria is only accurate in theory. The real world demands MUCH higher sample rates, and the structure of our brains reflects this.
I can and will go on.
As for my idea, its just an idea, and i have not presented it correctly, pretty much none of you understand what i mean, and thats down to me and not presenting my thought correctly.
Finally , yes i was insulted, i do not " chip away at the legs of giants " , that statement and others in the similar vein are designed to make me feel small, designed to grind a opinion into the dust. Not once have i heard "in my humble opinion" , seems to be very little humble here.
If one can not even admit the possibility of being incorrect ( like i have done countless times in this post ) then there is something wrong with that person.
I still cannot understand what you're saying - it's not English. And would you please finally answer where you pulled those 16 kHz and 4 kHz numbers from?
A NOS DAC just outputs a stairstep waveform. If that is subjectively 'musical' to you then great.
Could you provide any measurements or data on this?
Because in video we have high speed DACs that settle within ns, but for low speed audio signals we have to revert to ancient technology?
Also, oversampling creates a smoother low-pass filtered waveform which should be easier to follow than a pure stairstep curve with large instantaneous transitions.
And if you want to look at time spent at the wrong value then look no further than a NOS DAC. It spends by far the most time at the wrong value compared to the input waveform.
I have none to offer. You could acquaint yourself with a typical ADI TXDAC datasheet if you doubt that DACs running at faster update speeds perform worse (fewer ENOBs).
I'm aware of video but its not as demanding as cellular baseband applications in my understanding. You lost me with 'ancient technology' - cellular baseband DACs tend to use similar technology to (say) the Philips/NXP DACs i.e. segmented current sources, though of course with finer pitch CMOS processes and consequently faster update speeds (into GHz).
You lost me again with 'easier to follow'. Who or what is needed to 'follow' a DAC's output?
If you're looking at the stair-step waveform then you're looking in the wrong place as that's only an intermediate step. Look at the post-filter waveform then compare that with the input.
Last edited:
1) Is irrelevant since we quantize the bandlimited signal. As such, as long as you attenuate enough all aliasing will be in or under the quantization noise floor, let alone other higher noise floors down the component chain.
Even with infinitely high dynamic range you wouldn't have a point, because then we could also make use of infinite time.
So the theorem is right.
2) If you are saying that the math breaks down for cymbal sounds then that's simply wrong.
A low-pass filter doesn't care which instrument or how dense the spectrum is ... it simply attenuates high frequencies as designed such that it meets your requirements.
It is trivial to show how a low-pass filter reacts to a "discontinuity".
3) Your images are fine for what they are, but adding a sine doesn't remove all the images.
You mentioned discontinuities in 2). Well, that's what you got here. And Fourier will tell what that means in the frequency domain.
4) First of all, filters are hidden everywhere. The air filters. Your head filters, as do your ears. If you look at an HRTF you should see how messy sound arrives at our earcanals, which again are resonant filters ...
Once at the ear drum, the inner ear continues filtering. Most people have trouble hearing much higher than 16 kHz for example. To put it very simple, nerve cells that are sensitive to narrow ranges of frequencies will decompose the signal similar to a spectrum analyzer.
Only after this plethora of filters will the auditory nerve transport the signal to our brain, where filtering continues...
Our hearing and brain is not overwhelmed because it filters, decomposes the signal and immensely reduces the amount of information it finally receives.
Biases and expectation can actually make our brain ignore the actual sound and instead make us hear things that are not even there.
I have a high frequency impulse test file (24 bit 96 kHz) here: imp_urhp24.flac
I was asking specifically about the settling time and audio DACs, but alright.
Those are converters that go into the GHz range. Of course you cannot speed everything up arbitrarily, and even the general statement that faster speeds can reduce performance is not wrong but kinda misleading.
If the higher speed DAC outputs a waveform that resembles the input more closely than a NOS DAC then what does it matter?
If you love the roll-off and phase shift below 20 kHz then you could also do that zero-order hold oversampling and slow low pass filtering in software and probably even get cleaner results that way.
Exactly, so your argument that the DAC itself spends most of the time at the wrong value doesn't hold.
Besides, even if it would it is kinda irrelevant as well. If you added a high frequency sine wave inside the DAC, as suggested by vecna, the output would constantly oscillate around the already off value ... and you still probably wouldn't hear a difference.
I wouldn't even mind if the DAC spent 100% of the time at the wrong value, as long as there's a filter after the converter that cleans up the mess and outputs the desired signal.
Last edited:
Read post 60 for explain and sorry for my wrong English
Video application have different problems and need different approach.
Apples are not pears... do you understand this ?🙂
So you don't want to answer why you chose those numbers, so 16 and 4 kHz are completely arbitrary? Then what you originally said makes no sense.
In this spirit I could also say that "with most speakers you can't go above 12 kHz off axis", period. Pure nonsense.
Right, video has much higher speed requirements and not higher bit requirements (but sadly for many of today's popular songs even 12 bits would be enough). In audio we're going slower and easily achieve performance beyond 16, even 20 bits.
Also, my point was on settling time which you missed completely.
It's these blanket statements that I really dislike because they are superficial and misleading, especially to those who do not have the knowledge. Do you understand this?
And here is another example of exactly that:
Right, but there is a big difference in the spectrum of the output of a NOS DAC and a proper oversampling one, especially before final analog low pass filtering.
Do you understand why we oversample in the first place?
Last edited:
If you are human and not god with an analog filter 6 or 8th order is hard to exctract
over 16khz on nos dac .
The oversampling is a digital filtering interpolation but you need a good analog filter
The filter needs working at very high frequency because the output spectrum of
oversampling dac is too large , and you need a very good opamp .
The Nos have a good spectrum response at impulse and at steps input..
Was happen when you have a step on oversampling you have ringing and
probably the resonant response of output filter if you have a 4th order analog filter ....🙂
over 16khz on nos dac .
The oversampling is a digital filtering interpolation but you need a good analog filter
The filter needs working at very high frequency because the output spectrum of
oversampling dac is too large , and you need a very good opamp .
The Nos have a good spectrum response at impulse and at steps input..
Was happen when you have a step on oversampling you have ringing and
probably the resonant response of output filter if you have a 4th order analog filter ....🙂
You referenced post #60, right? Did you understand it? There may be a language barrier.
Even a NOS DAC will happily output over 16 kHz. Nothing changes above 16 kHz except more attenuation.
Look again at the image I posted in #15. Your 16 kHz sampled at 44.1 kHz would correspond to 0.36 Hz on that image.
Do you see how it happily outputs up to the Nyquist frequency (0.5 Hz in that image)?
Exactly because of the digital oversampling you do not need some really steep or complicated analog filter.
No, nothing here is "too large". That the analog filter can operate at high frequencies is a benefit!
Besides, even the NOS DAC outputs high frequency images. These come from the stairsteps, which are much longer for the NOS DAC.
Sorry, but no again.
Both an impulse and a step are already bandlimited to half the sampling rate, so the ringing is in the signal.
The NOS DAC just fails to reproduce it and instead outputs a ~22.7 us long step for a 44.1 kHz impulse. You again are fooled by "looks" of a signal, but the spectrum will show you that it is not the input bandlimited impulse that the NOS DAC is outputting.
It is the NOS DAC that theoretically requires a really steep and more complex analog low pass filter that will cause a big phase shift in the audible range (< 20 kHz).
Last edited:
What is this step you speak of? It is band limited?
Actually a MFB filter going over at maybe 70KHz or so is not that big a deal, doubly so if preceeded by some RC or RLC network to strip off the worst of the MHz and up hash.
If that step is properly band limited when it is created (Hint .....,0,0,0,0,0,0,1,1,1,1... is probaby NOT what you think it is), then it looks just fine, it is only when you create something silly in the digital domain that it looks odd when reconstructed).
Regards, Dan.
Actually a MFB filter going over at maybe 70KHz or so is not that big a deal, doubly so if preceeded by some RC or RLC network to strip off the worst of the MHz and up hash.
If that step is properly band limited when it is created (Hint .....,0,0,0,0,0,0,1,1,1,1... is probaby NOT what you think it is), then it looks just fine, it is only when you create something silly in the digital domain that it looks odd when reconstructed).
Regards, Dan.
Point A
The flat response is around 16khz on nos .. with human analog filter
Point B ans C
You need a working opamp at the spectral replays if you won't some attenuation
and with a very highest open loop gain
And on music signals is normal to find steps around 500lsb 🙂
Last edited:
No, it's not flat at 16 kHz. Did you understand any of the previous posts on this?
On the filters: doesn't the analog low pass only significantly attenuate at higher frequencies? Also, you can build fairly high order analog low pass filters.
They will significantly distort the waveform even far below its cutoff frequency, but it's possible. That's just a limitation of NOS.
No, what attenuates the images is the digital filter which can do away with the imperfections and problems of steep and complex analog filters.
Where does that number come from, what do you mean and why does it matter?
- Status
- Not open for further replies.