I disagree. upsampling is a term used in the studio and is specific to creating a specified sampling rate from another (upwards, downsampling when going down in rate) . Originally would have been to create a new file at a new rate, only more recently can be done on the fly on playback. Oversampling is a term used by engineers in digital audio to mean a sample rate increased by a fixed multiple and doesn't effect the storage format.
You can upsample and then that will oversampled during the conversion process.
They may, on the face of it appear to be the same, but the words are more specific and really used in different contexts.
You can look at dCS literature from the early 2000s to see distinct uses of the terms together for example.
Also oversampling can get you in trouble when trying to play higher rate material out of an old dac chip - a multiple will push it out of its operating frequency, you need upsampling to a specific max rate.
I also disagree that non-OS concerns some purist vision of no digital filters anywhere in the whole chain, from creation to playback. As far as I remember seeing it's first beginnings, it was about avoiding on-chip digital filters of about 20 years ago. Did it stem from Audio Note?
"Non-OS" is more useable than a term that is one that's more specific - hence I guess the re-interpretation of the whole thing to become something different. And you probably do find people aligning themselves with this new interpretation too, further confusing things.
Last edited:
I'd like to make note that the discussion for the past roughly 2 pages has been purely on the semantics of upsampling vs oversampling. Surely we can get past this without having too rigid of definitions?
Whether you do it on-chip, upstream, offline, sinc interpolation, zero padding, etc, it's going from a lower sampling rate to a higher sampling rate and needing filtering. Can we at least agree on that?
Whether you do it on-chip, upstream, offline, sinc interpolation, zero padding, etc, it's going from a lower sampling rate to a higher sampling rate and needing filtering. Can we at least agree on that?
You do know data converters are used outside of audio and there is established engineering terminology that existed well before these were used for audio? Yes, I am and have been very aware that audio marketing people decided to call sample rate conversion to a higher rate "upsampling". Your own definitions aren't even consistent, I can use an ASRC at an integer multiple of the original which is "oversampling" based on your arbitrary definition, by the way.
In any case, it does not matter. A box with an ASRC having output rate set higher than the input sample rate is not non-OS, period, end of story. Substituting ASRC ("upsampling") for a DAC's built in filter does not change anything. You can see this if you look into how the sample rate conversion is performed:
From this Analog Devices AES paper on ASRC - http://www.analog.com/media/en/tech...-articles/5148255032673409856AES2005_ASRC.pdf
"Smith and Gosset [4] presents a design for h(t) based on
the windowed sinc function, and an asynchronous
conversion method which computes runtime FIR
coefficients for the variable filter via linear interpolation
from a precomputed table. The method is designed to
be used for a very wide range of conversion ratios with
a single table, which requires that the FIR coefficient
generation handle a particularly general case. The
design presented in this paper is closely related to the
Smith-Gosset method (though it is derived more like the
Ramstad method), but the derivation will make some
simplifying and/or restricting assumptions that allow the
resulting algorithm to be implemented more efficiently.
The Analog Devices AD1896 hardware sample-rate
converter [11] implements a similar concept to the
Smith-Gosset design, but with higher-order
interpolation of the table lookups to keep the table size
down"
So, the "upsampling" done by the widely used chips like AD1896, SRC4192, etc. Will have the exact same perceived negative side effects that non-OS was created to avoid and might even be worse depending on the implementation.
As I already mentioned before, the first mention of this concept was in 1996 by Ryohei Kusunoki and here is his magazine article:
NOS DAC
Clearly, any box that does "upsampling" does not meet the criteria. This is not hard to understand.
As far as I know all CD players from the early 1980's except those from Philips had non-oversampling DACs. They did have analogue reconstruction filters, of course.
Yeah, they were trying to do the right thing. Not some faux zen "the ear is the reconstruction filter" nonsense.
Key word is trying 😉. I'm guessing it wasn't all that hot if you removed the analog filter either.
Definitions aid thinking and discussion, especially when all you have is words. Fuzzy definitions aid fuzzy thinking, whether accidental (e.g. newbies trying to understand) or deliberate (e.g. marketeers trying to ensure that their customers do not understand).DPH said:
If Feature X is deemed by some to be a Bad Thing, then some people will seek to redefine words so that Feature X is semantically removed from equipment even though in reality the electrical behaviour still uses Feature X. In the analogue world we have degeneration redefined by some as 'not feedback', so equipment full of emitter followers etc. can proudly declare itself to be 'feedback free'.
Here we seem to have DACs using massive oversampling declaring themselves to be able to implement 'NOS', when at best they can merely roughly simulate the output of a genuine NOS DAC. To me, 'NOS' doesn't necessarily mean no filters but in practice a lot of NOS DACs do omit the reconstruction filter and it could be this which people like. Perhaps filterless NOS should be called 'image-enhanced sound'?
That sounds better for marketing than treble-reduced sound, which is the other effect of using only a zero-order hold as reconstruction filter.
They are entirely consistant.
The difference is that you appear to be thinking in black / white terms.
Yes, you can call that oversampling if you want to. I have not stated definitions at all. I have provided context for why and when they are used deliberately to differenciate. If you can't relate to those different contexts then perhaps you don't have the experience. That's ok! I don't judge you for it.
So in the context of changing sample rate by a multiple, whether you'd call it oversampling or upsampling would depend on what you wanted to express - that you were doing something from a digital theory / digital electrical engineering point of view or whether you were doing something for a practical, studio-based sample rate change point of view (e.g. getting lots of files to be at the same sample rate despite originating from different sources at different rates). I can't immediately think why this would happen in the context of the studio though, it wouldn't make sense... therefore it wouldn't get called oversampling.
It is nothing to do with marketing - no need of conspiracy theories. It is practical, communication about music making in the studio! No conspiracy there. It has been like that probably for as long as there has been digital audio. The term is then used by equipment manufacturers to communicate to a certain audience using their language.
I'm sorry if you don't wish to learn a new thing. You can stick to engineering terms if you want. Depends if you wish to understand others or not...
There aren't criteria set out at all in the article. It is essentially musings on the processes of oversampling used in 1996. Mostly it is about a subjective feeling that digital audio with oversampling and FIR filters doesn't feel as "immediate" as analog and then finding a potential reason for it.
Despite the article, this link suggests that the feeling that oversampling removed certain qualities of the music were around before Mr. Kusunoki wrote his articles. That would suggest Mr. Kusunoki's reasoning should not be considered to be setting criteria for the practice anyway - these were just his thoughts.
Re: Non-oversampling DAC concept - Peter Qvortrup - Digital Drive
Last edited:
The article defines what the perceived problems are. Just because you don't understand it and how it relates to the filtering doesn't mean they aren't.
Sorry, you just don't know what you are talking about here.
Why wouldn't you? Wouldnt you want to hear that that effect sounded like considering it was so well known, to actually experience it?
Some have and feel that the negative effect is completely overstated and the benefits are better than the negative effects of removing it. That's just called experimentation, play, creativity if you will. It's a good feeling to not be bound by rules telling you you shouldn't do something because it's not going to work... try it anyway and see for oneself and you'll gain a deeper understanding that those who are just happy to quote the "rule" and you might also discover something new and interesting at the same time.
Last edited:
Here you go, from your cited Audio Note designer:
ANK Audiokits - Why is AudioNote's 1X oversampling unique?
When a set of samples is passed through a digital filter, what you get out won’t be an interpolated superset of the input samples, which is the fundamental premise of the whole technology, they will be an entirely new set of samples.
Therefore philosophically there is something wrong with digital filtering and this is proven in practical listening tests. Hence we do not oversample the input signal or digitally tamper with it at all. Hence 1x oversampling (not oversampled). And this also precludes the use of bitstream and delta sigma DACs, which rely on processing. Therefore we only use resistor ladder type DACs.
So, now both Audio Note and 47Labs, the companies that pioneered this (stupid) philosophy, both agree. It is crystal clear that "upsampling" violates the entire reason it exists. Audio Note goes as far to say that delta-sigma converters violate the principle alone, which I suggested several pages ago. There really is nothing more to discuss on this topic.
The article sets out his thoughts on what his perceived problems are... yes exactly what I said. So you agree he does not set out any criteria for others to follow.
He did not set out criteria for a movement or propose the conditions for everything called a non-OS DAC (a title is not part of a criteria). It is up to each individual to set out their own criteria and their own rules. If that means they are avoiding on-chip oversampling and digital filters that is up to them. They are not following a religion full of rules. It's just a vague description open to many people's different interpretations.
Why do you personally want to set a definition for non-OS? Does vagueness feel wrong to you, too unpredictable? Sounds like something you don't understand..
All the "just because you don't understand" is childish and doesn't belong here - it comes across as if you try to find ways to look down on people you have disagreements with! 😱
Gotcha, you just want to be able to call whatever DAC you want non-oversampling whether it is oversampling or not. This is a technical forum, not the Audio Asylum. Maybe you would feel more comfortable there.
- Status
- Not open for further replies.