> -sakset fra denne linken:
>
http://www.head-fi.org/forum/thread/280091/upsampling-oversampling-dac-tube-
> amp-best-of-both-worlds
>
>
> Upsampling is when the sampling frequency is raised from 44.1kHz to
> anything
> which is higher, could be 50kHz, 96kHz, or 100MHz. Oversampling is the
> subset of upsampling where the sampling frequency is raised by an
> integer
> multiple, in other words, 88.2kHz, 132.3kHz, etc. This is done so
> that the
> reconstruction and bandpass filters can more accurately put the
> signal back
> together and output it without running into ringing, noise, and phase
> issues.
>
> ...
> Upsampling or re-quantisation can, under very rare circumstances,
> improve a
> recording but only if the original recording has considerable jitter
> and the
> clock controlling your SRC process is considerably more accurate
> than the
> clock on the original ADC used for the recording. All of which is very
> unlikely, it's much more likely that upsampling is going to reduce the
> quality of the CD rather than improving it.
>
> ...
> The question that sschmeichel puts is an extremely valid one. In
> practice,
> upsampling cannot improve the quality of the audio file. However,
> upsampling
> will bypass the 44.1k reconstruction filter in your DAC and instread
> use the
> 96k filter. In some poorer quality DACs this may result in the audio
> sounding better because it's more expensive to create a good
> reconstruction
> filter at 44.1k than at 96k. So yes sschmeichel, under certain
> circumstances
> there can be a perceived improvement. Bare in mind though that with
> higher
> quality DACs upsampling is just as likely to cause a percieved
> degrading of
> the sound quality. This would be a more accurate representation of
> what is
> happening to the audio file. With the exception of bypassing poor
> reconstruction filters, upsampling can only increase quantisation
> errors. So
> for most people here, it is a toss up between introducing more
> errors by
> upsampling vs. a possible improvement from using a different
> reconstruction
> filter.
>
> ...
> I am not suggesting doing away with oversampling. Nor am I doubting
> that
> some people hear an improvement with upsampling. The likely
> explanation for
> the improvement in perceived quality at higher sample rates cites
> implementation limitations of the digital (or analogue) anti-aliasing
> filters within current Ooff-the-shelf¹ converter devices. These filt
> ers have
> traditionally been designed with very sudden high-frequency roll-off
> in
> order to maximise the available audio bandwidth at 44.1k. This
> requirement,
> coupled with practical limitations in filter complexity, have led to
> compromised designs with significant ripple in the passband (and
> sometimes
> inadequate attenuation in the stop-band). In addition, these very
> steep
> filters often result in unwanted temporal distortion: i.e. discrete
> pre- and
> post- echo effects. There is speculation as to why and how these
> might be
> noticeable, but experiments carried out with Ono-compromise¹ filter
> designs
> (at 44.1k) which eliminate these echoes, suggest that their removal
> renders
> the sampled signal subjectively indistinguishable from the original
> analogue. So perhaps very extended sample rates are not strictly
> necessary;
> on the other hand, an inadequate filter operating at extended rates
> produces
> echoes proportionately nearer in time to the actual signal than it
> does at
> lower rates, which are likely to be less noticeable. If this theory
> proves
> to be correct, it will cast doubt on the usefulness of the O4x¹ rate
> s (e.g.
> 176.4kHz or 192kHz). In other words, a higher sample frequency and
> therefore
> a filter with a higher cut-off point may not necessarily be the best
> way to
> go, depending on the quality of the implemented filters.
>
> ...
> The math is simpler and less prone to errors going from 88.2 to 44.1
> (or
> vice versa). There was a time about a decade ago where using high
> quality
> studio equipment you could hear a difference when converting to/from a
> multiple of 44.1k as opposed to say converting between 96k and 44.1.
> I feel
> the situation has improved in the last decade, although my guess is
> that
> using a multiple is still mathematically less prone to errors. The
> question
> is: Are these errors less significant than the benefits that may be
> obtained
> from an easier to implement filter at a higher sample rate. The answer
> obviously depends on the individual DAC and how it well it
> implements it's
> LPF at different sample rates.
>
>
> ...
> If you want the highest fidelity though, you would need to get a DAC
> with a
> good 44.1k reconstruction filter.