Upsampling the way to go? ?

Tim is correct ... upsampling refers to increasing the sampling frequency from 44 kHz to a higher number, and using the extra samples to interpolate.

The benefit is that the anti-aliasing rolloff filter is much simpler to implement since it can rolloff more gradually (since the sampling frequency is now much much greater than the maximum frequency recommended ... it's all to do with Nyquist's theory of sampled signals).

At the same time one can also increase the word length of each sample by adding dithered noise. Not extra information .. noise. Dithering is a wierd concept where adding some noise increases the perceived clarity. I'm not so familiar with it ... but as an experiment try looking at a pixelated image and then blur your vision and notice how the image becomes much more recognisable.

Neither upsampling nor adding dither and extra bits can add any information that was not originally present in the recording, neither can they increase the theoretical maximum dynamic range ... instead I think they help to mitigate some of the real-world problems associated with extracting the signal (the roll off filter being the biggest problem.)

Of course a well implemented non-upsampler will outperform a bad upsampler ... probably as much to do with the electronic design, analog output stage, noise from digital bleeding into analog .... it's just good design vs bad design. However since there are plenty of people out there who just have to own the latest buzz words upsampling will sell. Whether a good upsampler will outperform a good non-upsampler ... I really don't know ... probably down to individual taste.

This reminds me of the bitstream vs multibit arguments of about 10 years ago. Both can be done well .. both can be done badly.

All this said (and I have neither heard an upsampler nor SACD) I cannot imagine that upsampling can beat SACD since SACD encodes more of the original information and upsampling appears to "fake it". Perhaps SACD real world implementation has not yet quite caught up with the theoretical possibilities ?

Zaikesman ... I think you're being a bit pedantic. Increasing the sampling rate is the goal, and it is necessary to have some sort of interpolation in order to be able to assign a value to those new samples. One cannot interpolate without increased samples, one cannot increase the samples without interpolation of some sort.

Word length is a completely separate thing.

Using a multiple of 44kHz would seem to me to make the job of the upsampler easiest, especially with linear interpolation, since the original samples are preserved ... but I think a non-multiple could also be made to work, albeit with a bit more complexity.

Of course oscillators are used ... an oscillator is no more than a clock ... so they are saying they have extra clocks within the DAC ... I'm guessing from this that they are stating that they generate their own clocks at the higher sampling frequency instead of deriving them from the (potentially less accurate) digital source signal. Any reclocking DAC does the same.

I try to describe things as clearly as I can ... your response seems to be worded in an deliberately obscure way. Is there are reason for this ?

Actually I think Sean's (the other one) original post was spot on. The format is not the problem .. it's the implementation. That said SACD is a superior format, if not yet a superior implementation.

Zaikesman ... your accusations are quite valid ... I usually don't have much time to do a thorough read. Re-reading the thread we are in close agreement.

Thanks also to dan2112 ... this white paper was very interesting. Perhaps if I get the time and money I'll have a listen to an upsampler. However much SACD is theoretically better I'm certainly not in the mood to replace several hundred CDs, especially since many were (unfortunately) digitally recorded, probably at 44kHz !

http://www.dcsltd.co.uk/papers/aes97ny.pdf