Denon 2910 and 3910 upsample redbook??

No 2910 and 3910 do not up-sample Redbook. 3910 converts Redbook from 16 to 24 bit using 32 bit ADSP. Of course, at least 8 times linear oversampling is done in the Digital filter for CD which is a feature of the actual DAC. Further more the oversampled data is converted to Bitstream inside the DAC too.

When people refer to up-sampling they mean non-linear conversion of 44.1 to 96 or 192 kHz sample rate. As I mentioned in my previous posts, since 96 and 192 can not divide by 44.1 the actual CD (redbook) data is slightly truncated so the up-sampled version is actually worse than the original and this is the reason why the available up-sampling machines does not sound right (it is artificial to me). This is also the reason why the wiser DSP engineers prefer just data word expansion as there is no truncaton using this method and it still helps utilizing the latest 24 bit DACs which results in extracting even the last detail out of Redbook CD.

Regards,
Alex

Most clearly and simply put. Thanks Alex.

Alex,
What available machines out there do upsample and truncate the data?
Thanks.

I have to nit-pick with Alex a bit, although I think he basically has it down. The term "upsampling" applies to any interpolation method that injects additional samples into the 44.1kHz data stream BEFORE DIGITAL FILTERING. This same process is called "oversampling" if it is performed AFTER DIGITAL FILTERING in the DAC itself. Oversampling is always done at integer multiples of 44.1 as the DAC chip itself does not have the computing horsepower to interpolate at non-integer multiples. Upsampling can and is done at non-integer multiples of 44.1 (i.e. 96kHz and 192kHz) because it is usually handled by a dedicated IC which does have the required computing resources. The actual CD data is not truncated at all in non-integer (or any other) interpolation scheme. The original samples are all still there in the upsampled data stream, you've just injected new ones to make what one hopes to be a more accurate reproduction of the analog waveform. The truncation error Alex refers to for non-integer interpolation is inherent in binary floating-point representation of irrational numbers. For example, in upsampling 44.1 to 96 you need to inject 2.1768707.... additional samples per original sample. Of course there is no way to exactly represent the number 96000/44100 in binary. That means, depending upon the number of bits available for the computations, some floating-point computations will be truncated resulting in spacing between the samples that is not EXACTLY 1/96000. Of course, the truncation is generally happening in such a low-order bit that the error should be negligible. Of course since there does exist SOME amount of spacing error in non-integer interpolation many people make the hard-to-refute argument that all interpolation should be done at integer multiples of 44.1 whether it is during upsampling or oversampling. Of course the audibility of integer vs non-integer upsampling schemes (i.e. 88.2kHz vs 96Khz or 176.4kHz vs 192Khz) is certainly an open debate.

Meisterkleef I agree with you. Of course the Digital Filters (stand alone or built in the DAC) do not have the ability to perform non-linear interpolation. You are also right that the data truncation is happening in low-order bits but it's still present and I can hear it. It is most noticeable at the high frequencies which makes me crazy - I can not listen to it. I have tried some of the World's best re-sampling software using VERY sophisticated algorithms to convert 44.1 to 96 or 192 but the result is the same - I hear it. What I found also is that even a linear upsampling to 88.2 and 176.4 is not that good as some dynamics are lost. Playing with the above mentioned computer programs have proven to me that the best you can do to a CD is to convert it to 24 bit and keep the original sampling rate.

Regards,
Alex