The best CD Player for the money

Muralman - I read Kusunoki's article and have few problems with it. First is that he suddenly jumps to single picoseconds from 173ps by increasing number of bits to 20 and claims that it is not possible. we are not reading 20-bit from CD but 16. With 8x oversampling clock has to be 21.6ps accurate - quite possible. Next he assumes that FIR filters produce different delays for different frequencies - not true since they have even group delays. They delay sound constantly by the same amount of time (think of them as FIFO buffers) - no effect on sound. And most of digital filters have enough bits to avoid errors at 16-bit. He also claims that violating Nyquist causes inaudible upper frequncy repeats - it will cause foldback to low (audible) frequencies. If you output 40kHz signal with 44.1kHz clock you'll get 100Hz differential at full amplitude. In similar fashion 25kHz signal will produce 15kHz ghost at the same amplitude. There is not a lot of amplitude above 20kHz so folded frequencies are not a big problem but still sound is not as clean as it could be.

Kijanki, you are, of course referring to 'aliasing'?
Can these frequencies be interpreted as harmonics? If so, than I can see where they'd be bad juju.

Magfan - As they say "The beauty is in the eye of beerholder" - some people like NOS players while others like me like a lot of oversampling. I have test disk (Sheffield Lab Test CD) where they imitate NOS an OS with the sound of a drum. I thing that OS sounds better but it is strange to present only a drum.

Another interesting test is music recorded with different amount of THD. I cannot tell 0.03% from 0.1% but between 0.1% and 0.3% something strange happens. Both of them are very clean (very clean recording to start with) but 0.1% sounds a little lifeless compare to 0.3%. Now I can imagine why a little bit of THD or other crap might improve sound.

Magfan - I forgot to answer aliasing. In sampling theory, input signal frequencies that exceed the Nyquist frequency are "aliased." That is, they are "folded back" or replicated at other positions in the spectrum above and below the Nyquist frequency. In case of DAC it is easy to see what happens when you make drawing of a few cycles of sine-wave on the paper. Now imagine that you the frequency of the output samples is less than 2x signal frequency (period longer than 1/2 of the signal period). Lets take period of 3/4. Placing first point at zero we'll get second point at 3/4 (-1) next at next 3/4 (0) then +1 etc. When you connect all the points you will get sine-wave with +/-1 amplitude and frequency of 0.33 of the original signal frequency.

Let take signal of 33kHz and DAC update frequency of 44kHz (1.33x). In result DAC will output (instead of 33kHz) 11kHz (0.33x) at full amplitude (33kHz - 44kHz/2). Nyquist says that in order to preserve frequency information we have to either output them at >2x their frequency or to not have them at all (filter them out) otherwise they will fold starting from 0Hz (DAC will output 22.05kHz as 0Hz and 22.1kHz as 50 Hz, 22.2kHz as 150Hz etc.

Muralman - I read Kusunoki's article and have few problems with it.

I don't understand it either. He is mixed up in my way of thinking. He forgets that we use dither to get a lot more out of LSB's (to remove quantization errors). His assumption that the LSB is the error margin in what we hear is completely fallacious. This article seems to go against conventional engineering wisdom. It is either a mishmash of rather odd assumptions or I just can't follow it.