Just the opposite - we have to measure at 0 dBFS because there are thousands of CDs which reach this level regularly. And since intersample levels can reach + 3 dBFS both the digital processing before the final DA-stage as the analog electronics after the DA stage must be able to handle +3 dBFS levels. Many DACs prove that they can do this, some can't. Therefore it makes sense to perform also tests with maximum possible intersample values, as Nielsen did.My point us simple - we can not rely on measurements, performed at 0 dB FS. They are too close to the level of clipping
What you seem to miss is that when a sinus of 0 dBFS is calculated (for testing a DAC) this signal has no higher intersample values.
This is plain wrong. Fluctuations of the reference voltage lead to corresponding changes in the analog output voltage (or current) of the DAC. The analog stages afterwards must be able to handle this, otherwise their design is broken.and even a good “passing” unit can give us bad results due to different type of fluctuations. We also can not approximate such results for the same product, because tolerance in components can make it either passing or not.
You have to read the article again. Nielsen proved that DACs tested with 0 dBFS+ signals clipped. The whole article speaks about signals higher than 0 dBFS (hence the +), see the 3rd chapter of the abstract on page 1:Nielsen proved that 7 tested by him players perform really badly being tested at 0 dB FS. Are they all “bad players”? What if the same player will have enough room (again due to tolerance in components, for example) and pass?
This paper examines the sonic consequences when 0dBFS+ signals are reproduced in typical consumer equipment. The performance of a variety of domestic CD players exposed to such signals are presented and evaluated.