dCS 972 D/D converter Measurements part 2
Fig.5 dCS 972, spectrum, DC-22.05kHz, 1kHz at -0.1dBFS, 16-bit word length increased to 24 bits with triangular PDF dither (linear frequency scale, 22.5dB/vertical div.).
This does not mean that the 972 is not doing anything. The bottom eight bits are now active, with dither noise if not audio-related data, and I am told that D/A converters will usually behave in a more linear manner when fed this kind of datastream.
My next tests are not of relevance to the 972's ability in a domestic system to extract the maximum quality from CDs. However, some controversy surrounded the introduction of a 96kHz sample rate, due to the apparent mathematical difficulty in downsampling that rate to 44.1kHz. In the March 1997 issue of Stereophile (Vol.20 No.3, p.3), Robert Harley reported some experiments performed by James Moorer of Sonic Solutions that indicated that the 96/44.21 conversion was potentially damaging to audio. Steve Lee of Canorus responded in the May 1997 Stereophile (p.221), showing some spectra produced by a dCS 972 downsampling a 96kHz datastream to 44.1kHz. These spectra were free from the spuriae postulated by Dr. Moorer.
I investigated the 972's downsampling ability by using the 96kHz-sampled 1kHz tone at -0.1dBFS on Chesky's Super Audio Collection & Professional Test Disc DVD (CHDVD171). Interestingly, though the traycard proclaims this a "96/24 Super Audio Disc," the 972's bit-activity display indicated only 20-bit data words, at least as output by the California Audio Labs CL20's AES/EBU data jack. (California Audio Labs later confirmed that the C20 does indeed truncate the 24-bit data to 20 bits.) Fig.6 shows the spectrum of this tone, downsampled from 96kHz to 48kHz by the dCS 972 and analyzed in the digital domain by the DScope. Again, triangular PDF dither was used. You can see both that the noise floor appears random, and that the absolute level of the noise components is close to, if a little higher than, those in the 20-bit spectrum shown in fig.3.
Fig.6 dCS 972, spectrum, DC-24kHz, 1kHz at -0.1dBFS, 20-bit word length, downsampled from 96kHz to 48kHz (linear frequency scale, 22.5dB/vertical div.).
Changing the 972's output frequency to 44.1kHz and repeating the analysis gave the spectrum shown in fig.7. The noise floor no longer looks random; some discrete components are apparent. However, the overall level of these spuriae is still very low.
Fig.7 dCS 972, spectrum, DC-22.05kHz, 1kHz at -0.1dBFS, 24-bit word length, downsampled from 96kHz to 44.1kHz (linear frequency scale, 22.5dB/vertical div.).