dCS Purcell D/D converter Measurements
There are only three questions to be asked of a digital/digital converter set to upsample "Red Book" CD data: 1) does it add high-frequency information above the Nyquist Frequency (half the sample rate) of the original data? 2) Does it add low-level information below the original word-length limit of the original data? and 3) does it increase or decrease the level of word-clock jitter in the datastream? If it does increase the jitter level, then, regardless of any other effect it has, the sound may well be worse.
As you can read in our February 1999 review of the dCS 972, the answer to the first two questions is "no." The high-frequency limit of CD data upsampled even to 192kHz remains 22.05kHz; the noise floor of CD data that have had their word length increased even to 24 bits remains at the 16-bit level. So the only question I shall concern myself with in this sidebar is the effect of the dCS Purcell on the datastream's word-clock jitter.
As usual, I used the Miller Audio Research Jitter Analyzer to test the Purcell. This is a "virtual instrument," realized with a National Instruments DSP card fitted in a host PC. One of the analog outputs of a CD player or a D/A processor is fed to the card's input, and the Miller software averages sixty four 32-point FFTs on that analog signal to produce a high-resolution spectrum. The device under test is fed an analytical signal developed by British engineer Julian Dunn when he was at PrismSound: a high-level tone at exactly one quarter the sample rate (11.025kHz for CD data), over which has been laid the LSB toggling at 229Hz. The Miller software searches the spectrum for pairs of sidebands around the central tone, then attempts to identify their cause.
As the Miller analyzer needs to be fed an analog audio signal, the effect of a digital device like the Purcell can be assessed only by using it to feed a D/A processor. I used both the sample of the dCS Elgar Plus used by Jonathan Scull in his auditioning and the sample of the 96kHz-capable Musical Fidelity X-24K that I reviewed in February 1999.
Looking first at the results with the dCS Elgar, I established a baseline by looking at the jitter level and spectrum with the Elgar fed an S/PDIF datastream by a PS Lambda transport. The Elgar's analog output level was set to Low and its volume control to "0.0," to avoid clipping the input of the NI DSP card and to eliminate the volume control as an interfering variable.
The result is the black trace in fig.1. The absolute level of the jitter was 179.5 picoseconds peak-peak. While this is very low, it is about 40ps higher than the lowest I have measured. For example, the original sample of the Elgar that I reviewed in July 1997 gave about 135ps of jitter when fed an AES/EBU datastream by a Meridian 500 CD transport. The data-related jitter components, indicated in this graph with red numeric markers, were all at a very low level. The ±229Hz sidebands, for example (red "4"), contributed just 45ps of jitter to the total. The highest-level sidebands lay at ±15Hz (purple "1") and ±500Hz (purple "5"). The spurious tone visible at 9.8kHz seems to be characteristic of the Elgar.
Fig.1 dCS Elgar Plus, high-resolution jitter spectrum of analog output signal (16-bit/44.1kHz data, 11.025kHz at -6dBFS with LSB toggled at 229Hz). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz. Source is PS Lambda CD transport via 6' Apature S/PDIF cable. Grayed-out trace is with dCS Purcell inserted into the datastream, set to upsample to 24/192, and connected to the Elgar via two Madrigal AES/EBU datalinks.