The Jitter Game Page 4
Fig.16 Jitter Spectrum, Vimak DS-2000 processing 1kHz sinewave at 0dBFS measured at DAC (linear frequency scale, 0dB = 226.8ns).
Fig.17 Jitter Spectrum, Vimak DS-2000 processing 1kHz sinewave at 0dBFS measured at word-clock test point (linear frequency scale, 0dB = 226.8ns). Even lower jitter than fig.16, and now more random in nature.
Out of curiosity, I replaced the DS-2000's standard Crystal CS8412 "B"-version input receiver with the chip's newer, lower-jitter "C" version, borrowed from an EAD DSP-7000. The jitter was reduced at the DAC to 31ps, down from 34.8ns with the "B" version. The DS-2000's jitter spectrum is shown with the stock "B" chip (fig.16) and with the "C"-version input receiver (fig.18). Note the replacement of the many discrete lines in fig.16 with a 1kHz spike in fig.18.
Fig.18 Jitter Spectrum, Vimak DS-2000 processing 1kHz sinewave at 0dBFS with "C" version of Crystal CS8412 input receiver chip (linear frequency scale, 0dB = 226.8ns). Lower level of jitter than fig.16, but now a discrete, data-related, 1kHz component apparent.
Switching the "B" and "C" chips in the EAD DSP-7000 had a much more profound effect—suggesting that EAD's circuitry is better designed to resist jitter-inducing phenomena after the input receiver. Note that the DSP-7000 features the better "C" chip as standard. With the "C" chip, the DSP-7000's jitter was extremely low (168-210ps). With the "B" version, the jitter increased nearly fivefold, to 896ps. Note that in the Vimak DS-2000 the jitter decreased by about 10% with the better input receiver, but decreased by five times in the DSP-7000 with the "C" chip.
The DSP-7000's plot (with the "C" chip) is shown in fig.19 (worst case at -70dBFS). Overall, the DSP-7000 had very good jitter performance. Note that these measurements were taken with the DSP-7000's internal switch set to 8x-oversampling. The jitter halved at 4x, as would be expected. Here's why: if the jitter amount stays the same and the clock frequency is doubled, the jitter becomes a higher proportion (double) of the signal. This is perhaps one reason why the DSP-7000 sounds better in 4x-mode.
Fig.19 Jitter Spectrum, EAD DSP-7000 processing 1kHz sinewave at -90dBFS (linear frequency scale, 0dB = 226.8ns). Very low jitter level, better below 4kHz than above. Data-related components present at 1kHz and 2kHz, but still low.
The PS Audio UltraLink (reviewed in June 1992) had very low jitter of 139-177ps, this surprising in light of the fact that it uses the most jitter-prone receiver chip, the Yamaha YM3623B. Although the jitter spectrum was smooth when driven at full scale, there was a significant increase in both the number and amplitude of discrete-frequency jitter components as the input signal dropped (worst case was -70dBFS, shown in fig.20). Nevertheless, its jitter performance was excellent and among the best measured.
Fig.20 Jitter Spectrum, PS Audio UltraLink processing 1kHz sinewave at -70dBFS (linear frequency scale, 0dB = 226.8ns). Many discrete components, but overall jitter level quite low.
Because a CD player has no S/PDIF interface between the transport and processor, one would expect it to have low jitter at the DAC. This was indeed the case with the JVC XL-Z1010TN CD player. It had an astonishingly low 51ps of jitter and a very clean spectrum when driven by a full-scale signal. However, the player showed a significant rise in jitter level and an increasing number and amplitude of spikes as the input level was reduced. The 51ps figure increased to 713ps at -90dBFS. These plots are shown in fig.21 (at 0dBFS) and fig.22 (at -90dBFS).
Fig.21 Jitter Spectrum, JVC XL-Z1010 processing 1kHz sinewave at 0dBFS (linear frequency scale, 0dB = 226.8ns). Astonishingly low jitter following word-clock processing by proprietary "K2 Interface."
Fig.22 Jitter Spectrum, JVC XL-Z1010 processing 1kHz sinewave at -90dBFS (linear frequency scale, 0dB = 226.8ns). Much higher level of jitter compared with fig.21, and strong components at 1kHz and its harmonics, despite K2 Interface" processing.
I was eager to measure the XL-Z1010TN (reviewed in April 1990)—it uses a jitter-reduction circuit called the "K2 Interface" (described in my "Industry Update" in Vol.12 No.9). By comparing the jitter measurements before and after the K2 Interface, I could assess the K2's efficacy. Unfortunately, the word clock going into the first stage of K2 is 44.1kHz, a frequency that cannot be analyzed with the LIMD (it is designed for the more common oversampling systems). An analysis of the XL-Z1010TN CD player was instructive, however, in that it showed the effects of LIM alone, rather than LIM combined with the interface-induced, signal-dependent phenomenon described in the Dunn and Hawksford paper referenced in footnote 2.
The VTL Reference processor (reviewed in December 1990), which uses a similar receiver topology and the same Yamaha chip as the UltraLink, had a higher overall jitter level (540-992ps) than the UltraLink, but fewer discrete-frequency components. Also similar to the UltraLink, the VTL had better performance at high input levels (fig.23, 0dBFS), with the worst performance at -70dBFS (fig.24). Incidentally, the PS Audio UltraLink had a nearly identical spectrum at 0dBFS to the VTL at the same amplitude. Fig.23 is thus representative of the UltraLink with a full-scale input signal.
Fig.23 Jitter Spectrum, VTL Reference processing 1kHz sinewave at 0dBFS (linear frequency scale, 0dB = 226.8ns). Low level of jitter, mainly random in nature. Jitter spectrum of PS Audio UltraLink processing 1kHz sinewave at 0dBFS identical.
Fig.24 Jitter Spectrum, VTL Reference processing 1kHz sinewave at -70dBFS (linear frequency scale, 0dB = 226.8ns). Low-level data brings up level of jitter, but still basically random in nature.
With 51 jitter components rising 10dB or more above the overall level, the Sumo Theorem (reviewed in October 1992) had the highest number of periodic components of the units tested. Moreover, these components were high in level, particularly when driven by low levels: -70dBFS in particular (fig.25). This was surprising in light of the Theorem's excellent performance in the listening room. At high input levels, however, the Theorem's jitter spectrum looked substantially better (fig.26). The Theorem's jitter level varied enormously with input level; at full scale I measured 1248ps of jitter, and at -70dB (worst case) I measured 20 times that value (25,920ps). Despite my enthusiasm for the Theorem (I called it the best converter under $1000 in my review), I did criticize its tendency toward a slightly analytical and dry treble, particularly in relation to the smoother but less detailed CAL Sigma—which had fewer discrete-frequency jitter components and a lower RMS jitter level. (Sumo has reportedly reduced the Theorem's jitter.)
Fig.25 Jitter Spectrum, Sumo Theorem processing 1kHz sinewave at -70dBFS (linear frequency scale, 0dB = 226.8ns). High level of jitter with many discrete components.
Fig.26 Jitter Spectrum, Sumo Theorem processing 1kHz sinewave at 0dBFS (linear frequency scale, 0dB = 226.8ns). Increasing data level to 0dBFS drastically reduces the level of jitter.