Media Server Reviews

I examined the amount of timing uncertainty—jitter—in the Grimm Audio MU1's AES/EBU output using my Audio Precision SYS2722 system's digital oscilloscope function. I overlaid successive snapshots of the MU1's AES/EBU output, taken over a 60-second time window, to show what's called an "eye pattern." To make things as difficult as possible for the MU1, the AES/EBU link was a 45' length of Canare 110 ohm balanced interconnect and the data represented the 24-bit Miller-Dunn J-Test signal sampled at 44.1kHz. With an ideal transmission system, all the pulse transitions in the datastream will overlay one another to produce an image of a wide-open "eye," with just one trace visible.

Fig.1 was taken with the original 44.1kHz data (no upsampling) plotted over one "unit cycle." The eye is wide open, with no blurring of the leading and trailing edges. The average jitter level, assessed with a 50Hz–100kHz bandwidth, was 291.3 picoseconds (ps). Figs.2 and 3 show the eye pattern with 2× and 4× upsampling to 88.2kHz and 176.4kHz, respectively. Other than the reduction in the width of the eye, there is still no blurring, though the very long interconnect lengthens the waveform's risetimes. The average jitter levels were 291.3ps at 88.2kHz and 340.3ps at 176.4kHz. While these jitter figures are higher than Grimm's specified <0.6ps, the latter refers to the jitter on the MU1's master clock, while my measurements refer to the data-related jitter due to the transmission of the J-Test signal via the very long AES/EBU cable. For comparison, playing data representing a 24-bit 1kHz tone sampled at 88.2kHz reduced the jitter to 192ps, while the optical output of the Astell&Kern portable player I used for some of my auditioning had 1119ps of jitter.

Analyzed in the digital domain, with no D/A conversion, the Grimm MU1's "Pure Nyquist" filter used for the upsampling appears to be a very long linear-phase type, with a lot of pre- and postringing on either side of the single full-scale sample (fig.4).

While the upsampling doesn't add any data above the original sample rate's Nyquist frequency—22.05kHz with 44.1kHz data—this filter has an extremely steep rolloff, reaching full stop-band attenuation at 22.05kHz (fig.5). For reference, the red and blue traces in fig.6 show the ultrasonic rolloff of the MBL N31's Fast reconstruction filter with the same signal and plotted over the same octave between 15kHz and 30kHz. (Note that this spectrum was taken from the MBL's analog outputs; the ultrasonic noise floor is therefore higher than it is in fig.5, which shows the spectrum of the data analyzed in the digital domain.) The green and gray traces show the rolloff with the N31's Min (minimum phase) filter. The paradox is that I preferred this filter's slow rolloff with 44.1kHz data compared with that of the MBL's Fast filter, yet I preferred the upsampled MU1's ultrafast rolloff shown in fig.5.

Finally, I examined the question of whether the MU1's digital-domain volume control lost resolution at settings below the maximum. Fig.7 shows the FFT-derived spectrum of data representing a 24-bit 1kHz tone as transmitted via AES/EBU with the volume control set to “0” (red), "–10" (blue), and "–20" (green). Reducing the volume by 10dB lowers the 24-bit noise floor by 10dB; reducing the volume by 20dB lowers the noise floor by 20dB. The Grimm MU1's volume control does indeed preserve resolution at lower settings.—John Atkinson