The Fifth Element #16 John Atkinson August 2003
After John Marks enthused over the $1495 Grace 901 headphone amplifier in March (pp.45-47), he shipped the review sample (serial number 9085) to me. He wanted me to listen to the unit for myself, not only for pleasure, but also so I could decide whether his recommendation of a Class A rating in our April "Recommended Components" listing was appropriate.
Using two pairs of Sennheiser HD600 headphones, one fitted with the stock three-conductor cable, the other with balanced, dual-mono Clou cable, I compared it with my sample of HeadRoom's BlockHead ($3500), purchased following Jonathan Scull's rave review in June 2002.
To cut a long story short, JM was right. The Grace 901 has a rich, full-bodied presentation that fully justifies a Class A rating. Perhaps the balanced BlockHead edged slightly ahead in ultimate resolution, but at a considerable price premium. And the Grace's 96kHz-capable data input makes it a one-box solution for digital playback, eliminating the need for one pair of possibly pricey interconnects.
But echoing JM's enthusiastic encomium was not the prime purpose of this Follow-Up, which was to add some measured data to the magazine's database. For balanced analog signals, the 901 offered a maximum gain of 9.5dB in its Low gain setting, 20.1dB in High, allowing the stepped volume control to be used in its optimal range regardless of the sensitivity of the headphones used. There wasn't an exact unity-gain setting; it lay between the control's 4:00 and 4:30 o'clock positions. Neither set of analog inputs (balanced, pin 2 hot, or unbalanced), nor the digital input, inverted the 901's polarity. The input impedance at 1kHz measured a usefully high 19.7k ohms unbalanced, 39.5k ohms balanced.
The 901's output impedance measured a very low 1.4 ohms across most of the audioband, this rising slightly to 1.85 ohms at 20kHz. (This figure includes the series resistance of the leads connecting the unit to my Audio Precision System One.) The Grace should be able to drive long leads and all headphone models without breaking a sweat. (For the technically minded, its output stage uses an Analog Devices AD815 balanced line driver, capable of sourcing a minimum of 500mA.) The maximum output level into 150 ohms was the same in both High and Low gain conditions: 11.9V at 1% THD (fig.1), which will be more than enough to drive all available headphones to gray matter-compressing volumes.
Fig.1 Grace 901, analog input, THD+noise (%) vs 1kHz output voltage into 150 ohms in Low (top) and High gain conditions.
For analog inputs, the amplifier's frequency response was flat to 200kHz, its channel separation approaching 100dB (neither shown). The A-weighted signal/noise ratio with the input shorted but the volume control at its maximum (ref. 1V output into 150 ohms) was an excellent 93.3dBA in the Low gain condition. This worsened, as expected, by 10dB in High, and by another 20dB when the A-weighting filter was switched out of circuit to give an unweighted measurement bandwidth of 10Hz-500kHz. Distortion was extremely low into the laboratory 100k ohm load, and worsened only very slightly into 150 ohms, at low frequencies and above the audioband (fig.2).
Fig.2 Grace 901, analog input, THD+N (%) vs frequency (from bottom to top at 1kHz) at 3V into 100k ohms and 150 ohms (right channel dashed).
Fig.3 reveals that the highest-level component of that distortion at low frequencies was the second, at -80dB (0.01%); all other harmonics approached or were below -100dB. Intermodulation distortion was also vanishingly low under normal conditions. To generate the spectrum shown in fig.4, where the 1kHz difference component reaches a minuscule -96dB (0.0015%), I had to drive the 901 into 600 ohms—almost into clipping.
Fig.3 Grace 901, analog input, spectrum of 50Hz sinewave, DC-1kHz, at 3V into 150 ohms (linear frequency scale).
Fig.4 Grace 901, analog input, HF intermodulation spectrum, DC-22kHz, 19+20kHz at 12V into 150 ohms (linear frequency scale).
This measurement was taken using 24-bit digital data as the source. The Grace 901's digital circuitry is based on the Crystal CS43122 chip, a 24-bit, 192kHz-capable DAC with a specified dynamic range of 122dB. (The CS43122 uses a delta-sigma architecture with 5-bit internal words to achieve this performance.) With the High gain setting and the volume control at full, the output stage clips at -7.5dBFS for digital input signals. This is academic, considering that, at that point, the listener would be standing several feet from the headphones. With the Low setting, the maximum output level at 0dBFS was 8V RMS.
The frequency response at the 44.1kHz sample rate (not shown) was flat within the audioband, with 0.1dB drops apparent at 10Hz and 20kHz. Surprisingly, the infrasonic response was slightly more curtailed at a 96kHz sample rate (fig.5). The ultrasonic response continues the slight 20kHz droop to reach -0.5dB an octave higher, with then a steep low-pass rolloff evident.
Fig.5 Grace 901, digital input, frequency response at -12dBFS into 150 ohms, 96kHz data (right channel dashed, 0.5dB/vertical div.).
Fig.6 shows my usual spectral analysis of the DAC's noise floor, using a swept 1/3-octave bandpass filter, while the DAC decoded 16- and 24-bit dithered data representing a 1kHz tone at -90dBFS. The increase in bit depth doesn't lower the noise floor by as much as the very best D/As around—such as the Weiss Medea, which Kal Rubinson reviewed in February 2003, or the dCS Elgar Plus, which Michael Fremer reviewed in April. It also unmasks some very slight power-supply noise at 60Hz, 120Hz, and 180Hz. But the 901 is a) a headphone amplifier, and b) its price is a small fraction of those cost-no-object behemoths.
Fig.6 Grace 901, 1/3-octave spectrum of dithered 1kHz tone at -90dBFS, with noise and spuriae, 16-bit data (top), 24-bit data (bottom). (Right channel dashed.)
The 901's linearity error (measured with dithered 16-bit data representing a 500Hz tone) was vanishingly small to below -110dBFS (fig.7), and its reproduction of an undithered 1kHz tone at -90.31dBFS was essentially perfect, whether it was 16-bit data (fig.8) or 24-bit data (fig.9).
Fig.7 Grace 901, departure from linearity, 16-bit data (2dB/vertical div., right channel dashed).
Fig.8 Grace 901, waveform of undithered 1kHz sinewave at -90.31dBFS, 16-bit data.
Fig.9 Grace 901, waveform of undithered 1kHz sinewave at -90.31dBFS, 24-bit data.
I assessed the Grace 901's immunity to word-clock jitter with the Miller Audio Research Jitter Analyzer. Fig.10, a narrowband spectral analysis of the 901's analog noise floor while it reproduced a high-level tone at 1.025kHz, over which had been laid a 229Hz squarewave at the 1 LSB level, shows that the only sidebands of note were those at ±229Hz (red "3" markers), and that even these were low in level. The jitter level was a very low 190 picoseconds peak-peak. This figure was obtained with the 901 fed 44.1kHz data from a PS Audio Lambda CD transport via an S/PDIF link. Using a TosLink optical cable increased the jitter to 240ps, which is still very low, implying that the Grace's data-receiver circuit has good jitter immunity.
Fig.10 Grace 901, high-resolution jitter spectrum of analog output signal (11.025kHz at -6dBFS, sampled at 44.1kHz with LSB toggled at 229Hz). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
Like its sound quality, the Grace 901's measured performance belies its $1495 price. Nice. Very nice.—John Atkinson