Chord Chordette Gem D/A processor Measurements
Sidebar 3: Measurements
I carried out the testing of Chord's Chordette Gem mostly with Stereophile's loan sample of the top-of-the-line Audio Precision SYS2722 system (see the January 2008 "As We See It" and www.ap.com); for some tests, I also used my vintage Audio Precision System One Dual Domain and the Miller Audio Research Jitter Analyzer. To test the Gem in USB mode, I used my MacBook running Snow Leopard (OS 10.6) and Bias Peak Pro 6 to play WAV and AIFF test-signal files.
The testing of the Gem in Bluetooth mode was tricky, because of the mode's use of a lossy codec (COder-DECoder) to squeeze the audio data through Bluetooth's "narrow pipe." (A Bluetooth wireless connection appears to be limited to a maximum throughput of around 768kbps, which is not enough to handle uncompressed audio data of CD quality.) I used my iPhone 3G to test the default A2DP codec, and both the iPhone and a first-generation 16GB iPod Touch fitted with the Creative BT-D5 dongle to test the APT-X codec. I used iTunes on the portable devices to play test tones encoded with Apple Lossless, as well as the SignalSuite app from Faber Acoustical to generate digital test signals.
Apple's USB Prober utility identified the Gem as using a USB transceiver from "Burr-Brown from Texas Instruments Japan" operating in isochronous adaptive mode. It also indicated that the data were limited to 16-bit resolution and sample rates of 32, 44.1, and 48kHz, which is what is offered by the Burr-Brown PCM2704 the Gem uses. The Gem's specification of "DAC Sample Frequencies: 44kHz96kHz (sample rate determined by connection)" may be factually correct but could be misleading; though the actual DAC chip might operate up to 96kHz, all USB and Bluetooth data sources with which the Gem can be used will be limited to a maximum rate of 48kHz, due to the hardware issues.
The Gem's output preserved absolute polarity in all modes (ie, was non-inverting), but its maximum output level depends on the source used. When it was fed data via USB, a 1kHz tone at 0dBFS resulted in an analog level of 2.0V RMS. By contrast, the same signal fed to the Gem via Bluetooth from my iPhone gave, as Art Dudley noted, a lower maximum output level of 1.49V. With the iPhone fitted with the APT-X dongle, the maximum level was even lower, at 1.3V (footnote 1). (I first suspected that I had the iPhone's and iPod's volume controls set below the maximum, but these are disabled when the dongle is used.) The Gem's output impedance is specified as 75 ohms, but I measured a higher figure, which varied from 700 ohms at high and middle frequencies to 1655 ohms at 20Hz. This is still low enough to give no problems with typical preamplifier input impedances.
The channel separation (not shown) was 90dB LR and 94dB RL, both less than the specified 100dB. However, the separation improved to 110dB below 200Hz, and was still better than 78dB in both directions at the top of the audioband.
Like its maximum output level, the Gem's frequency response depended on the operating mode. With USB data at 44.1kHz (fig.1, blue and red traces), the response was flat and wide, with 0.25dB limits at 20Hz and 20kHz. The iPhone's default A2DP codec (green, gray traces) rolled off earlier at both extremes: at 0.25dB at 25Hz in the bass and at 3dB at 17kHz in the high treble. The APT-X codec (cyan, magenta traces) rolled off even earlier at low frequencies (0.25dB at 50Hz, 3dB at 12Hz), but extended higher in frequency at full level than with either of the other modes. Both lossy codecs have some slight passband ripple evident in the top octave.
Fig.1 Chord Chordette Gem, frequency response at 12dBFS into 100k ohms with: USB data (left channel blue, right red), A2DP data (left cyan, right magenta), APT-X data (left green, right gray). (0.25dB/vertical div.)
The three operation modes also differed in their handling of low-level signals. The middle traces in fig.2 show the spectrum of the Gem's output as it decoded 16-bit data representing a dithered 1kHz tone at 90dBFS via USB. The traces peak at exactly 90dB, implying minimal linearity error, and the spectrum is dominated by the dither noise used to encode the signal. By contrast, using the A2DP codec in the iPhone with the same data gave the top pair of traces in fig.2. The noise floor is higher, some positive linearity error is visible, and spectral components can be seen at both harmonic and nonharmonic frequencies.
Fig.2 Chord Chordette Gem, 1/3-octave spectrum with noise and spuriae of dithered 16-bit/1kHz tone at 90dBFS with: A2DP data via Bluetooth (top), PCM data via USB (middle), APT-X data via Bluetooth (bottom). (Right channel dashed.)
The bottom pair of traces in this graph show the spectrum when the same test signal was sent to the Gem via the APT-X dongle. All it shows is the background analog noise of the Gem itselfthe APT-X codec appears to go deaf with signals that drop below a threshold of around 65dBFS. To check this behavior, I repeated the test using a high-resolution FFT technique (fig.3). Other than the change in slope in the noise floor, due to the change in measurement technique and the change from a logarithmic to a linear frequency scale, the results were the same. The USB connection gave a single spectral spike at 1kHz with the noise floor that of the recorded dither (red trace), the A2DP codec (blue trace) gave a slight positive linearity error with a picket fence of distortion components, and the APT-X codec (green trace) showed just the analog noise floor, with no spike at 1kHz.
Fig.3 Chord Chordette Gem, FFT-derived spectrum with noise and spuriae of dithered 16-bit/1kHz tone at 90dBFS with: A2DP data via Bluetooth (blue), PCM data via USB (red), APT-X data via Bluetooth (green).
These results were also evident in the plots of linearity error against frequency (fig.4, USB; fig.5, A2DP; fig.6, APT-X). And while both the USB connection and the A2DP mode allowed the Gem to reproduce an undithered tone at exactly 90.31dBFS, with the three voltage levels clearly resolved (fig.7), the APT-X codec again was deaf to this low-level signal (not shown). Data with a depth of 24 bits were truncated to 16 bits, as expected from the Gem's use of the PCM2704 USB chip.
Fig.4 Chord Chordette Gem, USB connection, linearity error, dBr vs dBFS, 16-bit data (2dB/vertical div.)
Fig.5 Chord Chordette Gem, A2DP Bluethooth connection, linearity error, dBr vs dBFS, 16-bit data (2dB/vertical div.)
Fig.6 Chord Chordette Gem, APT-X Bluetooth connection, linearity error, dBr vs dBFS, 16-bit data (2dB/vertical div.)
Fig.7 Chord Chordette Gem, waveform of undithered 1kHz sinewave at 90.31dBFS, 16-bit data via USB (left channel blue, right red).
Why should the APT-X codec behave this way? The codec designer has to decide how best to use the limited bit budget available. I assume that, as it is unlikely for the codec to encounter low-level signals in music without there being higher-level information present, it just ignores those tones when they are on their own. I tested this by sending data representing a dithered 1kHz tone at 80dBFS, this time accompanied by a 19kHz tone at 20dBFS. The spectrum of the Gem's output, fed the data from the APT-X dongle on the iPhone, now does show the low-level tone, but with a slight negative linearity error and an overall rise in the noise floor, to around the equivalent of 12 bits resolution below 15kHz (fig.8).
Fig.8 Chord Chordette Gem, spectrum, DC20kHz, of 1kHz sinewave at 80dBFS and 19kHz sinewave at 20dBFS, APT-X data via Bluetooth (left channel blue, right red; linear frequency scale).
Because of the need to minimize the amount of data, lossy codecs are nonlinear devices, which makes measurement problematic. I examined the behavior of the lossy MP3 and AAC codecs in an article published three years ago, "MP3 vs AAC vs FLAC vs CD." In particular, I tried to reveal the choices made by the codec designers by using a special test signal comprising a series of 500Hz-spaced tones, but with gaps in the spectrum. Fig.9 shows the digital-domain spectrum of that signal, which peaks at 10dBFS, plotted up to 25kHznote that there is no spectral content above half the sample ratewhile fig.10 shows the spectrum of the Chordette Gem's output while it decoded those data via the USB connection. You can see that while some spurious tones do appear in the gaps between the groups of tones, almost all of these lie below 110dBFS, and that every one of the 500Hz-spaced tones is reproduced at the correct level.
Fig.9 Digital-domain spectrum of 16-bit multitone test signal at 10dBFS with frequency gaps (linear frequency scale, 10dB/vertical div.).
Fig.10 Chord Chordette Gem, spectrum of 16-bit multitone signal, PCM data via USB (left channel blue, right red; linear frequency scale, 10dB/vertical div.).
Fig.11 shows the data as transmitted by the iPhone's default A2DP codec and decoded by the Gem. The top group of tones above 15kHz is missing, the codec's designers obviously prioritizing lower-frequency data for the bit budget, given how few listeners will miss that top-octave information. The levels of the lower-frequency tones are almost all reproduced at the correct levels, but the noise floor has risen dramatically, even between the groups of tones, which is where MP3 and AAC at 320kbps behave more correctly (see figs.10 and 11 in the online article). If you look closely, you can see that the noise floor is about 15dB lower below 2.5kHz, which is where speech and music have the highest energy, than it is above that frequency, presumably to maximize sound quality for telephone voice transmission. But this is not high-fidelity performance. As AD wrote, "the music itself was now flat and undramatic . . . with grain audible between and behind the notes, where only dark nothing should have been."
Fig.11 Chord Chordette Gem, spectrum of 16-bit multitone signal, A2DP data via Bluetooth (left channel blue, right red; linear frequency scale, 10dB/vertical div.).
Turning to the Gem's performance with the APT-X codec, fig.12 reveals a different set of priorities on the part of the codec's designers. Not only does this codec preserve the levels of all but the highest-frequency tones, the noise floor is lower across the audioband, and dramatically so below 5kHz. The floor between the groups of tones is also much cleaner than it was with A2DP. I would have expected APT-X to sound better than A2DP, and indeed, AD wrote that "dynamic contrasts gained the most, and there was also better detail resolution with APT-X than without." But the performance is still compromised compared with uncompressed PCM sent to the Gem via USB.
Fig.12 Chord Chordette Gem, spectrum of 16-bit multitone signal, APT-X data via Bluetooth (left channel blue, right red; linear frequency scale, 10dB/vertical div.).
Because of the nonlinear nature of the lossy codecs, I examined the Chordette Gem's distortion signature with USB data. However, I did look at the APT-X codec's behavior with high-level 1kHz tones. Fig.13 shows that while a full-scale tone was accompanied by distortion harmonics, the highest in level lying around 66dB (0.05%) along with a rise in the lower-frequency noise floor (blue trace), these disappeared when the level of the tone was reduced to 10dBFS and the noise floor became uniform with frequency (magenta), though it lies around the 13-bit level. With an uncompressed full-scale tone via USB, all the distortion harmonics dropped below 96dB (0.0015%), though the noise floor looked somewhat granular (fig.14). This graph was taken with the high 100k ohms lab load; dropping the load to 600 ohms brought up the level of the second harmonic to 74dB (0.02%) and the third to 90dB (0.003%), suggesting that the Gem is breaking a sweat with this punishing load, but the higher-order and more audible harmonics remained below 100dB (not shown).
Fig.13 Chord Chordette Gem, spectrum of 1kHz sinewave, DC10kHz, at 0dBFS (left, blue; right, red) and 10dBFS (left, cyan; right, magenta), APT-X data via Bluetooth (linear frequency scale).
Fig.14 Chord Chordette Gem, spectrum of 1kHz sinewave, DC10kHz, at 0dBFS into 100k ohms, 16-bit PCM data via USB (left channel blue, right red; linear frequency scale).
Reproducing the demanding mix of 19 and 20kHz 16-bit tones via USB, the Gem produced a very low level of the 1kHz difference component, and the higher-order intermodulation products remained at or below 80dB (fig.15). But there is a rise in the noise floor around the fundamental tones that implies a small degree of noise modulation, perhaps due to jitter from the isochronous adaptive USB mode.
Fig.15 Chord Chordette Gem, HF intermodulation spectrum, DC24kHz, 19+20kHz at 0dBFS into 100k ohms, 16-bit PCM data via USB (left channel blue, right red; linear frequency scale).
I tested for jitter rejection by feeding the Gem 16-bit/44.1kHz data representing the J-Test signal (a high-level tone at exactly ¼ the sample rate combined with an LSB-level squarewave at 1/192 the sample rate). The results are shown in fig.16, with the APT-X connection shown in green, the A2DP connection in blue, and the USB connection in red. The expected rise in the high-frequency noise floor with APT-X obscures any jitter sidebands, though the central spike of the HF tone is narrow, suggesting that there is no random low-frequency jitter present; the A2DP codec is significantly worse, with massive "shoulders" around the central spike, as well as sideband pairs. Curiously, the same spurious tone at 7.5kHz is present as with APT-X. Perhaps this has something to do with Bluetooth operation.
With USB data (fig.16, red trace), the noise floor still looks somewhat granular, though actual sideband pairs are few in number, lying at ±74Hz and ±229.7Hz. The former is presumably related to USB operation; the latter is the frequency of the LSB-level squarewave; something is accentuating them even though there is no biphase-encoded datalink present with the USB connection, as there is with an S/PDIF or AES/EBU datastream, for which this signal is diagnostic. The Miller Analyzer estimated the level of these sidebands as 751 picoseconds peakpeak; this is moderately high compared with the best-performing DACs, but not too bad for an adaptive USB connection.
Fig.16 Chord Chordette Gem, high-resolution jitter spectrum of analog output signal, 11.025kHz at 6dBFS, sampled at 44.1kHz with LSB toggled at 229.7Hz: 16-bit APT-X data via Bluetooth (green), 16-bit A2DP data via Bluetooth (blue), 16-bit PCM data via USB (red). Center frequency of trace, 11.025kHz; frequency range, ±3.5kHz.
Chordette's Gem is well engineered, its shortfalls in measured performance stemming from the mandatory lossy codecs used to implement the Bluetooth link. But these shortfalls are severe enough that I feel Bluetooth capability, as convenient as it might be for wireless iPod use, becomes moot in the context of high-end sound.John Atkinson
Footnote 1: Though my comments throughout this sidebar refer to APT-X, this is based on the assumption that the Creative Music BT-D5's implementation of the APT-X codec is representative.John Atkinson