Do Hard-Disk Drives Sound Better than CDs?
A reader's post of an interesting and relevant question regarding the sound quality of hard-disk-based music servers versus CD
(http://www.avguide.com/forums/cd-player-vs-computerserver-system) prompted me to post my experience with the subject as this blog. This text originally appeared in The Absolute Sound Issue 177 as part of my review of the Sooloos and QSonix music servers.
There’s been much debate about the sound quality of CDs played back by a conventional transport versus the sound quality of files made from the same CDs, copied to and then played back from a hard-disk drive. Many contend that hard-disk drives sound better, much as CD-Rs sound better than the source CD from which the CD-Rs are made.
I tested this contention by comparing the sound of the Sooloos and Qsonix servers to a state-of-the-art CD transport (the Esoteric P-03), alternately feeding the music server under evaluation and the CD transport to an Esoteric D-03 digital-to-analog converter through the same digital interconnect.
Listening to the same CD from Sooloos and from the transport, I heard a subtle but noticeable improvement in sound quality from the music server. I heard more space, air, bloom, and soundstage depth when the audio data were read from the Sooloos hard drive rather than from the CD. The hard-drive-sourced sound had better resolution of low-level detail, particularly reverb decay, which is why the presentation sounded more airy and spacious. The presentation was gentler, a little more laid-back and relaxed, and had a greater sense of ease. The upper midrange and treble were smoother and more “organic” sounding when the data feeding the DAC was sourced from Sooloos’ hard-disk
drive. I repeated this comparison with the Qsonix server and heard a similar increase in resolution, improvement in spaciousness and depth, smoother textures, and more relaxed presentation. In fact, the two music servers sounded very much alike.
Much of the current speculation as to why this should be the case focuses on the fact that the Exact Audio importing software used in Sooloos and many other hard-drive-based systems can read sections of the disc multiple times if data errors are detected. The implication is that random bit errors affect sonic qualities such as soundstage depth and the reproduction of timbre.
I disagree. CD’s error-correction system can completely correct burst errors of up to 4000 successive bits. My experience performing bit-for-bit comparisons between source data and replicated CDs when I worked as a CD mastering engineer suggests that bit errors are virtually nonexistent. If CDs were rife with errors, computer software couldn’t reliably be delivered on CD (CD-ROM does, however, have an extra layer of error detection and correction but is invoked only when discs have been mishandled and become damaged).
Let’s perform a thought experiment. Suppose that uncorrected bit errors are common in CD playback. What are the sonic consequences of those errors? If a 16-bit word’s least significant bit (LSB) were changed from a zero to a one (or from a one to a zero), an amplitude error would be introduced that is one part in 65,536. (The number of amplitude steps in a quantization word is 2n, where n is the number of bits in the word; 216 is 65,536.) You would never hear an amplitude error that small in a single sample. But let’s say the bit error occurred in the word’s most significant bit (MSB). The value of the MSB is the value of the other 15 bits combined plus one (32,769). The MSB represents just over half the amplitude of a full-scale signal. Changing the MSB from a zero to a one, or from a one to a zero, would cause the signal level to suddenly increase or decrease by half the full-scale amplitude, producing a loud click at every error. Do you hear clicks when playing CDs?
Note also that bit errors would cause momentary jumps in signal amplitude, but would not affect factors such as soundstage depth or instrumental timbre. Bitstreams read from different media (CD, CD-R, hard disk) can sound different from one another, but not because of data errors. The most likely explanation is that hard drives deliver a bitstream with greater timing precision (lower jitter). If the bits are the same, and the sound is different, the only thing left is jitter.
The Esoteric P-03/D-03 combination’s separate clock link allowed me to test this hypothesis. This separate clock link slaves the transport to the processor’s clock, a feature that greatly reduces jitter in the D/A’s word clock. (The word clock controls precisely when the DAC converts the incoming samples to analog, which is the point where jitter matters.) The arrangement obviates the need to extract a clock from the S/PDIF interface, a primary source of jitter in separate transports and D/A converters. In addition to avoiding the jitter-inducing effects of the interface, another advantage of a separate clock line is that the D/A converter becomes the timing reference for the entire system. That is, the D/A converter generates the timing reference for the crucial digital-to-analog converter stage, and the transport must lock to this reference. In a conventional connection between a transport and D/A converter, the D/A converter must lock to the transport’s timing reference. The result is higher jitter in the clock that controls the actual digital-to-analog converter chips. The separate clock line eliminates these two shortcomings.
I compared the sound from the music server’s digital output to the Esoteric P-03 transport with and without the separate clock line (you can engage and disengage the clock with a front-panel button). Engaging the Esoteric’s clock (lowering the jitter) when listening to a CD rendered an improvement in sound quality virtually identical to what I heard when the music was being streamed from the music servers. This suggests that jitter is, indeed, the explanation for why hard-disk drives sound better than optical discs.
Note that in this experiment, the music server’s output was subjected to the jitter-inducing S/PDIF interface while the transport’s output was not. One might reasonably infer that jitter in the signal recovered from an optical disc is at least as significant a source of sonic degradation as the S/PDIF interface. The ideal solution is hard-disk-based storage with no S/PDIF interface. It’s not just the S/PDIF interface that’s a problem; any asynchronous connection between source and DAC is potentially problematic, including the popular USB connection for connecting a computer to an outboard DAC.
We will increasingly listen to music from hard-disk drives; it’s good news that their advantages don’t come with a sonic penalty. Indeed, hard-disk drives actually deliver better sound than optical disc.