Chord Electronics DAC64 D/A processor
Yet my experience and that of Stereophile's reviewers has been that, yes, they still sound different. What has happened is that the cost of owning a state-of-the-art CD-playback component has dropped significantly, compared with only a few years back. So when British amplifier manufacturer Chord Electronics showed its first D/A processor at the 2001 Consumer Electronics Show, I was not too surprised that this statement product was intended to sell not for $20,000 or $10,000 or $5000, but for $3040.
The DAC64 looks as if carved from a solid chunk of aluminum, the only visible highlights being a recess for the gold-plated logo and a convex glass window, though which the circuit board can be seen, illuminated by LEDs of various colors. In fact, these internal LEDs provide feedback to the user: When you turn the DAC64 on, blue LEDs light up. When the DAC64 locks to a data source, red LEDs add a purple hue. When either of the RAM buffers is selected, yellow LEDs illuminate.
The rear panel features three digital inputs—TosLink optical, S/PDIF electrical via a BNC jack, and AES/EBU on the usual XLR—selected by a three-position toggle switch. Although the specification states the DAC64 will accept 192kHz-sampled data, the necessary second AES/EBU jack doesn't exist. A second toggle selects between no RAM buffer and either of the two RAM settings. Both single-ended (RCA) and balanced (XLR) outputs are provided.
In its announcement of the DAC64, Chord fired a salvo in the numbers wars by talking about a "64-bit" DAC. This sounds like overkill, considering CD's and DVD's respective 14- and 16-bit limits. And, of course, 64-bit performance implies a dynamic range that might well allow the Big Bang itself to be captured in full fidelity. But what is actually meant, as Chord's literature carefully explains, is that the digital reconstruction filter used in the DAC64 is realized using a 64-bit DSP core, and that the Pulse Array D/A chip uses seventh-order noise-shaping realized with 64-bit mathematical precision. Both mean that any mathematical error due to the filtering and noise-shaping calculations will be way below the analog noise floor, and therefore inconsequential.
The digital filter used by Chord is of a type new to me. Called a WTA filter, for "Watts Transient Aligned," it is said to minimize timing errors, therefore reducing the need for large numbers of filter taps to be used to achieve a given performance. Nevertheless, the DAC64's filter uses 1024 taps, compared with the 256 typical of many commercial digital filters. The DAC64 doesn't offer HDCD decoding, however.
I referred above to the RAM buffer. This is basically arranged as a FIFO (First-In, First Out) store. In theory, the clock accuracy with which the data are clocked into the FIFO doesn't matter, as the data are clocked out with a high-precision local crystal, which in turn should reduce jitter to vanishingly low levels. In practice, there has to be some means of locking the local clock to the long-term-averaged clock of the incoming data, which will mean low-frequency jitter might still propagate to the DAC chip. But because Chord uses a very much larger FIFO than is usual, all jitter in the incoming data above a very low frequency should be rejected.