Wadia 2000 Decoding Computer Page 4
The DACs in the Wadia are proprietary 18-bit units, a "transversal, summed-multiport delay-line" design said to have a perfectly rectangular impulse response. The DACs are connected to the processor circuitry by 40 Megabaud optical couplers, but a surprising item is the absence of any deglitching and sample/hold circuitry. A glitch is an undesirable byproduct of switching, as in on-off, or one level to the next. It's particularly nasty at high frequencies, and the 2.82MHz rate used here certainly qualifies.
Wadia indicated that the capacitors needed to fix the problem are more of a headache themselves. At these switching rates, the dielectric memory is too long and the caps are forever lagging behind. Instead, Wadia chose to work on optimizing the very fast settling times. Don Moses pointed out that the 64x-resampling rate actually helps. The short intervals inherently limit the voltage-level changes to very small increments. Better control of these increments is therefore possible.
And yes, rather than merely truncating the 36-bit digital words used internally in their processing chips to the 18-bit length appropriate to drive the DACs, Wadia chooses to add dither and round off the words in order to avoid the high distortion levels at the LSB level due to the truncation. The LSB is very importanthere the information of the soundstage, the very high harmonics, and low-level information in general are encoded. To reintroduce quantizing distortion at this level by truncation would offset the advantages gained by resampling.
Those of you who would like more information about the inner workings of the Wadia processors should read Martin Colloms's review of the Wadia 1000 (Vol.12 No.9), the little companion of the Wadia 2000. MC covers a lot more ground technically which, I'm sure, will complement some of the ideas I've covered here.
The 2000 unveiled
The Wadia 2000 processor is comprised of four individual components, themselves grouped into two subsystems. In the first group we find the DigiLink 30 fiber-optic interface and input selector switch unit, and its separate regulated power supply. The second group consists of the processor chassis, along with a dedicated outboard power supply all its own.
The DigiLink 30 is a coaxial-to-fiber-optic cable converter and a selector switch for as many as three digital input sources, these connected by BNC sockets (footnote 6). For now, it easily accommodates a CD player and one or two R-DAT transports. Should satellite broadcasts be available in the future, the DigiLink 30 is ready for that. Its output is a high-quality fiber-optic driver used to couple the digital bitstream to the processor proper over a silica (not plastic) fiber-optic cable.
The DigiLink 30 addresses two significant problems. First, it eliminates potential ground-loop conditions by optically decoupling the CD transport from the circuitry of the processor. Second, coaxial cables for transmitting the digital data stream have been found to be critical and should be kept as short as possible. By locating the DigiLink 30 at the CD transport, the coax cable can be as short as 1', but since the supplied fiber-optic cable is 50' long, distance limits in the home are eliminated.
Both the CPU and the DigiLink 30 must be among the most substantially constructed pieces of equipment in the high end. In each case, Wadia took a solid chunk of aluminum and milled out a heavy-duty, one-piece shell. The five sides of the chassis are, therefore, a single piece of aluminum. A 0.125"-thick cover plate, held down by eight screws, completes the assembly.
The processor chassis is divided into compartments for each of the three main sections. Thus the components for processor, DACs, and audio output stage are separated. The components in the Wadia 2000 are of the highest quality and workmanship. I did not feel comfortable, however, with the insulation-displacementtype connections used to connect the three sections. It may be very convenient for implementing upgrades, but is that worth possibly compromising the connections?
I'm particularly concerned with the audio signal path. By the time you get from one board to the next, a total of six contact interfaces are encountered. That's two solder junctions, two connector contacts, two insulation-displacement interfaces, and a length of wire. A bit much, in my opinion.
Wadia is most emphatic about the application of an eight-layer printed circuit board. This technique allowed them to implement independent ground planes for not only the digital, but also the left and right audio channels. In addition, power distribution takes place on independent power planes.
The separate power-supply module is connected to the CPU by three separate cables: one for the digital circuitry and the other two for each of the left and right audio channels. A dedicated transformer is allocated to the digital circuits to prevent contamination of the audio portion by digital byproducts. Two more transformers feed the left and right audio channels. A total of 17 voltage regulators are used in the CPU alone, in addition to the proprietary pre-regulation at the AC line level, where power-line disturbances are minimized.
Footnote 6: Though Wadia refers to these inputs being AES/EBU compatible, this is not strictly true; the AES/EBU serial digital data standard also refers to the use of XLR connectors and balanced operation.John Atkinson