Manufacturer's Specifications:
Input Resolution: 18 bits, 128 times oversampling.
Amplitude Response: 5 Hz to 20 kHz, +0.1,-0.5 dB.
PLL Re-Clocking Jitter: Less than 50 picoseconds.
THD + N: Less than 0.004% at 1 kHz.
EIAJ Dynamic Range: Greater than 100 dB.
Channel Separation: Greater than 96 dB, 5 Hz to 20 kHz.
Internal Precision: 56 bits (338 dB).
Dither Type: Triangular probability density function or narrow-band weighted; user-selectable.
Maximum Output Level With 600-Ohm Loads: Balanced, +22 dBV; unbalanced, + 16 dBV; headphone, + 18 dBV.
Minimum Load Impedance: Balanced, 150 ohms; unbalanced. 300 ohms; headphone, 8 ohms.
Input Sampling Rates: 32, 44.1, and 48 kHz.
Dimensions: 17 1/2 in. W x 4 3/8 in. H x 18 1/4 in. D (44.5cm x 11.1 cm x 46.3 cm).
Net Weight: 42 lbs. (19.1 kg).
Price: $5,000.
Company Address: 12 Alfred St., Baldwin Park I, Woburn, Mass. 01801.
As a trained electronics engineer, I have always been a bit skeptical when reading some claims made for so-called high-end equipment, particularly when the claims lack scientific basis. My skepticism turned into unbridled enthusiasm when I encountered the Vimak DS-2000. Its designers were obviously schooled in the scientific approach to handling digital audio signals. yet at the same time they still know the value of simplicity, physical beauty of product, and ergonomics. Much more than a digital-to-analog converter, the DS-2000 incorporates control functions (such as volume and balance and selection of up to seven digital inputs) that allow it to serve as the only interface between your power amplifier and digital program sources.
Thick. solid aluminum surrounds the DS-2000's steel frame, isolating the internal circuitry from electromagnetic interference and providing a high thermal mass. Digital, analog, front-panel, and power-supply board assemblies are all shielded in separate sections, and the entire inner framework is copper plated to reduce electrical and magnetic interference. Separate digital and analog power supplies are used. and optical couplers are provided. to eliminate electrical connections that could carry noise between the analog and digital circuit boards.
Once you've selected a digital input signal. it's read by a digital audio interface receiver that extracts the sampling rate clock and up to 24 bits of audio data (if the source provides them), using a low-jitter phase-locked loop (PLL). From there, it goes through a digital signal processing (DSP) section that performs volume, balance, polarity inversion, and dither computations with 56-bit internal precision, according to Vimak. The volume control is a hybrid digital type: in the upper part of its range, it varies the reference voltage to the D/A converters, while at lower settings it multiplies the digital signal by a gain coefficient of less than 1. Dither is automatically switched on when the input source has 18-bit or greater precision or when the volume control is in its lower range. The user can choose either broadband triangular dither (which minimizes the audibility of quantization but raises the noise floor about 9 dB) or dither weighted towards the inaudible part of the spectrum, which is filtered out in conversion but uses ultrasonic frequencies some listeners may believe are audible.
The next circuit section, a delta-sigma modulator, up samples the audio data by a factor of 128 and generates separate one-bit outputs for left and right channels. At this point, the signals are fed through opto-couplers to the analog board, where the signals are reclocked by another PLL, to limit jitter. The reclocked signals for each channel are then fed through four balanced, single-bit D/A converters of the pulse-density-modulation type, whose outputs are summed in the analog domain The summing accomplishes some of the needed low-pass filtering, minimizes the effects of any nonlinearities in individual D/A converters, and cancels switching glitches. After additional analog low-pass filtering, the left and right signals are fed to separate output driver stages for the balanced, unbalanced, and headphone outputs. These details are covered more fully in a brochure, "DS-2000: Insights into Theory and Design," available on request from Vimak.
While the front-panel layout seems simple and uncluttered, the versatility of the Vimak's few controls and switches is awesome. The power switch is at the far left. To its right are five input selectors and a "Program" button, about which more in a moment. The display at the center of the panel shows the digital sampling rate, the input in use, and all current control settings. An ''Error" light in the display indicates defects in the source data or that the selected source isn't playing.
Opening the hinged panel below the display discloses the stereo headphone output jack, the auxiliary optical and coaxial inputs and the pushbuttons to select them, a polarity-inversion key, and a "Fixed/Variable" switch. This switch disables the volume and balance controls, allowing the unit to be used with an outboard preamplifier.
The right-hand section of the front panel contains a -20 dB muting switch, the balance control, and the volume control. When used in conjunction with the "Program" button, the balance control takes on additional functions: Varying display brightness, selecting the type of dither you want, setting the "Automation ID Number" (to give the unit a unique digital address for use in automation systems, and for factory diagnostic use), and changing the displayed source indications to the names of your actual sources.
(Names can be up to 20 characters long and use any of 70 alphanumeric characters, including both capital and lowercase letters.) The DS-2000 accepts up to seven digital inputs: Toslink and glass optical, two coaxial, and one AES/EBU input on the back panel plus the optical and coaxial inputs behind the hinged section of the front panel. The rear-panel outputs include balanced XLR and unbalanced phono jacks plus computer-style DB-9 "Automation" input and output jacks for connection to RS-232 or RS-485 computer serial ports.
At the moment, these ports are only used by Vimak in factory testing, but built-in software routines will eventually allow the DS-2000 to be linked to the factory via modem for diagnostic tests, and the software is being changed to allow displayed source names to be typed in from a computer keyboard. The company also plans to offer interface units to link the DS-2000 to some home-automation systems.
The supplied infrared remote duplicates all of the front panel input selectors, the volume and balance functions, the polarity-inversion and muting functions, and the "Program" key. Combined use of the remote's "Program" and "Balance" buttons enables you to access the previously described programming functions just as easily as from the front panel itself. There is also a "Mode" button reserved for future applications.
Measurements
I approached bench testing of the DS-2000 in two ways. From a practical point of view, I wanted to see how the unit would perform when fed with a digital input source, such as my reference CD player's digital output terminal. Then, from a purely theoretical point of view, I wanted to compare that performance with what the DS-2000 would do when fed with a digital signal generated by my Audio Precision test equipment. It should be noted that the AP equipment generates 24-bit data words. The AP then sets the dither level and assumes that the receiving device will simply truncate at the set level. However, the DS-2000 extracts the full 24 bits, as mentioned earlier, and feeds the data to its internal DSP circuits. Since data is present below the 18-bit level, the DSP will automatically add dither of its own, effectively adding dither on top of dither. This will result in misleadingly high noise and THD readings. The problem was avoided by "tricking" the Audio Precision unit, setting its output word length to 25 bits. This prevented dither from being added to the output data and allowed the DS-2000 to properly dither the data internally.
The frequency response when decoding the digital output from my reference CD player, using the CBS CD-1 test disc (Fig. 1A), is down 0.25 dB at 20 kHz, well within Vimak's published specifications. Response to the digital output of test signal with a 48-kHz sampling rate (Fig. 1B) is virtually the same. The DS-2000 provides a digital output as well as digital inputs, so it can serve as a control center for digital signals. I therefore measured response at the digital output using my Audio Precision equipment to convert the signal back to the analog domain. This response (not shown) was ruler flat to 20 kHz, suggesting that the slight high-frequency attenuation seen in Figs. 1A and 1B was caused by the analog low-pass output filters or by the analog amplifier stages.
I used the same three setups to measure THD + N (Fig. 2). With input from the CD-1 test disc and my CD player, THD + N remains at around 0.003% over most of the audio spectrum, reaching the rated 0.004% only at 10 kHz. With a 48-kHz test signal from my generator, THD + N is even lower, remaining at around 0.002% for the left channel and 0.0026% for the right channel over most of the frequency range. With the same 48-kHz input signal, mid-frequency distortion at the digital output measures between 0.00005% and 0.00006%, nearly 125 dB below reference level! Next, I checked THD + N versus signal amplitude, using track 18 of the CD-1 test disc as a signal source. This track provides signals of decreasing amplitude from 0 dB (maximum) to -90 dB. The THD + N, expressed in dB below maximum recorded level, generally hovered around the 95 dB mark, regardless of signal amplitude. This corresponds to a THD + N percentage of 0.0018%. To isolate THD proper from quantization noise, I used the Audio Precision's FFT facilities to conduct a spectrum analysis of residual harmonic distortion components of a 1-kHz signal at full amplitude. The only significant harmonic (at 3 kHz) had an amplitude of 108 dB, corresponding to a true THD percentage of 0.0004%. A spectrum analysis of a-60 dB, 1-kHz signal (Fig. 3) revealed no outstanding harmonic components and showed a noise percentage of around 0.04% relative to the-60 dB signal. This is nearly 68 dB below the-60 dB signal level! Channel separation, plotted versus frequency in Fig. 4, reveals a separation of approximately 115 dB at mid-frequencies. At 16 kHz, channel isolation was still 105 dB. The curve slopes suggest that if my test disc had a 20-kHz signal on one channel, the separation I'd measure at that frequency would still be greater than the 96 dB Vimak claims.
Fig. 1--Frequency response for digital output from test CD (A) and 48-kHz
signal from test generator (B), both measured at analog output of DS-2000.
Fig. 2--THD + N vs. frequency for, top to bottom, signals from CD, signals
from test generator at 48-kHz sampling rate (both measured at analog output),
and test generator signal measured at digital output. Note that bottom
of scale is 0.00001%. Solid curves are for left channel, dashed curves
for right channel.
Fig. 3--Spectrum analysis of 1-kHz signal at-60 dB recorded level, measured
at analog output.
Fig. 4--Separation vs. frequency.
Fig. 5--Spectrum analysis of residual noise when decoding "no signal" CD
track.
The A-weighted signal-to-noise ratio, obtained using the "silent" track of the CD-1 test disc and with the signal applied to the DS-2000 via its coaxial digital input, was 103.4 dB on the left channel and -101.2 dB on the right channel. A spectrum analysis of the residual noise as a function of frequency (using a third-octave bandpass filter to track the plot) is shown in Fig. 5. The left-channel hum component at 60 Hz reaches a peak of only -104 dB, while the right channel seems better isolated from the power-line frequency.
Fig. 6--Deviation from perfect linearity, using undithered signals from CD.
Fig. 7--Linearity, output vs. input, for dithered Figure 6 shows deviation
(or, in this case, lack of deviation) from linearity over the amplitude range
from 0 to 90 dB, using undithered signals from the CD-1 test disc. These
results are clearly superior to those from any other digital device I recall
having tested over the years. At -90 dB, deviation amounts to less than -0.3
dB, providing ample proof of the superiority of one-bit D/A conversion, at
least as it is implemented in this superb Vimak D/A converter circuitry.
To further confirm the DS-2000's superb linearity, I used the Audio Precision
system to generate a dithered digital signal of gradually decreasing amplitude
and then compared input and output levels for this signal (Fig. 7). Linearity
is virtually perfect from 0 down to -110 dB. Note, too, that linearity is
virtually identical for the left and right channels.
The only other bench test I made was to assess master clock accuracy, which turned out to be within 0.0002% of absolute accuracy. Translated to musical terms, this means that a middle-A tone (440 Hz) would be reproduced as 439.99912 Hz. The difference would be indistinguishable even to musicians possessed of perfect pitch.
Use and Listening Tests
I found the DS-2000 to be simple to use, once I familiarized myself with its somewhat obscure programming functions. My problem in conducting listening tests was: What source material would I use to fully demonstrate the excellent qualities of the DS-2000? Listening to CDs was all well and good, but I knew that the limiting factor would be the inherent quality of the CD itself. The same would hold true for the small collection of DAT cassettes that I have amassed over the last couple of years. Still, listening to pure tones generated by my test equipment would hardly make sense. I finally arrived at a compromise solution: I would listen to CDs and DATs, using the DS-2000 for D/A conversion, and compare the sound quality with that obtained from the analog outputs of my reference CD player (which, at its introduction a couple of years ago, had a suggested retail priced of around $1,200). The CD I chose was Engineer's Choice (Delos DE 3506). This disc's wide variety of classical selections was engineered by my good friend John Eargle, one of the most skilled recording engineers I know. Selections chosen included the "Gun Battle" from Copland's Billy The Kid, the Scherzo from Shostakovich's 10th Symphony, and for evaluating the sound of a piano, Rachmaninoff's Prelude in G Minor. Could I hear a difference between the two modes of sound reproduction? You bet I could! In the "Gun Battle" sequence, the percussive sounds seemed to have a better attack or transient response when reproduced using the DS-2000. The Shostakovich selection, though highly dynamic when heard from either setup, seemed to exhibit greater clarity and definition of individual instruments when played through the DS-2000. And as for the piano selection, all I can say is that it sounded more like a piano when heard through the DS-2000. I repeated the tests using the headphone outputs of the Vimak as well as the CD player, and my observations and conclusions were consistent with the tests using loudspeakers.
My chosen prerecorded DAT was one issued early on by GRP, Digital Duke (GRT-9548), featuring the music of the great Duke Ellington played by the Duke Ellington Orchestra. After the experience with the CD, I was not surprised that "Mood Indigo,' my favorite tune on this tape, sounded mellower and less strident through the recorder's optical digital outputs and the DS-2000 than via the recorder's analog outputs.
In short, the improvements in sound quality that I heard convinced me that the folks at Vimak have not only come up with a better way to decode digital audio signals but that devotees of the purest form of sound reproduction would do well to audition the DS-2000 (and, when available, the somewhat lower cost DS-1800). Such auditioning may well lead to the conclusion that, if anything, the DS-2000 is underpriced compared with some other high-end D/A converters that offer neither the versatility and features of the DS-2000 nor its superb sound quality.
-Leonard Feldman
(Source: Audio magazine, Nov. 1992)
Also see:
Vacuum Tube Logic Straight-Line D/A Converter (Nov. 1992)
TEAC Esoteric X-1 CD Player (Nov. 1982)
= = = =