Meitner IDAT D/A processor Page 2
Although the IDAT implements its filters with the same the Motorola DSP56001 Digital Signal Processing (DSP) chips found in other custom-filter processors, they are used in a unique way. To understand how the IDAT's filter works, a little background in digital filters is helpful. Digital filters come in two varieties, Infinite Impulse Response (IIR) and Finite Impulse Response (FIR). IIR filters can be optimized for time-domain performance; they can have less ringing, overshoot, and pre-echo than FIR types. This makes them ideal for processing waveforms with steep attackslike musical transients. IIR filters, however, have less than ideal frequency-domain characteristics.
Conversely, FIR filters exhibit excellent frequency-domain behavior but generally offer poor time-domain performance. Virtually all CD players and digital processors use FIR filters. Squarewaves subjected to an FIR filter are afflicted with classic overshoot and ringing called Gibbs' Phenomenon; the FIR filter's impulse response shows such time-domain distortion as pre- and post-echo.
The solution to this quandary is to use both types of filters. In the IDAT, a detector looks at the nature of the audio signal and directs the digital audio data to the filter best suited for that type of signal. Musical components that are fairly continuous in level with no transients are routed to the FIR filter; signals with steep leading edges are sent to the IIR filter. The two filters' outputs are then combined and output to the DACs. The musical signal is thus processed by the optimum filter type for that kind of signal.
The word "Intelligent" in Intelligent Digital Audio Translator refers to the detector that recognizes the type of signal and routes it to the appropriate filter. This detector, which is implemented in software and held in EPROM (Erasable Programmable Read-Only Memory) chips, is the subject of a patent application. When I look at the IDAT's bench performance, it will be interesting to examine its squarewave shape and frequency response. If the IDAT's filter works as claimed, the measurement results will be unlike those of any other digital processor.
A few other notes on the filter: it requires four DSP56001 chips running at 50MHz. This is a serious amount of computing power, equivalent to 90MIPS (Million Instructions Per Second). Museatex is so proud of their invention that they put large windows in the front panel so you can see the DSP chips cranking away. The filter and detector were designed by Ed Meitner and Robert Gendron, the latter writing the DSP code.
To prevent the large amounts of digital noise generated by four DSP chips from getting to the DACs and contaminating the analog signal, the entire digital input and DSP sections are optically isolated from the DAC stage. A row of opto-couplers appears on the DSP filter's outputs, resulting in complete physical isolation between digital and analog electronics. Meitner went to great lengths to keep the analog and digital circuitry isolated. In addition to the opto-coupling, the power supplies (and their cables) are separate, the grounds are isolated, and the analog electronics are housed in the separate audio modules well away from the digital board.
The IDAT's D/A conversion stage is also unique. The circuit, which uses four Analog Devices AD1860 20-bit DACs (per channel) in a novel configuration, avoids a problem inherent in multi-bit DACs called "zero crossing distortion." In conventional DACs, distortion occurs at the point where the waveform crosses the zero line between positive and negative polarity. This distortion is conceptually identical to crossover distortion in class-A/B power amplifiers. Unfortunately, zero crossing distortion has a greater effect on low-level signals; the waveform discontinuity becomes much larger in relation to signal at low levels.
In the IDAT, two DACs per channel are used for each half of the balanced data signal instead of one. One DAC converts the positive-going portion of the waveform, the other converts the negative-going portion of the waveform. One DAC idles at full-scale positive, the other at full-scale negative. This technique shifts the MSB (Most Significant Bit) transition (where zero crossing distortion usually occurs) to two-thirds full-scale, rather than the zero crossing point where the waveform is more affected by this distortion. The DAC's intrinsic linearity error thus occurs at high signal levels where it is less audible than at low signal levels that are more easily corrupted. In fact, no MSB trimmer is needed: the linearity errors that would grossly distort low-level signals are inconsequential at high signal levels. Moreover, using dual 20-bit DACs in this way reportedly results in 21-bit conversion accuracy.
This dual-DAC technique is implemented twice per channel: The DACs are fully balanced, with one DAC pair handling one polarity of the balanced signal and a second pair converting the other balanced signal polarity. The IDAT thus uses a total of eight DACs. Note that the single-ended outputs are derived from the balanced signal.
The DAC outputs are summed at the current-to-voltage converter (I/V), an Analog Devices AD845 op-amp chip. A trimpot scales the DAC outputs so that the LSB (Least Significant Bit) values from each DAC match. Unlike conventional DACs that produce zero crossing distortion if the trim is misadjusted, the IDAT will produce second-harmonic distortion if out of alignment (footnote 2).
The word clock driving the DAC is processed by an additional C-Lock stage that removes jitter and Logic Induced Modulation (LIM) products generated in the DSP chips. Incidentally, the rear-panel digital outputs (used for driving a DAT machine or other digital recorder) are processed with C-Lock T (for Transmit) to assure a low-jitter digital output signal. Under "Measurements" (below), I examine the jitter-reduction capabilities of C-Lock.
The analog output stage is also based on the AD845 op-amp. Two AD845s per channel are biased into class-A with an active current sink. Given the IDAT's sophistication, I was surprised to see op-amps instead of a discrete output stage. While the AD845 is a superb op-amp, it's hard to believe that any monolithic device can exceed the sonic performance of a discrete stage (footnote 3). According to Ed Meitner, a good ceramic (not plastic) op-amp, implemented with careful attention to the power supply, low feedback, and pulled into class-A operation, can outperform a discrete circuit.
The direct-coupled analog stage uses an unusual output filter. Rather than low-passfilter the output to remove the high-frequency noise and 8x-oversampled images, the IDAT uses a notch filter that removes energy between 235kHz and 405kHz. This passive notch filter (inductors and capacitors) is reportedly better than a low-pass filter due to its linear phase and absence of group delay. In addition, it is not directly in the signal path. The capacitors used in the notch filter are biased with a DC voltage. Ed Meitner believes this technique alleviates many of the sonic problems associated with capacitors.
Finally, de-emphasis is performed in the digital domain by the DSP chips, and a pair of muting relays prevents noise and glitches from appearing at the audio output jacks.
In short, the IDAT's design and execution are unlike those of any other digital processor extant. The sheer number of unusual and innovative circuits in one product is surprising: the custom input receiver, input-stage C-Lock, dual DSP filters with intelligent detector, class-A DAC scheme, C multipliers, word-clock C-Lock stage, biased capacitors, optical isolation between analog and digital circuitry, and output notch filter. The IDAT is the full embodiment of Ed Meitner's ideas about digital converter design (footnote 4).
Build quality is first-rate, and the IDAT is ergonomically simple. I have a quibble about the amateurish owner's manual, however. For $15,000 one should expect a complete and professionally prepared owner's manual, not the cursory, 11/8-page instruction sheet supplied. Further, the IDAT put out an audible glitch when it locked to incoming data from the Theta Data transport. It also had trouble locking on to the signal from a Meridian 602 transport's coaxial output.
I've been listening to the IDAT for the past few months in my usual reference system. The IDAT's $14,950 price tag pits it squarely against the $13,950 Mark Levinson No.30, the revelatory digital processor reviewed in Vol.15 No.2 that forced us not only to rethink our standards of digital replay, but to completely restructure "Recommended Components." You may recall that the No.30 became the sole digital product in Class A. If the IDAT doesn't equal or exceed the No.30's extraordinary musical performance (or if it achieves Class B performance), it will be impossible to recommend considering the unit's hefty price.
After hearing the IDAT in isolation and in comparison with the Theta DS Pro Generation III ($5400 fully loaded), I knew that it was a contender for Class A. The IDAT's presentation was extraordinaryand eminently musical. Yet it was only after extended comparisons with the Mark Levinson No.30 at matched levels that I became intimately familiar with the IDAT's strengths and weaknesses.
Footnote 2: A similar technique is the basis of the popular Burr-Brown PCM63 "Colinear" DAC. It should be noted that Ed Meitner's technique, which he calls a "Class-A Data Scheme," preceded the PCM63.
Footnote 3: Virtually all digital processors have an op-amp in their signal pathseven those with discrete output stages. The op-amp appears right after the DAC, converting the DAC's current output to a voltage. Because the DAC outputs a waveform with a rectangular shape, a very fast circuit is needed to avoid slewing distortion in the current-to-voltage converter. Monolithic op-amps are said to be typically faster than discrete circuits.
Footnote 4: See my interview with Ed Meitner elsewhere in this issue.