Audio Alchemy DTIPro 32 Jitter Filter
With the introduction of Audio Alchemy's Digital Transmission Interface (DTI) more than three years ago, the company created an entirely new category of hi-fi product: the jitter filter. The original DTI was a good start, but didn't always improve the sound of the better-quality digital front-ends.
The DTIPro, released 18 months later, bore little similarity to the rather simple DTI. The Pro added a more sophisticated dual Phase-Locked Loop (PLL) input receiver for much lower jitter, and also performed a new type of Digital Signal Processing (DSP) that Audio Alchemy called "Resolution Enhancement." This processing reportedly increased the resolution of the compact disc's 16-bit digital words, approximating 18- or even 20-bit resolution. It did this by looking at the 16-bit words over a time window, and calculating the additional bits that would have been present had the signal been originally encoded with higher resolution than 16 bits.
In my November 1994 review of the DTIPro (footnote 1), I called it a "breakthrough in digital audio reproduction" and "a musical revelation." When inserted in the playback chain between a transport and digital processor, the DTIPro significantly improved the sound of my system. Moreover, experimenting with the output word length suggested that the Pro's benefits were indeed rendered by Resolution Enhancement, not just jitter reduction.
Shortly after the DTIPro was launched, Star Semiconductor, the manufacturer of the Pro's DSP engine, temporarily went out of business. Although Audio Alchemy scoured the globe and hoarded all remaining Star DSP chips, it was clear that the DTIPro's days were numbered.
Audio Alchemy took the opportunity to redesign the Pro with a 32-bit DSP chip from Texas Instruments, hence the product's new name, the DTIPro 32. The new chip reportedly has 50% more computing horsepower to run the Resolution Enhancement algorithm. The Pro 32 also includes some operating improvements over the Pro. Fortunately, I still have an original DTIPro on hand for comparison with its successor.
The DTIPro 32 is housed in Alchemy's standard 8.5" wide and 2" high chassis. From the front panel, the DTIPro 32 looks identical to the DTIPro. The front panel's 10 LEDs indicate the input selected, power to the 32's three main sections, when the unit is locked to a source, and whether the phase or polarity has been inverted.
The lock indicator actually requires two LEDs, marked "Primary" and "Secondary." The Primary lock shows that the Pro 32's first PLL stage has locked to the source, and the Secondary LED indicates the low-jitter "double-lock" condition. The Pro 32 should double-lock to most CD transports. More on this dual PLL later.
The tiny rear panel is consumed by input and output jacks. Input is via coaxial (BNC) jack, TosLink optical, or Alchemy's I2S bus. The addition of the I2S bus input is new on the Pro 32, and is included to accept I2S output from Alchemy's forthcoming CD transport. For an additional $179 you can swap the TosLink input for an ST-Type optical jack. The Pro 32's output appears on a BNC jack, AES/EBU connector, ST-Type glass optical jack, and I2S bus. A DC input connector, which accepts a plug from the Pro 32's small external power supply, finishes the rear panel.
An I2S connector looks like an S-Video jack, but has five pins rather than S-Video's four. The five pins carry left and right audio data on one line, bit clock, word clock, master clock, and the emphasis flag. By separating the clock and transmitting it independently of the audio data, the unit receiving an I2S-formatted signal need not "lock" to the incoming datastream and "recover" a clock. The result is nearly total immunity to interface-induced jitter. Indeed, the whole idea of extracting a clock signal from the audio data is a fundamentally flawed approach to transmitting digital audio. Note that when one of Alchemy's processors with I2S input is connected to the DTIPro 32's I2S output, the familiar and comforting "lock" LED doesn't illuminate on the processor's front panel. Don't worry; the processor will still decode the I2S data.
Seven user-selectable operating modes optimize the Pro 32's output for your digital/analog converter. These modes set the output word length (16, 18, 20, 22, or 24 bits) and other output conditions. One mode turns off the dither, but performs Resolution Enhancement. The output word length in this no-dither mode changes with the program source, with the algorithm deciding how many additional bits it can reliably interpolate. Turning off the DTIPro's dither can be a benefit with some digital processors that are already dithered internally.
The seventh mode turns off the Resolution Enhancement processing for HDCD$r playback. HDCD decoding requires that the control code buried in the least significant bit of the CD's 16-bit words arrive at the HDCD filter in your digital processor. Any change in the datasuch as that introduced by Resolution Enhancementcorrupts this control code and prevents HDCD decoding. This is why the Pro 32 offers the HDCD bypass mode. Note that this HDCD bypass option wasn't offered on early versions of the Pro.
Choosing the correct output word length is crucial to getting the best performance from the DTIPro 32. Let's say you have a digital processor with 18-bit DACs, and you set the Pro 32's output word length to 20 or more bits. The DACs will simply truncate (cut off) any bits below 18, introducing noise and distortion. As JA described in his review of the Meridian 518 Mastering Processor in the January '96 Stereophile, truncation also hardens midrange textures and reduces the sense of space.
Conversely, setting the Pro 32's output word length to 18 bits if you have a true 20-bit processor prevents you from experiencing the full benefits of Resolution Enhancement.
Just because your digital processor has 20-bit DACs doesn't mean that it will pass 20-bit data from input to output. The NPC 5813 and 5813 digital filters, for example, truncate the incoming data to 18 bits. The older Yamaha YM3623 input receiver chip will pass only 16-bit data. For comparison, the Crystal Semiconductor CS8412 and UltraAnalog AES21 input receivers will pass up to 24-bit data, as will the Pacific Microsonics PMD100 HDCD decoder/filter.
There's yet another trap to be aware of. Digital processors with the Crystal or UltraAnalog input receivers, the PMD100 filter, and 20-bit DACs still may not pass 20-bit data. Some processors that have been retrofitted with the PMD100 weren't redesigned to pass the PMD100's full resolution to the DACs. Two such examples are the PS Audio UltraLink Two and Enlightened Audio Designs DSP7000 Series III. Each of these processors has only a 16-bit data path from the digital filter to the DACs. Consequently, they'll truncate 20-bit input data, even though they have a 24-bit input receiver, 24-bit digital filter, and 20-bit DACs. Any processor designed from the ground up around the PMD100 (as opposed to an existing design retrofitted with the PMD100) should pass at least 20-bit data. The only way of finding out which "20-bit" processors truly pass 20-bit data is by the measurements included in reviews. Reports from readers suggest that some processor manufacturers either aren't aware of their products' capabilities, or provide misleading information.
Footnote 1: See my review of the DTI in Vol.16 No.5, DTI measurements in Vol.16 No.11, and DTIPro review in Vol.17 No.11.