July 2000Assemblage D2D-1 Sample-Rate Converter
by Doug Blackburn
In the beginning, Sony and Philips said, "Let there be 120cm digital audio discs and one-box players to allow us to hear this marvel of modern technology." Audiophile companies then split the one-box players asunder into transport and DAC in separate boxes connected by a single cable. Then jitter was heard around the world, and it was a very bad thing. Soon there were third boxes to smite jitter a powerful blow. As digital audio aged it became wiser and grew to 24/96. But the world was puzzled by the change because in the new beginning there were only one-box players again. But soon the sun shone on separate 24/96 transports and DACs. And the world was happy again, until it realized that 24/96 was also plagued by jitter and suffered old 16/44.1 discs that sounded not as fine as the newest 24/96 discs were known to sound throughout the land.
The Assemblage D2D-1 is the first 24/96 "third box" product that Ive encountered. While de-jitter boxes for standard CD transports and DACs have been around for a while, nothing like the D2D-1 has existed for 24/96. In the old days, jitter reduction alone was enough functionality to get us to part with a moderate chunk of pocket change. Today, the D2D-1 entices us with jitter removal, sample-rate conversion, data-word interpolation and digital-format conversion -- all for about the same chunk of slightly inflated pocket change that the early 16/44.1 jitter-removal-only boxes cost.
To use the D2D-1 in your system, the minimum requirement would be a separate 16/44.1 transport and 16/44.1 DAC. However, this configuration wont let you use the D2D-1s upsampling capability. To make the D2D-1 worthwhile, you really should have a 24/96 DAC. And to get the full measure of sonic bliss, having a 24/96 transport adds the last little bit of finesse to the sound. Lets look at each feature of the D2D-1 before getting into the sonic performance.
The D2D-1 accepts any standard PCM digital sampling rate: Mini-Discs 32kHz, CDs 44.1kHz, DATs 48kHz. You can select whether the output signal is 48kHz or 96kHz, or not converted at all. Internal circuitry uses mathematical algorithms to add sample points to the digital signal. To upsample to 96kHz, the D2D-1 outputs 2.18 samples for each original 44.1kHz sample. This is tricky because 96kHz and 44.1kHz samples dont match in time very often. This means that the 96kHz output mode is really doing a lot of remodeling of the digital input signal by placing 96,000 samples per second where 44,100 samples per second originally existed. You could, in theory, do this operation alone and output 16-bit/96kHz digital audio, but all current upsampling products combine sample-rate conversion with data-word-length interpolation, which is .
Some background. Heres what a random 16-bit data word looks like for CD audio:
There are 65,536 different values represented by the 16 digital bits (2 raised to the 16th power). Each of these values represents a voltage in the analog output signal. If the DAC IC outputs two volts maximum, then each different value for the 16 bits represents .0000305 of a volt (this isnt a precise analysis, just a general conceptual overview).
0000000000000000 = zero volts
If you increase the word length from 16 bits to 24 bits, the number of different voltages you can represent increases from 65,536 to 16,777,216 (2 raised to the 24th power). Each step in a 24-bit word would represent .00000006 volts, still using our 2-volt output model, compared to the .0000305 volts for each step in a 16-bit word (using the same 2-volt output example). You can see that the granularity in voltages representing the musical signal is incredibly fine when the data words are 24 bits long. Take an example where the audio signal wants to be .0000455 volts (still in our 2-volt example). With 16 bits available, software has to decide whether to make this voltage become .0000305 or .0000610 volts. There is no way with 16 bits to make a voltage that is .0000455 volts. But in the 24 bit world, you have an extra 256 different voltages available between .0000305 and .0000610 volts. One of them will be almost exactly .0000455 volts.
Twenty-four bits is really higher in resolution than any consumer analog or digital audio products can achieve due to limitations in current electronic technology. Most digital products with claimed 24-bit performance lose probably three to four bits in the noise floor. But nevertheless, it is advantageous and comforting to have a digital standard that actually exceeds the capabilities of performance in consumer audio components. Im not sure you would want to listen to music that requires all 24 bits to reproduce anyway; the loudest sounds would be as loud or louder than the loudest noise you ever heard in your life, well beyond the threshold of pain. The quietest sounds would require the total silence of an anechoic chamber to be able to hear them. What 24 bits brings to the table is headroom and footroom, which make digital audio more forgiving and easier to work with and offers an improvement in resolution even if limited to 20 or 21 bits of effective resolution.
The circuitry in the D2D-1 creates a higher resolution digital audio bitstream by analyzing sequential groups of 16-bit data words and generating interpolated (best guess) 24-bit data words to replace the original 16-bit words. The process does have some margin for error, but it is vanishingly low. Keep in mind that in converting 16-bit audio to 24-bit audio does not increase the resolution of the reproduced audio. You cant have more than 16 bits of resolution when you start with 16 bits of data, but the higher resolution digital bitstream can be more optimally transmitted and processed to make small improvements in sound quality.
At 44.1kHz, 16-bit samples are .0000227 second apart (four zeros) compared to jitter, which is in the range of eight or nine zeros after the decimal point. At 96kHz, samples are .00001042 of a second apart. Taking the worst-case jitter of 2000 picoseconds in 16/44.1 audio playback devices, your samples could occur at any time base between .000022701 and .000022609 of a second, a 2000-picosecond window. Is eliminating this jitter really audible? Im not sure that the definitive testing has been done yet, but there are postulations that the jitter is often correlated to something -- power-supply noise, signal content, etc. -- so that patterns appear in the jitter. When this happens, there are possibilities for jitter to become more of an issue than it would be if it were a totally random event. There is no question that technically zero jitter is absolutely the best way to reproduce PCM digital audio. But questions remain. How audible is jitter? How low is low enough?
The D2D-1, if it performs to the level specified by Sonic Frontiers -- less than 2 ps jitter at the output -- would be a stunning accomplishment. Previous 16/44.1 de-jitter devices struggled to hold jitter below 100 ps at their outputs. This would make the D2D-1 the most impressive jitter-reducing product Ive ever encountered. It takes some stunning technology to be able to measure jitter at the 2 ps level -- technology that I have no access to. Sonic Frontiers supplies a comprehensive white paper with the D2D-1 on the unit's technical capabilities, with supporting measurements made with an Audio Precision System-2 test center.
The D2D-1 uses a dual-PLL (phase-locked loop) configuration for reducing jitter. This idea is not new -- other products have used the same kind of circuitry -- but its implementation in the D2D-1 would seem to be more robust, if the 2 ps specification is accurate. There is an internal jumper that can be defeated when you use a transport with greater than +/- 200ppm absolute error, in which case the second PLL may not lock.
As if all of the features I've mentioned weren't enough, the D2D-1 also allows you to connect and select between several different digital inputs and drive all outputs simultaneously. You could connect a Mini-Disc, DAT, DVD player and CD player for example. The D2D-1 then outputs whatever input you have selected. This means you can use one DVD-based transport and via the D2D-1, connect it to a high-performance 24/96 DAC for high-quality stereo listening while simultaneously connecting to a surround processor or A/V receiver. The D2D-1 can be your digital switcher and distributor. The D2D-1 supports two infrequently seen copper-wire digital-transmission interfaces: I-squared-S and I-squared-S-Enhanced. These are different enough to justify their own paragraph.
I-squared-S is an internal data transmission format that had been used in some CD audio circuitry, perhaps from the dawn of CD players. The now-defunct Audio Alchemy realized that they could use that digital format for connection of transports and DACs and avoid at least two digital-format conversions. All that was required was a different cable than common copper coax, glass optical or plastic optical. I-squared-S connections assign clocks and data to separate ordinary wires in a multi-wire cable. Assemblage has taken I-squared-S one step further with the I-squared-S-Enhanced connections, the best-sounding, least-cable-sensitive digital transmission Ive experienced so far. I-squared-S-Enhanced places the data on three separate mini-coax cables within a single cable. Your options for I-squared-S-Enhanced cables at this time appear to be limited to the Sonic Frontiers/Assemblage I-squared-S-Enhanced cable ($50) or to a SUN (the computer manufacturer) video monitor cable, which has three mini coax connectors and several pin connectors housed in a DB-25-size shell, the size used for most parallel ports on PCs. You cant use a parallel cable, however; you must use the three-mini-coax specialty cable. I used a SUN monitor cable and the Assemblage/Sonic Frontiers I-squared-S-Enhanced cable during the evaluation and found virtually no difference in sound quality. Both cables sounded better than any other copper cable I had on hand, including the $269 Cardas Lightning digital coax cable. As far as I know, the Assemblage DAC 3.0 (starting at just under $1500) and the Sonic Frontiers DACs (more expensive) are the only DACs currently supporting I-squared-S-Enhanced.
Front and back
The front of the D2D-1 is a mass of LEDs and buttons which allow you to choose, and then see from across the room, which of the various modes and connections are in use. LEDs for the digital input and sampling rate of the incoming signal reside on the left side of the faceplate, while LEDs to indicate that the dual PLLs are locked to the digital signal and the mode in use are on the right. The D2D-1 allows you to choose how data will be output. The Trans mode does no sample-rate conversion or interpolation to the data, but does still make use of the PLLs for reducing jitter. You would use this mode when you play HDCD or 24/96 software, or if you are using the D2D-1 with a non 24/96 DAC. The 48K mode upsamples the data to 48kHz, which is only a little greater than CD's resident 44.1kHz format, while 96K upsamples to a full 96kHz and interpolates to 24 bits.
Around back are the various inputs and outputs, some of which I've mentioned already. The D2D-1 can take a digital signal in BNC/coax, AES/EBU, TosLink or AT&T glass formats and output it to BNC/coax, AES/EBU, I-squared-S or I-squared-S-Enhanced. There's an IEC power-cord receptacle but no power switch -- when the unit is plugged in, it's on. Assemblage includes gold-plated BNC adapters with the D2D-1 for those of us with RCA-terminated coaxial digital cables.
I used the Cardas Lightning digital coax from the transport to the D2D-1 and Nordost Optix (a $99 video cable which happens to make a very good-sounding digital coax cable) from the D2D-1 to the Assemblage DAC 3.0. I listened to some favorite discs -- Squirrel Nut Zippers Hot [Mammoth Records MR-0137], Vivaldi Oboe Sonatas [Harmonia Mundi 907104], Daves True Story Sex Without Bodies [Chesky 96/24 Super Audio Disc] -- and others mentioned below using a female-to-female RCA bypass I fabricated to allow both digital cables to remain in the digital signal path when the D2D-1 was removed. This was necessary because bypassing either one of the digital cables to connect the transport directly to the DAC 3.0 caused as much or more change in sound quality than inserting or removing the D2D-1. This is one reason products like the D2D-1 sometimes get overly gushing reviews: The reviewer changed two things at one time, by adding or removing the jitter-reducing component and an extra digital cable, instead of just one. Im sure the female-to-female adapter I made has some sound of its own, but it is far less than removing one of the digital cables from the signal path. The use of this tool made the comparisons much easier to perform and confirmed that the D2D-1 was indeed doing something audible to the digital audio signal. I could listen to a track through the D2D-1, then bypass the D2D-1 by inserting the female-to-female adapter to connect the Cardas and Nordost cables.
The money paragraphs
All the cool technology and features of the D2D-1 arent enough to justify its existence. There must be a sonic payoff thats perceived to be commensurate with or greater than the selling price of the product would. At $699, I expect clear and obvious improvements in sound quality -- not necessarily huge improvements, but ones that I dont have to go crazy trying to listen for. The D2D-1 delivers about what Id expect for a $699 expenditure. It didnt blow me away or bring tears to my eyes. You arent going to read a bunch of lurid phrases designed to induce an obsessive longing to own a D2D-1 in this review. But still, I very much like what the D2D-1 does for digital audio and it will be very difficult, if not impossible, to give up the D2D-1, especially considering that it assists both music and movies via separate connections to DAC and surround processor.
The things that are most noticeable when using the D2D-1 are improvements in the spatial presentation, focus and bottom two bass octaves. Im not sure why the bottom octaves should improve this much with the D2D-1 since the low frequencies are the easiest, least-demanding frequencies to digitize in the first place. Bass frequencies are slow and lazy, and 44.1kHz is plenty of bandwidth for digitizing them. Yet I hear clear and rather obvious improvements way down there in a system configured to hear the bottom two octaves. The R.E. Designs SCPA 1 preamp and the pair of Vandersteen 2Wq subwoofers really did reveal what was happening down low. The Audible Illusions M3A tube preamp rolls off enough in the deep bass that the improvements down low were not nearly as obvious.
The two bottom octaves gain significant articulation and power, giving low-frequency sounds considerably more of a real feeling. In fact, some of the subsonic noises are so realistic youd swear that they were really being generated by physical activity in your house and not by a subwoofer. This is really something with movies that have low-frequency content. You swear your house is exploding or doors are slamming or that people are moving around in rooms above or below. To get this bass detail with DVD movies, you have to use the Trans mode, which bypasses the processing circuits and only de-jitters the signal before it is decoded by the surround processor.
Spatial improvements were of the type that allowed me, for the first time, to experience an indefinable satisfaction with the sound of classical music on CD. I enjoy classical music when performed live, but neither LP nor CD recordings satisfy in the same way. I dont expect the recordings to sound like live performances, and they dont. But I do expect that indefinable satisfaction to be there, and I never get it from recordings. Using my favorite performance of Beethovens Fifth Symphony [Carlos Kleiber, Vienna Philharmonic, Deutsche Grammophon 447400-2], a 1975 analog recording with typical precise-but-dry DG sound, along with several other recordings of Debussy and Vivaldi, I was able to experience that satisfaction factor on a very consistent basis with the D2D-1 in use. For the first time, when the orchestra works itself into a frenzy and pauses for an instant, nobody playing a note for half a measure or so, you can hear and feel the hall very clearly in the decaying sound. Without the D2D-1, this decay is truncated so quickly that the space seems small and closed in. Delineation and space between instruments improves also. So while the actual effect on the sound wrought by the D2D-1 may not be huge, this result of making classical music sound like it is recorded in an actual concert hall for the first time might be something that some people would consider a huge accomplishment.
The improvement in focus the D2D-1 brings is analogous to bringing an image into focus optically. Vocals and instruments shrink down in size which clears a little space in the soundstage. This gives a sense of openness and definition of location that previously was somewhat masked. I never thought my digital sound was less than properly focused in the past, yet the D2D-1 reveals clearly that digital audio focus can be improved upon, even if you dont notice it as a problem now.
I detected no change in tonality when using the D2D-1. If you are looking for a particular type of sound in your system, the D2D-1 will be something you can count on to not make fundamental changes. It outputs a sound that is tonally virtually identical to whatever you put into it. Its my experience that many listeners may think that a product isnt doing anything beneficial if there isnt a tonality change that is easy to notice in demos. Im glad that Sonic Frontiers/Assemblage didnt crank any tonal changes into the D2D-1. The result is more subtle but it will be much easier to live with over the long haul. So far, Id have to call the D2D-1 completely neutral-sounding. The improvements you hear from instrument to instrument arent tonal changes; they are small resolution improvements that make formerly homogenized details audible without manufacturing anything that shouldnt be there.
I heard about half the improvement made by the D2D-1 from the jitter removal alone, with the other half from the 44.1-to-96kHz and 16-to-24-bit conversion. The I-squared-S-Enhanced between the D2D-1 and the DAC 3.0 (coax from transport to D2D-1) produced perhaps another 10% improvement over coax-only connection. This was not a huge thing, but for $50 or less for the cable, it certainly was significant.
How does a D2D-1 sound when upsampling 16/44.1 discs compared to 24/96 discs and what, if anything does it do for 24/96-disc sound? Id peg the improvements you hear from upsampling 16/44.1 at about 50% of the difference between a 16/44.1 track and a real 24/96 version of the same track. Thanks to Classic Records demo discs, Ive been able to experience that difference. Personally, while I very much like the sound of 24/96 audio compared to standard CDs, the magnitude of difference is not large there either. When playing 24/96 discs, I did not have a clear preference as to whether to set the D2D-1 to 96kHz or to set it to Trans mode where it would de-jitter only. I did not spend a lot of time agonizing over that since I dont listen to 24/96 discs all that often compared to standard CDs, but I think I preferred the Trans unprocessed mode by a very small degree. There was a very slight difference in sound quality between the two modes, almost vanishingly small.
Bringin it all back home
The four functions performed by the D2D-1 add up to a worthwhile pile of goodness for your $699. It would have been a harder call if the D2D-1 had only half the features. But given that it can increase the sampling rate to 96kHz, increase data word length from 16 to 24 bits, attenuate jitter and output to multiple digital formats simultaneously, the D2D-1 is a high-value package. You wont find huge or massive improvements in sound quality, but you will find an honest improvement to both focus and spatial characteristics. In addition, the D2D-1 does something unexpected in the bottom two octaves; it makes sounds down there far more audible and far more detailed than you are used to hearing. This can be good when the sounds are intentional, but it also uncovers lots of uncorrelated ultra-low-frequency errors that exist in many recordings. If your system doesnt reproduce the bottom two octaves (20Hz to 80Hz) with authority, you may not notice the newfound capability.
If you have a 24/96 DAC, the D2D-1 is definitely worth considering. If you also have a 24/96 transport, the D2D-1 is worth thinking about really hard. And if you also need a way to feed both a high-quality stereo DAC and a surround processor and/or remote receiver, you probably already want a D2D-1 in the worst way. I can confirm that you would be quite happy with its addition to your system.
Copyright © 2000 SoundStage!
All Rights Reserved