First, let's look at a common misunderstanding: upsampling is not oversampling. In the case of the latter process, data is multiplied, sometimes with interpolation algorithms operating to provide data that looks more complete and can be better decoded to analog; however, that data is still going into a 44/16 DAC, so a good deal of the oversampling is, in my view, wasted, even if there are are audible benefits from such interpolation algorithms as Pioneer's Legato Link process.
With upsampling we are interpolating data in a similar fashion, but probably with more sophisticated predictive algorithms, and that data resolution is kept, not truncated in the conversion process. The question is asked by skeptics: "How can you increase resolution by multiplying the sampling rate and adding bits?" And the second question is often an assertion: "You can't possibly make a CD sound better by upsampling to 24/96!"
The answer requires a bit of technical stuff. According to a 1997 white paper by Michael J. Story of dCS, the originator of upsampling techniques in 1998, the problem of the CD standard starts with the sharp anti-aliasing filtering required to keep frequencies above 20 kHz out of the audio output signal.
In general, anti-aliasing filtering needs to be quite vigorous - typically from some fraction of a dB down at 20 kHz to -100dB or so at half sampling frequency (22.05 kHz for CDs). Sharp filtering causes a ringing transient response...The effect is well known and unavoidable, and tends to be dismissed as a mathematical irritation with no audible effect - because the frequency response is flat."
The ringing contains energy, and we can plot energy against time. For anti-aliasing filters we get the sort of shape shown in figure 3. This shows that although the energy in the input transient is concentrated at one time, the energy from the anti- alias filter is spread over a much longer time - the audio picture is `defocused'. We might be tempted to argue that the energy is ultrasonic, but this is certainly not the case at 44.1 or 48 kS/s - our bandwidth constraints mean that to get good anti-aliasing, we must filter as fast as we can, and only pass the audio bandwidth. Ergo - any energy in the output signal is in the audio band. At sample rates above the standard, the energy in the ring still has the full bandwidth of the passband - math tells us so. We can also note that the energy in the ringing is large - for a sharp filter it can be 12% (-9 dB) of the energy in the main lobe."
To paraphrase Story, the brick wall filtering that has been the main preoccupation of digital designers since the early 80s has consequences that cannot be overcome at current sampling rates, especially in the decode/playback process. At that point in time (1997) he was arguing simply for higher sampling rates, but a year later was actively experimenting with upsampling, something Steve Lee (then dCS and Nagra distributor for the US) and I talked about extensively when recording in Bellingham in the summer of 98. Further, the CD standard has the problem of inadequate transient response that causes time smear that is audible. When the transient parts of a musical sound that has wide bandwidth are not reproduced together, the result is sound that is not natural, because things are not all happening at the right time. This leads to the listening fatigue that was characteristic of early digital. It's gotten much better, but the only way to get rid of the ringing and consequent smearing of energy is to move the filtering process well out of the audible band. And the only way to do this is to double, or even quadruple the sampling rate.
Now, upsampling cannot eliminate the problems of the filtering done in the encode/record process, but it can deal with those in the decode/playback final stage, in the process moving the filtering up to 44.1 or 48 kHz from 22. Also, with the clever interpolation programs developed by dCS and, more recently by other companies, data can be added in a way that makes the 96- kHz DAC able to wring the greatest amount of detail from the incoming signal while maintaining transient speed and accuracy.
And this brings us to the Assemblage D2D-1, which passes digital signals, but processes them in a number of ways. According to the manual, it is "really 4 products in one! The D2D-1 is a state-of-the-art Sample Rate Converter (Up- Sampler), Jitter Attenuator, Data Word Length Interpolator, and a Digital Transmission Format Converter."
There are 5 digital inputs: AES/EBU coaxial, S/PDIF coaxial via BNC jacks (gold-plated RCA adaptors are supplied), Toslink plastic fibre optic, and ST glass fibre optic. Any of these will appear at all 4 outputs, whose formats are AES/EBU, S/PDIF (BNC again), I2S (an input is found on the Assemblage 3.0 DAC), and I2Senhanced, "an actively buffered and expanded version of I2S", (also found on the DAC 3.0 and the Sonic Frontiers Processor 3).
A Crystal CS8414 input receiver is the first of two phase-locked loops, the second (both have lock LEDS) a discrete PLL with custom made voltage control oscillators; it is in the latter circuit where the jitter is reduced to a claimed .02 picoseconds.
There are 3 output modes: TRANS passes the signal through, including HDCD information, or you can choose to output 48 or 96 kHz signals; 48K upsamples and interpolates to 24 bit; and 96K does the same at this frequency. HDCD coding will not be passed in the latter two modes, though the signal is processed and passed through. Not only is this a useful box to maintain and improve data integrity, it's a valuable professional tool for moving data around safely from format to format; and then there's the upsampling.
I was also supplied an Assemblage DAC 2.6 with a 96/24 daughter board replacing the standard Pacific Microsonics HDCD filter/decoder chip. That chip was also provided, and is easy to swap in and out. That was good, because the 96/24 chipset was DOA out of the box. When I tried the HDCD one the decoder worked fine, though, understandably, not at 96 kHz. The Parts Connection kindly sent me another 96/24 IC by courier, and we were in business.
Or so I thought. The refitted decoder worked well for a couple of days, but the DAC 2.6 was running disturbingly hot, and one minute it was working and then not, on the third day. I checked the fuse and it was definitely blown, and I opted to not replace it because of the obvious thermal problems. I'll come back to the DAC 2.6 presently.
Luckily, I also had on hand the MSB Link DAC III, which is also 96/24. MSB has been promising internal retrofit upsampling boards for months, but they haven't delivered yet. The Link III can also be retrofitted for Virtual Dolby Surround. Our review sample also included the Nelson Upgrade, which adds a balanced AES/EBU input, and an improved power supply. A future upgrade is said to be 192 kHz decoding, but as little program material is available, I'm not exactly anxious about this one, though having recorded and heard 192 kHz, I could get interested if it takes off.
In the here and now I spent several afternoons listening to various CDs through this combination, some of the time also including my Meridian 518 in the chain to see if it would make things even better, or perhaps worse. The 518 dejitters and allows various noise-shaping algorithms, also allowing emphasis encoding and decoding, and like the D2D-1 does bit interpolation up to 24 bits. All this starts to sound like a digital daisy chain...hey, I could have added another jitter box (a Monarchy DIP was at hand), but this seemed enough firepower to do the job.
As it turned out, the 518 was a valuable addition, especially when playing our own CDs, some of which are pre- emphasizedto increase low-level resolution at high frequencies). I started with one of these, Awake My Heart with the Bell'Arte Singers, using de-emphasis in the 518 and outputting 24-bit words with Noise Shape C. However, the listening started with the direct output of the 1- bit DAC of the Pioneer PDR-05 CD recorder. When I switched to the upsampled version it was like turning an audio light on in the room. Yes, it was fairly subtle to a casual listener, but distinct when I switched back and forth: there was less glare in the midrange, better deep bass (especially the ambient foundation of the church), more natural sibilance, and a better developed soundstage both laterally and front-to-back. One could also hear a better sense of dynamics, with less congestion in loud vocal passages. At the lower end of the dynamic scale there seemed to be more contrast, a deeper silence in the background that allowed the decay of notes to be more readily heard. Everything just sounded more realistic.
Next I listened to our newly released Bellingham Sessions 2, with the memory of hearing comparison of 44.1, 96, and 192 kHz reproduction in my head. It was my impression that the upsampled 44.1-to-96 didn't have quite the vitality and snap of the original, but was quite an improvementover the straight Pioneer DAC's output. Percussion was more delicate and sweet, and other musical details got sorted out better, as the Brits say. With the Link III DAC sitting on top of my Pioneer/HHB D9601 96 kHz DAT recorder, another option occurred to me: why not feed it the AES/EBU output of the D2D-1 (the S/PDIF coax fed the Link). Then all I had to do was put a DAT into the deck and put it into record standby. Voila...another 96 kHz DAC, albeit a 16-bit one. And even throwing away 8 bits, I thought the D9601 sounded a little sweeter and more detailed than the Link III, though it was very close.
Numerous other CDs followed, including Jennifer Warnes' The Hunter (fabulous deep bass on all the percussion), Kind of Blue (the CBS Mastersound version: the high hat cymbals on So What just came alive!), and quite a few others. Another Miles Davis CD, Some Day My Prince Will Come (regular CD reissue) always sounded mediocre, the muted trumpet quite distorted. Well, the upsampled playback was free of much of these ugly harmonics and non-harmonic digital grunge, literally proving how audible aliasing artifacts can be (but you could still hear the analog tape saturation, in fact, hear it more clearly!). I always glance at my 1/3d octave analyzer when playing Miles' recordings just to see the bouncing lights going right up to 20 kHz. Another CD that I have in both CD and LP, Maazel's Mahler 4th with the Vienna Philharmonic, demonstrated that though the upsampling process brought the sound quality closer to that of the LP, real 96 kHz (or Sony's SACD/DSD process) would be needed to close the gap.
I also listened to several tracks from our Test & Reference CD (yes, #2 is coming this fall), including Loon's Tunes, and heard things in surround sound through the Cantares SSP-1 (Fall 99) that I'd never heard before. The upsampling brought out subtleties in the wind and rain sounds, and made the echoes of the loon calls off the rock face much more distinct. The thunder got deeper and more powerful.
Overall, I find 96/24 upsampling to be the most exciting thing in digital audio this year, because it allows our existing CD libraries to be much better appreciated, and in this case at a relatively modest cost. It could be said, I suppose, that doubling the sampling rate doubles your pleasure. And having a 96 kHz DAC doubles your fun by allowing you to listen to the Chesky and Classic 96/24 audio DVDs (see SuperSounds).
I'll say more about the excellent Link DAC III with its upsampling board next issue, but for the time being I'm going to live with the D2D-1 in my system.
Just after I'd finished writing all this, another DAC 2.6 arrived via UPS, so I made it another daisy in the chain (actually, I replaced the Link III with it). The heart of the 96 kHz conversion is the new Burr-Brown PCM1704 chip. In the past, this company's DAC chips have been highly regarded, but I found myself not entirely happy with what I was hearing. It sounded quite forward and bright, not as detailed as the 9601, and seemed to lack the sophistication of sound of the Link III. I swapped these two back and forth a few times to compare their sound character, comparing both with the Pioneer DAT's DAC. It seemed I always ended up with the D9601, with the Link coming in second.
When I voiced my thoughts on this to Sonic Frontiers/Parts Connection honcho Chris Johnson, his reply was to send me yet another DAC 2.6, this time containing the Signature Parts Upgrade Kit (SPUK) and the Turbo Mod. Here's how Chris described these upgrades in a subsequent E-mail: "The SPUK contains custom MultiCap signal path capacitors (12 pcs.); 10 Caddock non-inductive foil resistors (in signal path); 6 Sanyo OS-CON ultra low ESR/ESL power supply regulator output caps; 4 IRF Hexfred ultra fast, soft recovery rectifying diodes; 4 WIMA power supply bypass caps; Kimber 19 awg KCAG pure silver output wiring; 4 LINEAR TECHNOLOGY ultra low impedance, high current power supply regulators; 2 pcs. of AD811 I-V stage op amp upgrade.
The TURBO mod replaces the "other" 4 op amps in the unit, which are currently very good Burr-Brown types (OPA134), with the very best op amps, by any manufacturer, the Burr-Brown OPA627 ...these retail for over $15 US each (about 4 times more)! Measured S/N improves by 8 to 9 dB through their use!!!!! FYI, even the $16,000 US Mark-Levinson No. 30.6 doesn't even use the OPA627...they use 134 and the 604 op amps, which are a fraction of the price, and discernably lower in performance, both empirically, and subjectively."
You know, you gotta like a DAC with SPUK, especially TURBO SPUK, and I did. I don't think I've ever heard such a demonstration of the importance of parts in a digital component before. It was the same DAC with the same Burr-Brown PCM1704 heart, but the lung, liver, and whatever else transplant created a completely new sound: more alive, more vibrant, more detailed, and more dynamic. I put the Link III back into the comparison loop, and the Turbo SPUK DAC 2.6 clearly eclipsed it with both fed by the D2D-1. Even the DAC in the D9601 seemed to pale by comparison. I think that The Parts Connection should simply forget about the regular DAC 2.6 and just sell the upgraded version for 96/24 use.
I suspect that the main reason for the better sound lies with the analog output: the improved Burr-Brown OPA627 op amps. But the Kimber silver wiring out of them couldn't hurt, either. To sum up, I've now heard a level of digital reproduction from CDs that I didn't think possible via these two little boxes from Assemblage.
Upsampling Update On The MSB Link DAC III
I was finally able to get my hands on a Link III with the upsampling board installed, this time a Half Nelson, which includes the internal parts upgrade, but not the AES/EBU input. I was able to set it up for level-matched comparisons with the Assemblage D2D/DAC 2.6 combo, these fed and level matched via the Meridian 518, while the Link was getting the same signal from the Toslink optical out of the Pioneer PDR-05 CD transport.
This should have given the Meridian/Assemblage system a sonic advantage, but on some program material I heard the opposite, the Link III's upsampling seeming to give the music a bit more life and micro-dynamic subtlety on Canadian guitarist Ray Montford's new CD, One Step Closer (about which you'll read more in these pages; it's a wonderful acoustic guitar program of original music). Much of the time it was a tossup, for example, on orchestral music. Both got the textures right and most of the details, with a very un-digital kind of atmosphere.
Unfortunately, I received the Link III too close to deadline to spend more than a few hours with it before writing, but I can say that it is least the equal of the Assemblage, both using pretty much the same digital goodies inside, and both built with great attention to parts quality and circuit design. And finally, I was struck by how close both came to the fabulous dCS gear sonically. Great digital audio has never been so affordable.
http://www.audio-ideas.com/reviews/digital-sources/msb-link-iii.html