Preface: After I wrote these comments I found an article by Dan Schwartz at the abso!ute sound that very closely mirrors my own experiences with high resolution digital, calling it "a third kind of sound". I differ from Schwartz in that I find vinyl's distortions to be too gross to be appealing. His point about each generation having it's own sonic frame of reference is spot on however. Mine was previously CD.
I should also mention that my DAC-3 is upgraded above and beyond the regular DAC-3 upgrade kit since I've changed to Burr Brown's highest "K" grade PCM1704 DAC chips and highest "B" grade OPA627 op amps, ultra-fast HexFRED rectifiers in all 3 raw power supply sections, and high-performance Linear Technology LT1086 and LT1033 voltage regulators in both analog supplies. I've also added extra chassis dampening, etc.
Hi Chris, Eddie, Glenn & Peter,
Here are some listening notes on the D2D-1 jitter
reducer/upsampler/interpolator. Please feel free to publish my
comments with or without attribution.
I connected the low-jitter, separate-clocked I2Se cable from the D2D-1 to my parts upgraded Assemblage DAC-3. My Pioneer DV-414 has both Toslink and S/PDIF outputs so I was able to roughly A/B test the bypassed upsampler by leaving the DAC S/PDIF input connected via BNC from the DVD player and sending the (jittery) Toslink through the D2D-1. With the D2D in Transparent mode (no interpolation or upsampling) jitter reduction alone over Toslink versus non-reduced BNC was very noticeable. Much less grain and hash on the music, i.e. cleaner. Better instrumental and voice timbres. A very noticeable and good improvement.
Unexpected to me, when I sent both the BNC and Toslink from the Pioneer to the D2D-1 (i. e., both jitter reduced) the Toslink sounded much worse than the BNC. Electrical as opposed to optical S/PDIF transceivers are less jittery, and the BNC resulted in a much cleaner sound than Toslink optical with *both* sent through the D2D/I2Se. The difference was smaller than the jitter reduction of the D2D but I was quite surprised to hear the fairly sizeable difference between Toslink and BNC on the D2D inputs. Why this was unexpected is explained later.
The D2D-1 has no optical outputs (wisely since they're high jitter) so I was unable to compare Toslink from the D2D to DAC, but sending the Toslink around the D2D directly from the transport to the DAC-3 sounded terrible! This would be the highest jitter configuration and it sounded really hashy and mushy in comparison to the lowest jitter BNC + D2D + I2Se configuration.
So both the D2D jitter reduction only and BNC seem significant. The difference between BNC and Toslink is a lot less noticeable without the D2D (i.e. direct from DV-414 to DAC-3) so this is another way to confirm that the D2D in transparent mode (no interpolation or upsampling, only jitter reduction) improves the signal significantly, making the differences between BNC and Toslink more apparent. Another way to say this is that the D2D improves resolution enough just through jitter improvement to make the jitter differences of the different transmission methods far more apparent and audible.
Part of the reason for my surprise at the differences between Toslink and BNC was that I had hoped the jitter reduction would have improved the Toslink so that its high jitter relative to copper was inaudible, but it didn't. Perhaps the jitter reduction on the output of the D2D due to the dual PLLs and I2Se interface is of a larger magnitude than the jitter reduction of the inputs. The input jitter reduction is significant and large, but it's not infinite, so starting with the lowest jitter input appears to help significantly. (I wonder if you can get more improvement by daisy chaining D2Ds?)
I also tried the 24-bit 48k interpolation/upsampling. It does change the sound to be softer yet more detailed, but strangely it also shifts the imaging/reverberation around so the original spaces don't quite sound the same. Individual instruments/voices are noticeably more pleasing and distinctive and resolution is noticeably enhanced, but ambience seems slightly sacrificed or at least changed from the original 44.1 x 16. On the other hand the clarity and improved emotional impact of the upsampled and interpolated 48kHz makes it more fun to listen to. All this is with the PMD-100 (HDCD) filter in place which maxes out at 48kHz.
44.1 HDCDs with the D2D (i.e. jitter reduction only) sounded *much* more stable and quiet than without it. Perhaps the jitter is more noticeable with these effectively 20-bit, higher resolution recordings. It would seem logical that a higher resolution recording would be more affected by jitter, and indeed that seems very strongly the case. In other words the jitter improvement with HDCDs (and no interpolation) seems larger than for regular CDs without interpolation. Acoustic instruments sound smaller, more distinct and a lot less bloated. Once could say the air around the instruments was much tighter and cleaner. The improvement of HDCDs over regular CDs was much more noticeable using the D2D. It's worth mentioning that the HDCDs I used were mostly Reference Recordings made by Keith Johnson which are probably recorded with extremely good original fidelity and resolution. In other words, better reproducing chains can do more with better recordings.
I might wish there were a 88.2 output option and upsampling selectable separately from interpolation but I don't know if the Crystal 8420 will do that and it would certainly make a more complex box. In other words I'd like to know the difference between only upsampling or only interpolation. Perhaps it's not possible to meaningfully separate the operations, for example if added samples must also be interpolated. I'm afraid the DAC-4 will need to be a complex microprocessor controlled box to really take advantage of the many features. And I do think it would be best if all the features were integrated into a single box. Yes this would make the marketing more challenging and harder to "part out" into different boxes like DAC vs UBJ vs D2D.
Changing to the Burr Brown DF1704 96k filter and listening to Chesky and Classic Records 96 x 24 DVDs in transparent mode (jitter reduction only), 96 x 24 through the D2D is a heck of a lot smoother than without. Sound is cleaner and more listenable, less hashy yet also less hard; more "analog" and less "digital". Adding the D2D clears up a large amount of haze and fuzz and makes sounds rounder, fatter and more whole dimensional. The sound on Chesky's Sara K Hobo DVD is deeper, cleaner, quieter, more stable, spacious and full bodied. More analog if you will. It sounds very much like a high quality analog master tape... if you took away tape's noise, static, dropouts, wow & flutter, speed variations, and frequency perturbations and limitations. :-) The obvious conclusion is that high resolution digital can be better than the best purely analog media, if used properly.
In 96k output mode it's not clear if the Crystal 8420 is modifying a 96k input signal, but it does perhaps sound a bit different from transparent mode. I preferred transparent for 96 x 24. Changing from "Transparent" to "96k" mode (probably interpolated?) seemed a little cloudier versus transparent handling of 96 x 24 input.
CDs upsampled and interpolated to 96 x 24 were startling. Very different sounding from merely jitter reduced 44 x 16. Softer, clearer, cleaner and heck of a lot more emotionally involving. I almost couldn't help singing along with Joe Walsh, Don Henley and pals on The Eagles' Hell Freezes Over. A lot more instrumental and vocal details and character could be picked out and easily identified. 96 x 24 interpolation wrought out similar improvements on a number of CDs. The ones reasonably well recorded sounded much cleaner, with better timbral accuracy, depth, reverberation, space, clarity, etc. If I had to pick a single adjective it would be clean. The best news is that a sizeable portion of my existing CD collection is made much cleaner, as if a large layer of grunge were removed.
(However, grunge music is also much cleaner! Several guitar riffs on Pearl Jam's Evenflow were more clearly Jimi's Voodoo Child reinterpreted. For that matter the European Polydor HDCD (immensely better than the U.S. MCA release) of Jimi Hendrix Ultimate Experience was almost painful to listen to for all the hackles on the back of the neck it raised, a la the goose pimple effect on overdrive. This was without HDCD decoding but with 96 x 24 interpolation. I'd really like to hear it with HDCD decoding at 96 x 24; see next.)
The Burr-Brown DF1704 currently used for 96 x 24 filtering in the DAC-3 (and DAC-2.6) to me does not sound as smooth or sophisticated as the PMD-100 when both are run at 44k or 48k, so we'll have to wait for the PMD-200 input board for the DAC-3 before we can try the HDCD filter set on either natural or interpolated 96 x 24 signals. I'm looking forward to that. (I'm referring of course to the HDCD's generally better implemented filters and not the HDCD resolution enhancement process which is presumably CD-specific.) Chris and designer Peter Schut mentions the DAC-3 will need a new input board for the PMD-200. Given that all my circuit improvements are on the power supply and converter/output boards, I'm definitely going to get one with all upside.
The sound of fully interpolated and upsampled CDs is definitely more pleasing, but I have a nagging wonder if it is any more accurate. Can harmonics lost in the original 44 x 16 analog-to-digital conversion be resurrected this way? To me information theory says that information once lost should be gone forever, lost to quantizing noise and recording input anti-aliasing filters. The sound is definitely sweeter and more involving but I had some doubts as to whether it's more accurate or realistic. On the other hand it's an awful lot of fun to listen through.
Update #1: I'm listening to the Bainbridge Mozart Piano Concerto #13. This is a very well recorded 1987 CD using the highly linear 4 channel 16-bit Colossus recorder and minimal miking. In an age of bad digital recordings it was my benchmark for how well they could be done, though on some equipment it can occasionally sound bright and harsh. Over-steely string sound in particular is a good indicator of system harshness.
Over Toslink directly from the Pioneer to the DAC-3 it's unbearably hashy, grainy and fuzzy (this is after listening at 96k for a while). Sent over BNC and I2Se, with D2D providing jitter reduction at 44.1k (no upsampling), it's immensely cleaner though still a bit grainy. Upsampled and interpolated to 96k the images and soundstage do shift a bit, but the sound is much cleaner.
The magic of the performance comes through clearly. Instrumental timbres are great. String sound is a touch bright in places probably due to the mikes, but overall it's really pleasant and realistic. It's hard not to get involved in the delicate beauty of the slow movement especially when Jeremy Menuhin speaks through the piano unaccompanied. Crowd noise from this live recording is also a bit bright but extremely realistically rendered. The rolled-off off-axis response of the presumably cardioid-set Neumann TLM 170i main mikes is strikingly clear when the audience applauds at the end of the concerto. One can clearly make out individual hands clapping near the mikes and pretty far into the rest of the hall. The sonic character of the hand claps are spot on in being rolled off and phasey, bearing in mind they are probably in or near the null of the mikes. The point here if course is that the outstanding D2D and DAC-3 playback technology lets a great recording come through in all it's beauty and detail. I don't think I've ever heard the behavior of microphones reproduced this well.
I'm still concerned about the apparent shift in sound space when interpolating, but it's clear the interpolation and/or upsampling can do some great things, especially on good recordings. One important mechanism I didn't mention earlier is that raising the sampling rate moves the aliasing noise further from the audio band and allows for either more benign filtering and/or simply moves the effects of filtering further from the recorded harmonics, either of which should be good for sonics. In other words, even if the information content is not actually increased by upsampling, or higher original sampling rates do not actually record more audible information, moving the filters much higher in frequency gets their effects that much further away from the music which has got to be a good thing. Upsampled audio does sound significantly cleaner and more open, though recordings initially made with more than 16 bits still seem to have greater depth, space and timbral information. In short, recording with more bits seems good, with possibly diminishing returns at more than 20. The jury may be out on higher sampling frequencies recording more useable information, but the resulting benefits of reduced filtering effects are likely significant.
Update #2 on the shifting space/image issue: downsampling a 96 x 24 DVDs to 48k results in a flatter, more "in your face" presentation, very much the opposite of the change in space when upsampling CDs to 96k. Interestingly this also gave the DVD a more "CD-like" sound in that it was harder and more "digital" in the sense of over-bright and too present. Changing to the lower sampling rate makes the sound closer in perspective, less subtle, less natural, in a smaller space, and much more similar to 44k CD listening. So the opening of spaces caused by upsampling may not be unnatural, or at least the sample rate conversion process seems to have a roughly symmetrical effect up and down. All the results of upsampling are in the direction of better sound and the opposite is also true of downsampling. This makes me a lot more comfortable saying the spatial improvements from upsampling are not necessarily artificial side effects and may be meaningful. My initial reaction to upsampling not sounding as accurate may simply have been due to my familiarity with 44 x 16 audio. The lower resolution CD format, with all its flaws, had become my reference. 96k is better, upsampled or native.
My recommendation would be to not even think about a DAC-3 without a D2D-1. The improvement is wonderful. After the pure 96x24 DVDs, I didn't think CDs could sound this good. Either interpolated or just jitter-reduced, the improvement D2D-1 makes on CDs is very large.
I haven't really finished my comments on the DAC-3 alone, and I'll need to rethink their presentation a bit. It makes more sense to discuss the system as a whole (i.e. D2D and I2Se and DAC-3), but I haven't decided how to integrate that content with these comments.
I spotted another couple occasional operational glitches. When changing rapidly between S.R.C. modes the 96k mode occasionally collapses the sound into nearly mono and certainly lower bit length. I'm not sure I can explain why this might happen without a logic analyzer, yet it does on occasion. Selecting another mode fixes it. It's almost as if the SRC chip is getting confused and resorting to a very low resolution fallback mode.
Also after playing a 96 x 24 DVD on 96k mode and not changing any settings trying a CD next left audio muted on a couple occasions. This also happens on power up occasionally if one selects an upsampled mode possibly before the D2D is fully booted or before an input signal is asserted. Again this can be cleared by cycling through SRC modes, but it is a minor operational bug.
Also the I2Se cable I got from you last year had a ferrite choke on one end which I assume should go on the load end to block RF picked up in the cable from entering the DAC. You may want to make a note about this if you feel it's important. Perhaps the current cables have chokes on both ends as I've seen on some computer cables.
Thanks for another great product. Together these guys really form an awesome and synergistic digital system. Thanks for making life a lot more enjoyable!
Jeff C.