Upsampling or Oversampling? Letters
Bob Katz on upsampling
Editor: In January of the first year of the third millennium, John Atkinson reviewed some new DVD-As that had been originally recorded at 44.1 or 48kHz and then upsampled to 88.2 or 96kHz, and stated that there would be no potential advantage to doing that. Actually, there is significant sonic advantage to remastering and reissuing entire catalogs that were originally recorded at 16-bit/44.1kHz.
Here's a summary of why even those recordings that do not contain high-frequency information above ~20kHz can sound much better if reissued at the higher sample rate and word length:
• The digital processors available to master recordings have improved significantly since many older recordings were originally produced. For example, my recording of Clark Terry Live at the Village Gate (Chesky) was succeeded one year later by Clark Terry: The Second Set. Both recordings were mastered from the same original set of DAT tapes—just different songs from the same night. In one year, Chesky's dithering, DSP, equalization, and other digital processing had improved so much that the second recording sounds warmer, more open, clearer, and more dynamic—all from improved digital processing. And this was in the early '90s. Time has not stood still, and the digital processors of the year 2000 are significantly better than most of the mastering tools we used in 1990.
• According to papers written by Dr. James (Andy) Moorer of Sonic Solutions, and verified independently by myself and many other engineers, there is considerable advantage to double-sampled processing. Even if the original frequency range is not extended past 20kHz, the quantization distortion is considerably reduced—because this distortion is now spread over twice the frequency range, and the human ear really only perceives the distortion below about 20kHz.
Quantization distortion is a deadly digital artifact that produces edgy sound. Instead of "natural" harmonics that are integer-related to the musical frequencies, it comprises enharmonic products of the musical source beating against the sample rate itself. When the sample rate is doubled prior to DSP processing, enharmonic distortion is greatly reduced, and the sound becomes significantly more open and clear. This is true even if the source was originally 16-bit. It should also be noted that the DSP processing inevitably produces a longer output word length, which is meaningful.
• If the sample rate remains at double the source rate, and the word length is left at 24 bits, then it will not be necessary to add another generation of sound-veiling 16-bit dither at the end of the process.
Skipping the degrading dithering step to produce the DVD-A is another often-overlooked advantage. Thus, consumers should not scoff at DVDs that have been digitally remastered from original 16/44.1 sources. They will be getting real, audiophile-quality sonic value in their remasters.
• Next, I note that you have been exploring upsamplers. Of the commercial units you have reviewed, to the best of my knowledge, all of these samplers and upsampling converters, except for the dCS 972 and Purcell models, are based on the Crystal Semiconductor CS 8420. This chip is an asynchronous converter. It has excellent sound, but its imaging suffers compared with that of a synchronous converter because it is continually estimating the incoming sample rate in order to reduce incoming jitter. And the filtering in an asynchronous converter can never be as ideal as that in a synchronous converter.
I have done the listening tests, and, under identical jitter conditions, recordings that have been upsampled via a good-quality synchronous converter sound about "two points" better, with a wider, more stable soundstage and imaging. Of course, some people may prefer the sound of the asynchronous converter on some recordings, because the vagueness of imaging is aesthetically pleasing (shades of vinyl?). Professionals who upsample will generally avoid asynchronous converters because we want to get those last two points of sonic performance.
• Last, if I may put in a commercial plug, a new processor is now available that is a genuine breakthrough in DSP technology: the Digital Domain Model DD-2 Ambience Recovery Processor, for which I have applied for a patent in 2001. This new process can literally restore and enhance older 16- and 24-bit recordings that were perhaps made with inferior A/D converters or less-than-ideal mike techniques or rooms (happens to the best of us!). It is now possible to enhance and recover the lost ambience in recordings, remastering them to standards befitting a 24-bit world. Of course, it's never as good as if it were recorded at 24/96 in the first place, but it's a big, big step forward.
Hope all of this clears up what's been happening on the professional recording front. Best wishes to Stereophile magazine for a happy and successful new millennium.—Bob Katz, Digital Domain, firstname.lastname@example.org
Didn't that other Bob say it first?
Editor: First of all, I'd like to say I have enjoyed Stereophile for nearly 10 years now. This is the first time I've taken the opportunity to write, though.
After having a conversation with a dealer regarding upsampling, I read JA's December 2000 "As We See It" once again. I kept having the feeling I had heard these words before. I went through some old audio mags and came across the September 1998 issue of [the now defunct] Fi magazine. In a "Guest Editorial," Robert Harley seems to have said basically what JA said concerning the improvement one hears being due to better digital filtering than just upsampling to 24/96.
Now, to my admittedly limited technical knowledge of the inner workings of digital audio, it seems the biggest difference in what was said in both articles is that Mr. Harley said this over two years ago, after hearing one particular component.—Jay Willingham, Macon, GA, email@example.com
I can't recall the Harley article, Mr. Willingham, but I am sure you are right. That both Bob and I independently reached the same conclusion is an indication that we are probably correct. As David Rich of The Audio Critic, I think it was, said many years ago, "in the next century, all audiophiles will be listening to will be different digital filters."—John Atkinson