"What makes the most difference in high resolution digital recording systems--the bit depth or the sampling rate?" That was the question that I wanted answered as I entered the Digital Conversion Systems (dCS) demonstration room at the 100th Audio Engineering Society last fall (1997) in New York. dCS makes high-end digital converters for classical recording and has recently made a splash with their 192 kiloHertz converters, choosing to skip just a doubling of the current 48 kHz sampling rate (96 kHz) and go straight to a quadrupling of the smaple rate. Many of their reasons for 192 kHz sampling are psychoacoustic and invove the localization of sound sources, dealing with the time delay between the ears and how our brains use that information to properly ascertain positioning within the stereo sound field.
The tests were conducted in a small room (hotel room size) and I listened from a 1/2" analog source (a recording of a small jazz ensemble), through the dCS converters (A/D then D/A), and we started at a resolution of 24-bits at 192 kHz sampling and went down from there. As I sat down to listen, I made some surprising discoveries. Here are my observations.
1) 24/192- It sounded very good, not identical to the original but close. (I was to find out later that this had less to do with the sampling frequency (Fs) than with the "sound" of the dCS converters.) It was definitely within the acceptable range. I heard no more difference than I would expect to hear comparing an analog 1/2" playback to the source mix bus from the console. The digitized signal was definitely missing the characteristic digital side-effects that to which many of us are accustomed.
2) 24/96- My concern here was primarily to see if there was a big difference between the 96k and 192k Fs. There was very little difference to my ear between 96k and 192k. I heard the same differences between the analog source and the digitized return that I had heard at 192k. This surprised me somewhat, because I thought the difference would be subtle but audible, even under the hotel room conditions we were listening in. Let me say here that I was looking for noticable differences, not microscopic differences. The conditions were not conducive to absolute scrutiny, but that wasn't what I was going for.
3) 24/48- Now I began to hear a big difference. The imaging and the top end got very strange. I will go into that phenomenon later.
4) 16/96- At this juncture, I went ahead and asked the dCS personnel whether the Fs or the bit depth made a greater difference in the sound. Their response was that the sampling frequency made a far greater difference. I doubted this, because I had heard the progression from 16-bit to 18-bit to 20-bit converters and knew how much difference those made. So I asked them to prove it. We listened to the original source material followed by the digitized version at 24/48, then back to the analog source, followed by 16/96. I was SHOCKED. The difference between the source and the 24/48 was clearly noticable. The difference between the source and the 16/96 was MUCH more subtle. The strangeness on the top end that was so evident at 24/48 was almost completely gone at 16/96. So much for my theory.
4) 16/44.1- After that revelation, I skipped past the 16/48 comparison and went straight to the CD standard, 16-bits at 44.1 kHz. This was the most horrible revelation of all. The original analog source was played to re-establish the benchmark. Then we switched to the 16/44.1 digitized version. It was DISGUSTING, to put it politely. The wash and sway (how about those technical words) on the imaging was HORRIFIC. And the overall sound was extremely undesirable. The biggest revelation was the top end. There was a definite comb-filtering effect that cause everything above 10K to swim randomly, causing sonic localization to go out the window. I couldn't define it, but I've noticed it many times before. I've never been able to A/B it and pin down the cause. This was the biggest revelation of the whole AES trip. I came away determined that the CD standard, which I had always thought was inadequate, was simply unusable for master recordings. The grief that I felt over all the incredible music that had been recorded this way and all that was forever lost between the original performances and the 16/48 or 44.1 masters was overwhelming. I wanted to go back. (Don't we all?)
I took some consolation in the fact that my most glorious works of the 1980s and 90s were all recorded on analog tape and simply mixed to digital, most to DAT using external converters. At least I can still go back and reclaim the resolution that is there on the original masters.
The other major revelation that this listening test proved to me is that after all these years of recording with digital and all the research and money that has gone into digital recording, and with $20,000 worth of the dCS converters and stable external clocks and even more money in workstations, and 24-bit, 192 kHz converters and Dual Nagra recorders, we have finally accomplished a digital reproduction chain that sounds "as good as analog." Almost seems a little silly doesn't it.
The dCS people took great pride in one thing about their demonstration though. As they were A/B'ing their digitally converted output to the analog input, and without indicating which was which, they would ask the listeners to identify which playback was the analog source and which was the digitized version. Most people thought the dCS converter chain sounded more "analog" than the analog, which always elicited big grins on the faces of the dCS staff. I found that interesting, but laughable. I look forward to the day when I can sit down and perform the same test and A/B the analog source with the digital return and shrug my shoulders and say I can't tell. To me, that is the ultimate goal.
3D Audio Inc