Digital Yesterday: Steady Progression
When digital audio was new, you could pretty much chart the steady, linear progression of the technology for about a decade. From the first 14-bit multiplexed non-oversampling DACs in CD players in 1982, to the fully realized, 8x oversampling, 20 bit ladder DACs in the top DACs of the early 90s, there was clear and steady progress:
14 bit multiplexed D/A converters in CD players, no oversampling, brickwall filtering
The first 16 bit converters, still with no oversampling and brickwalled
16 bit converters with 4x oversampling, to eliminate the brickwall filter
Standalone DACs with 18 bit converters and 4x/8x upsampling
Standalone DACs with custom DSP filtering, 20 bit converters and 8x upsampling
And, along the way, you could chart the course in measurements. D/A converters got more linear, less noisy, and achieved higher performance by every measure. New versions of the old products performed better, because the multibit technology behind them was improving. Publications like Stereophile started measuring jitter, which raised awareness of its importance and led to jitter numbers steadily decreasing.
The result? By the early 1990s, it was possible to get 19+ bits of linearity out of multibit converters—a huge leap forward from the 13 or so bits of early CD players.
Progress wasn’t only made on the playback side, either. Mobile Fidelity contracted Mike Moffat (yes, our Mike Moffat) and Nelson Pass to create their GAIN system, an insane recording chain with a real 16-bit oven-controlled multibit DAC that output linear PCM with no missing codes up to 500kHz rates. This multi-chassis product took up almost a full equipment rack…but it was what was necessary to do good 16-bit ladder analog to digital conversion. Arguably, it still is.
Now, of course, there was only one problem with all of this progress: price.
Check the historic price of a PCM63 D/A converter, and you’ll quickly realize that it’s something that will never appear in an iPhone (nor would it fit.)
So, what to do? D/A chip manufacturers came to the rescue with products based on 1-bit sigma-delta modulation. These products were less expensive, easier to use, and more highly integrated. And they measured pretty well.
Another leap forward? In one way, yes. Without sigma-delta D/A converters, we wouldn’t have the wide range of DACs and ADCs we have today. Your smartphone has a DAC in it with specs we would have killed for in 1990. The analog to digital converter inside it may even output 24 bit samples, at higher sample rates than we would have ever imagined.
And we can’t underemphasize the impact of sigma-delta technology. It has allowed us to create more DACs (and ADCs) more inexpensively, with higher performance than we would have guessed, 20 years ago.
But we did lose something in all of this progress.
Digital Today: The Lost Decades
Today, it’s largely a sigma-delta world.
Recording. Most recording studios use analog to digital converters that employ A/D chips that use an intermediary multibit sigma-delta format before their PCM output. Note that this isn’t DSD. And note that even sigma-delta can have shades—single bit, multibit, etc.
Mixing. From there, the PCM output is mixed/mastered in PCM (pretty much all mixing and mastering is in PCM…yes, even recordings that end up as DSD.)
Playback. From there, it’s typically going to be compressed and downloaded or streamed to a player using a multibit sigma-delta D/A converter.
Or, in the case of some crazy audiophiles like us, it’s stored uncompressed, maybe even in high-res, before going to a DAC with a fancy multibit sigma-delta D/A converter.
Or, in a literal handful of cases, it might go to a true multibit R-2R converter, just like the old days. But that’s a fraction of a fraction of a percent.
“So, who cares what it is, I just want good sound!” you say.
And we agree! We’re far too wrapped up in formats. Take that format-proselytizing energy and aim it at the studios. Lobby them to produce better recordings. That will produce greater benefit than any format “regime change.”
But…here’s the deal (and here’s where we get philosophical.) In today’s sigma-delta world, we’ve lost something that we consider important: the original samples.
They’re destroyed by upsampling, they are destroyed by asynchronous sample rate conversion, they’re destroyed by sigma-delta D/A ICs. What you hear is an interpretation, a guess, at what the original content was (they don’t call them successive-approximation converters for nothing.)
“But this can’t possibly matter, it’s hard to measure the distortion of your typical ASRC, for example,” some will say.
Hard to measure doesn’t mean it isn’t there, we say.
Bottom line, it’s a mathematical fact that samples that have passed through a digital filter, an asynchronous sample rate converter, or a sigma-delta modulator are not retained. There is no closed-form solution to the math.
“And why should that matter to me?” you ask.
Maybe it doesn’t. Maybe the approximation is good enough.
But maybe it isn’t.
And this is where we get to the core of what Yggdrasil is about: what if we haven’t been hearing everything PCM is capable of, because we’ve been hearing it on delta-sigma technology that throws away the original samples?
Yeah. We know. We’re crazy.
And perhaps we are. Perhaps it will make no difference at all. Perhaps it won’t be important to anyone other than us. But the fact is: we have a solution to retain the reproduce the original samples, without the drawbacks of a non-oversampling design. It is in Yggdrasil. And we’ll see what you think, very, very soon.
And that is the absolute core of our digital philosophy: retaining the original samples, all the way through to the output.
“But, It Doesn’t Matter, Because…”
Because this position, this philosophy, is so counter to the currently accepted wisdom, I’ve prepared a quick discussion of possible objections to it, for your convenience.
“It doesn’t matter anyway, because everything comes from a delta-sigma ADC these days. Do you have any original bits at all?”
Actually, this isn’t entirely accurate. There are still multibit ADCs out there, though they are probably thin on the ground. There are also plenty of recordings made with multibit ADCs, including Mike’s GAIN system. They don’t disappear when new technology appears. And, you know what? Instead of being fatalistic and negative, we’d like to consider the best-case scenario: that we actually push PCM’s capabilities forward to the point where new multibit ADCs appear.
“But how can those old DACs possibly perform better than the best of today? They’re only 20/48. We have 32/768.”
Going from 16/44 to 20/44 actually makes more difference than anything else, when it comes to digital. Why? Although the Nyquist theorem says you can perfectly reconstruct a waveform from digital with 2X the sample rate, it assumes an infinite-bit ADC with no quantization error. The more levels, the less the quantization error. 16 bit = 65536 levels, 20 bit = 1048576 levels. 24 bits is 16 million+ levels, but nobody has ever achieved 24 bit linearity, period. The best DACs are about 19.5-20 bits, even after 20 years of “progress.” (Hence, “the lost decades.”) Higher sample rates are nice for analog filtering, but limit the amount of horsepower a digital filter can bring to bear…and it takes up more storage space. So that’s a tradeoff. And “32 bit?” LOLOLROFLCOPTER. There will never be any 32 bit music. Because physics.
Just IMO.
