Audio Asylum Thread Printer
Get a view of an entire thread on one page
|For Sale Ads|
In Reply to: RE: Best processors -but best tech ? posted by J. Phelan on February 21, 2017 at 11:04:08
DACs are probably the most complex consumer stereo component. As such, it is exceedingly difficult to assign technical cause to subjective listening result. On top of that, digital signal processing is a non-intuitive technology. Concepts like sampling itself, dither, and noise-shaping are just a few which are commonly misunderstood. What intuitively makes sense with analog often is incorect with respect to digital. For example, analog processing typically exhibits increasing distortion with increasing level, while digital processing typically exhibits decreasing distortion with increaseing level. For another, adding noise (dither) reduces digital distortion. In addition, audiophiles often believe that simpler is necessarily better, so, more digital processing is necessarily worse than less digital processing. Except, that isn't necessarily, or even typically, true with digital.
There's little that seems 'natural' about a digitally sampled system in the first place. More digital processing can be better than less, and often is, but it depends on the implementation details. R-2R (only one kind of multibit), or some other multibit converter technology isn't inherently superior to 1-bit sigma-delta, it depends on the details. 1-bit sigma-delta converters inherently exhibit perfect linearity, but also exhibit very high quantization noise. Full resolution multibit converters are the opposite, they exhibit low quantization noise but less than perfect linearity. Multibit sigma-delta is a hybrid of the two, intended to bring together the benefits of each technology while minimizing their drawbacks.
It would be convienent if we could determine a laundry list of technologies which would ensure our obtaining subjectively satisfying sound from a given prospective component, but I know of no such list as yet.
I agree with your post. Many of the assumptions made about computer audio are just copy throughs from data processing technology and not related to sound quality, electronics or acoustics considerations.
Thanks for saying that delta/sigmas have very high quantization noise. As you know, quantization is a cornerstone of digital audio.
Look at the specs -the Burr-Brown 1704 measured much better than delta/sigmas. But delta/sigmas invaded high-end audio. Partially due to availability. But then engineers *ignoring the specs* and marching forward with cheaper chips.
Why start off on the wrong foot ?
Delta/sigma replaced ladder chips because of cost. R2R chips became very expensive and Delta/Sigma are very cheap
But also look at the distortion specs. for PCM1704, especially at lower signal levels, such -60dB. The distortion is greater than it is for a modern sigma-delta converter.
As far as the engineers, don't be too hard on them. They get driven by market pressure too. Those that pay the bills get to call the shots - customers. The market wanted lower cost converters still having good distortion specs. Noise-shaped sigma-delta modulation (SDM) converters typically exhibit much lower audio band distortion than do full resolution multibit converters.
Full resolution multibit can also made to have low distortion, but typically requires time consuming costly trimming, possible by hand, depending on the design target. In addition, the greater the number of bits, the more difficult is achieving a commensurate reduction in distortion. Therefore, 24-bit sample size further accelerated the issue. SDM enabled low audio distortion conversion at a low chip cost despite it's other evident drawbacks. I believe that was the primary reason for the market takeover by SDM converters. Today, the market is dominated by hybrid multibit sigma-delta converters.
In high-end products, I don't see any 'cheaper' processors.
Delta/sigmas have much higher (measured) noise and distortion and that's final. Lynn Olsen's reports on PFO show this...
Believe what you will, but that's factually wrong. An PCM1794A, or an AD1955, or the Sabre chips, etc. produces lower in-band noise and distortion, especially at low signal levels, than does any legacy mass production full resolution multbit converter, including the PCM1704. Read the various datasheets for yourself, unless you believe they all lie. The SDM converters do have greater out-of-band noise - meaning, measurable yet inaudible ultrasonic noise. That said, the multibit converters do have their objective advantages. Subjective sound assessments, of course, vary.
BTW, Because some high-end DAC box is costly usually has little relationship to how costly is the DAC chip being utilized.
If you believe cheap chips (which discard data) are superior than pure-resistors, that's your business.
Data sheets mean nothing -they cover up and highlight what they want. It's independent study that counts.
"Mountains and Fog" -2 part article at PFO and "R2R/multibit vs. Delta-Sigma" at Head-Fi are good places to start.
As noise is (slowly) dropping in digital processors, we're hearing differences in DAC design. Could we have done this in the 1980s and 90s ? I doubt it...
If you believe so robustly on what the web says about dac chips, then there is not much to be said.
It is not chip technology but package implemetation, including the digital filter, I/V, analog stages, and digital input conditioning that most affect SQ.
Well, then, there's no sense in my trying any further to open an closed mind.
Post a Followup:
Post a Message!
This post is made possible by the generous support of people like you and our sponsors: