|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
73.249.178.108
In Reply to: RE: Upsampling making R2R obsolete?.....is this the revolution? posted by J. Phelan on April 08, 2016 at 15:04:16
Off loading the DSP has been common since the beginning. The early DAC chips didn't even have an onboard digital filter.
All the top CD players 15 years ago were using fancy DSPs in front of standard DAC chips to implement the reconstruction filter. Wadia had their DigiMaster + BB 1704. Simaudio and several others were using the Pacific Microsonics PM100/200 with a BB 1704. Audio Aero were using an Anagram algorithm on a SHARC DSP with an Analog Devices DAC. And the gold standard of digital playback for most of the last decade was EMM Labs, which upsampled Redbook to DSD. So the technology hasn't been changing much. Things have been getting better via refinements to existing approaches, no revolutions.
And 45 MHz is nothing special. Most parts support MCLK of 1024fs which is 42 MHz just for 44.1k Redbook. Some ESS DACs can take over a 100 MHz clock if you disable the internal oversampling and it seems that most people working with them are using MCLK in the 80-100 MHz range. No reason you couldn't go even higher, but you just increase the RF noise on your audio board.
Follow Ups:
EMM has been the standard -but they only sample at 5.6 Mhz.
Before their (brand new) DA-2, who else sampled at 45-100Mhz ?
36.864 MHz was a common maximum MCLK frequency a decade ago, which allows operating up to 768fs for a 48 KHz PCM input, 384fs for a 96 KHz input, etc. Some current gen Cirrus and AKM DACs have maximums in the 40-50 MHz range. For example, the CS4398 can operate at 256fs at 96 and 512fs at 192, which is 49.152 MHz. ESS is the only one I know of that goes beyond that.
Assuming you want to achieve the maximum SNR in the pass band, there is a tradeoff involved with increasing the frequency multiplier. Using a higher frequency multiplier allows for better noise shaping. As you know from SACD, 64fs requires very aggressive noise shaping to achieve good performance up to 20 KHz. When operating at 256fs, the noise shaping can be a lot more gentle and push the noise further away from the audio band. But as you increase the system clock frequency, it's harder to control the analog noise from the clock and harder to control jitter. For this reason, the best Redbook performance is often achieved at 192fs or 256fs.
..ok, but DACs -capable- is one thing, actual DSP-work is another.Besides Meitner and PS Audio, I found (Antelope) Zodiac Platinum. All upsample to Quad DSD (11.2Mhz).
It's hard to find DACs that do this, let alone one that processes to 45Mhz.
Edits: 04/11/16 04/11/16
Wadia was doing 64fs with a digital filter implemented by FPGA way back in 1988! That's the same rate as DSD. Krell and some others were doing the same by 1990.
By the time SACD launched in 1999, it was common to convert 44.1 KHz PCM at 256-512fs, even with R2R ladder DACs. That's 11.3 to 22.6 MHz, 4-8x higher than DSD at the time of SACD's introduction. A popular high end combination was the Burr-Brown DF1704 filter and PCM1704 DAC, which was released in 1998 and worked at 512fs for Redbook and 256fs for 96 KHz PCM.
And a decade ago, 768fs was becoming common, that's 33.87 MHz.
As released at 64fs in the SACD format, DSD was way behind the state of the art. It was limited by the storage inefficiency of the 1-bit format and trying to fit 8 channels (5.1 plus stereo) on DVD-based media. Quad DSD would have been more in line with the state of the art circa 2000, but it would have been stereo only.
These days, quad-DSD playback is still operating at a slower clock rate than modern PCM playback, but we're at a point of diminishing returns where increasing the clock rate further doesn't necessarily improve the filtering but may increase noise.
Dave thanks for the post. I found the whole thing incredibly informative.
Does the very high sampling/oversampling rate in PCM R2R dacs allow for a low order gentle filter like DSD?
Do you think there is anything intrinsically more linear about one format versus the other? I mean in reproducing a waveform, or does it just come down to the sound of filters?
Thanks,
------------------------------------------------------------
Big speakers and little amps blew my mind!
Does the very high sampling/oversampling rate in PCM R2R dacs allow for a low order gentle filter like DSD?
No for the digital filter, yes for the analog filter.
There is a gentle low order analog filter on the output of an oversampling DAC. It can be gentle because it's not the reconstruction filter. All it's doing is filtering ultrasonic noise. With some R2R DACs, you might not even need it.
But the digital reconstruction filter will still be a brick wall or close to it. Oversampling at higher rates and with longer filters allows for a more ideal filter response. It also allows for better noise shaping. But you're still stuck trying to fit a steep cut-off between 20-22 KHz regardless of how fast you oversample. Such is the nature of Redbook unfortunately, and there is no way around using brick wall filters without accepting aliasing distortion or rolling off below 20 KHz.
Same thing if you're converting 44.1 KHz PCM to DSD, the conversion will involve a brick wall filter.
If you're sourcing DSD from DSD or analog or high sample rate PCM, then you can avoid it.
Do you think there is anything intrinsically more linear about one format versus the other? I mean in reproducing a waveform, or does it just come down to the sound of filters?
Are you comparing typical oversampled Redbook playback vs. Redbook upsampled to DSD? Or R2R vs. delta-sigma? Or more generally PCM vs. DSD?
Thanks for the response.
Hi res PCM vs DSD.
One tenth of an octave (20 to 22kHz) would require one hell of a brickwall filter. JA mentioned in another post that DSD provides an improvement over PCM in time domain errors. I suppose that could be called a form of linearity. Nelson Pass published a paper that has some graphs showing how steep filters for speakers obliterate the phase coherence of a speaker. My limited knowledge leads me to think that phase and time domain response are probably closely tied? Your opinion? I have used both gentle and extreme slopes with speakers, and neither is perfect or perfectly terrible. Just different with different tradeoffs.
Thanks again for answering my basic questions.
------------------------------------------------------
Big speakers and little amps blew my mind!
I suppose he is referring to the ringing impulse responses of the anti-aliasing and reconstruction filters, which is particularly bad for Redbook due to the need for steep filter responses. For higher sample rate PCM, 88.2 KHz and above, the filters do not need to be steep. And even if the filters are made to be steep, there isn't much musical content up at 44 KHz or above to incite ringing, and the ringing would be ultrasonic anyway. So in my opinion, it is impossible to make Redbook anti-aliasing and reconstructions filters truly transparent, but it should be easy to do that for PCM at 88.2 KHz and above.
I enjoy both DSD and hi-res PCM, but I think DSD 64 (the SACD format) is a little bit compromised relative to hi-res PCM. Going forward, I could be happy with either hi-res PCM or higher rate DSD. I think that a DSD-only DAC operating at 256fs could be superior to hi-res PCM due to simpler implementation, but at the expense of storage space and download bandwidth.
I found the following Audiostream interview with Mr. T. Of special interest was the following;
"If we convert from 24-Bit at 352.8kHz (DXD-PCM) to 1-bit at 2.822MHz (DSD) - we need to throw away around 99.96% of the amplitude information the PCM format is capable of, while we are only having 12.5% of the time domain information that the DSD system is capable of. If we convert to DSD from DXD, that is 1-bit at 2.822MHz to 24-Bit at 352.8kHz - we need to throw away 87.5% of the time domain information of DSD, though we can theoretically remap all of this into the amplitude domain. So in effect we get the worst of both formats, rather than the best of one."
Maybe higher rate DSD would even things some, but I find it interesting. Also very few albums for sale at 352kHz. I would bet Sony thinks they can remap into the amplitude domain, or they wouldn't have put it forward as an archival format, but then again maybe they would. I have read on a forum for recording engineers where they say DSD was meant to be converted to PCM, but PCM was not meant to be converted to DSD. They certainly seem to do it though.
-------------------------------------------------------------------
Big speakers and little amps blew my mind!
Thorsten's numbers are misleading because he's equating timing resolution with sample rate and amplitude resolution with sample size. I believe he knows better but is just trying to oversimplify for a non-technical audience.
Ultimately, amplitude resolution and time resolution are characteristics of the encoded waveform, not its encoding format. Amplitude and time resolution are limited by the digital noise spectra. So for example, 1-bit DSD has better amplitude resolution than 16-bit PCM, and 352.4 KHz 24-bit PCM certainly has better timing resolution than DSD (even 192 KHz PCM probably does).
Using delta-sigma modulation with appropriate noise shaping, one can trade sample size vs. sample rate without loss of resolution. That's how delta-sigma DACs work.
Very informative reading here.
nt
I didn't know this. But in recent years, co. were not running as high. Even DCS upsampling-devices were processing at DSD (2.8 Mhz).I guess these earlier rates were overkill. And so much for 'HQPlayer'- it looked like a breakthrough.
In your view, what was the biggest bottle-neck of Red Book ? The whole act of conversion ? Is Equibit the way to go ? (Still has a clock, but now controlling the output stage). Is this a better way to clock ?
Edits: 04/11/16 04/11/16
DCS is sort of an odd duck in that they use a proprietary 5-bit discrete ring DAC operating at 64fs, which is quite slow by modern standards.
As I was saying before, higher clock rates are pretty much standard fare. At the present time, 768fs (33.9 MHz) is typical for those using current CS, TI/BB, AKM, and Wolfson delta-sigma DACs. ESS is 80 MHz.
For older equipment, here is a link that has a nice summary of clock rates:
http://www.trichordresearch.co.uk/cd-player-list/
In my opinion, the biggest problem with Redbook is the sample rate is too low, which means the anti-aliasing and reconstruction filters will never be fully sonically transparent. Worse yet, the anti-aliasing and reconstruction filter responses overlap and interact because they're both occupying the same frequency space, which may be why some filters sound better with certain recordings but not others. The low sample rate guarantees that there will never be any consensus "best" reconstruction filter. As long as the format is around, I think we're doomed to go on iterating and tweaking different filter designs.
Post a Followup:
FAQ |
Post a Message! |
Forgot Password? |
|
||||||||||||||
|
This post is made possible by the generous support of people like you and our sponsors: