DSD comparison to PCM.

[LynnOlson]

Thorsten discovered this blast from the past:
http://web.archive.org/web/20071011...tre.org/surround2002/technology/page_07.shtml

"Incapable of reproducing the same transient twice." Like the famous description of NTSC as "Never Twice the Same Colour", or Ford as "Fix Or Repair Daily".

 

[Ragnar]

New article out at Positive Feedback:

I suspect mid-grade DSD DACs mimic the modulation noise of 2-track 15ips analog tape, thanks to all that randomized distortion. Magazine reviewers trip off on that sound, and associate it with "analog".

That's the sound I hear when listening to DSD - analog tape modulation noise. Of course, I've only listened to Sony's idea of flagship SACD players and some inexpensive ones from Pioneer and OPPO. I'll have to beg or borrow a real $15K DSD dac so I can hear the truth.

I guess it's time to visit Bruce.

 

[Bruce B]

I guess it's time to visit Bruce.

Come on down!!

 

[ack]

New article out at Positive Feedback:
http://www.positive-feedback.com/Issue66/pcm_dsd.htm

I'm pretty sure that the proposed dither-noise spectra would change the sound of mid-to-upper-mid delta-sigma DACs in some kind of way (maybe worse, maybe better). Class D amplifiers improved quite a bit when spread-spectra modulation (octave-wide ultrasonic noise) replaced a single ultrasonic tone; I'm guessing that opamps would also respond to the same approach. If you gotta have slewing, better to spread it around evenly.

Of course, it's better not to have slewing, but that costs more.

P.S. The title is partly tongue-in-cheek. I don't think this technique is going to make a generic audiophile DAC sound like a Playback Designs with 128fs DSD source. No. But I do think it'll make PCM on a mid-grade DAC sound better ... well, more "analog-like", which can be taken several different ways.

I suspect mid-grade DSD DACs mimic the modulation noise of 2-track 15ips analog tape, thanks to all that randomized distortion. Magazine reviewers trip off on that sound, and associate it with "analog".

I just finished reading parts 1 and 2 on PFO - this is the most impressive audio technical write-up I have ever seen; I felt I was studying for college exams again, and it took me over 3 hrs. Thanks!

 

[LL21]

I just finished reading parts 1 and 2 on PFO - this is the most impressive audio technical write-up I have ever seen; I felt I was studying for college exams again, and it took me over 3 hrs. Thanks!

+1!...still reading it!

 

[LynnOlson]

I'm still learning this stuff myself, so don't be surprised if some errors have crept in. Thorsten brought my attention to Andreas Koch's FFT graph of DSD performance. He estimates it has 33 dB (!) of FFT Gain, presumably through very long averaging times. Although not shown in the graph, this degree of averaging would give conventional 24-bit PCM a noise floor of -177 dB, far below the noise floor of any analog electronics.

If we subtract the 33 dB of FFT Gain from the graph, the noise performance in the audio band starts to look like real-world DSD, and the noise at 1.4MHz rises to the -6dB figure you'd expect if the system dither is at the optimum 50% modulation level.

My takeaway from this is that DSD-narrow (conventional SACD and 64fs DSD-download) is surprisingly fragile. A recording made in DSD-narrow, converted to DSD-wide (Sonoma) or DXD (Pyramix), edited and mixed in that format, and then re-converted back to DSD-narrow, is going to suffer a substantial gain in noise levels. 3 dB? 6 dB? I'm not sure.

Every conversion into the DSD-narrow domain requires a re-dithering with the peak energy around 1.4MHz, for the simple reason that the noise is the carrier for the audio information in the 1-bit DSD system. No noise, no carrier, and no audio information. The 1.4MHz noise has to be there to modulate the DSD system.

By contrast, dither-noise linearizes PCM at the lowest levels, but it is not required for successful operation. The earliest (admittedly bad-sounding) digital recordings were made without any dither at all; the better-sounding ones had about -80 dB of noise from the analog master tape, which served the same purpose as dither (before the necessity of dither was discovered).

Thorsten also mentioned that some software-player vendors are digitally filtering DSD above 30 kHz with a 24 dB/octave lowpass filter before converting the DSD to PCM. This might be a mistake; the noise in DSD carries audio information, and perhaps should be left intact to be lowpass filtered after the physical conversion into the analog world.

Then again, with PCM, we have to worry about images from ultrasonic content, which is an argument for using the highest possible frequency if DSD is going to be converted into PCM. There are a lot of tricky issues that arise if DSD is converted into PCM, or vice versa, and these are made much worse if the PCM has a non-integer rate like 96 or 192 kHz. (In other words, if your software or hardware converter offers a selection of sample rates, select 88.2/24 or preferably 176.4/24, or best of all, full DXD at 352.8/32. A few PCM DACs support this rate through the USB 2.0 interface.)

Aside from this theoretical stuff, there's a big difference in the sonics of consumer-grade and pro-level DSD equipment. My guess is the pro stuff has better noise-shaping algorithms and analog electronics that are free of slewing artifacts, which gets rid of most (or all?) of the noise-modulation blur you hear in consumer gear.

My impression was the Resonessence Invicta was pretty close to the pro-grade Playback Designs ... but this was hardly an A/B comparison, separated by several months, 1500 miles, and two completely different systems. David's system is all-transistor with giant 800-watt MBL amps and MBL speakers, while my system has 20~30 watts of direct-heated triode power, and speakers that are about 8 to 10 dB more efficient than the MBL's.

 

[ack]

So do you have a conclusion wrt DSD-Narrow vs PCM? It seems to me this extreme noise shaping required for DSD and single-bit delta-sigma PCM, coupled with Putzey's DSD claim that it is in its place in mastering from analogue, renders them a flop for reproduction (we kinda knew that single-bit PCM is not good, ever since the mid-80s when those portable players hit the market), so the battle shifts to the new DSD-Wide vs multi-bit delta-sigma PCM or hi-rez PCM??? Other than Analog Devices' laser-trimmed ladders of yore, who else makes them today? I believe dCS's Vivaldi uses them and they are in fact discrete ladders?

 

[LynnOlson]

What I heard at David's place with the Playback Designs was very, very good. We only listened to 64fs and 128fs DSD-downloads; no SACD's, Red Book CD's, or high-rez PCM, so I can't say anything how the PB does with those sources. The deficiencies you hear from consumer-grade one-box SACD players (I have one myself) were not there.

There was a surprising and entirely unexpected improvement in 128fs DSD sources; remarkably, it sounded like the very best studio-grade PCM, or a direct mike-feed. 64fs DSD sounded more like a 1/4" 15ips 2nd-generation copy of a studio master tape ... very good, but not like a mike-feed. Both were a huge step up from consumer-grade SACD, which I don't find all that amazing.

Here in Colorado, I had access to many of the same tracks I heard at David's place in Portland, but I wasn't able to persuade Pure Music and the Invicta to play 128fs DSD tracks. But I was able to audition the Invicta on 64fs DSD, and it sounded a lot like the 64fs DSD tracks I heard at David's place, allowing for the big differences in systems. The Invicta definitely preferred DSD to PCM source, for what it's worth, and the sound quality from the little SD cards was as good or better than the USB connection to the MacBook Pro.

My takeaway from this is the ESS Sabre 9018 is a really good DSD converter, so long as the associated digital and analog electronics are top-flight. The usual murk-n-mush from delta-sigma converters attempting to play SACD's isn't there. Whatever ESS did with the noise-shaping algorithm, it appears to work. My guess is the ESS noise-shaping algorithm is dynamic, so it easily adjusts to different sources.

 

[Orb]

Here are Dustin Foreman's patents that are core to ESS sabre dacs.
http://patft.uspto.gov/netacgi/nph-...=1&f=S&l=50&Query=in/Dustin+and+Forman&d=PTXT

The last one on the patent list had me laughing though as it is so out of relation to electronics, industry risk of being such a techie I assume
Cheers
Orb

 

[Bruce B]

What I heard at David's place with the Playback Designs was very, very good. We only listened to 64fs and 128fs DSD-downloads; no SACD's, Red Book CD's, or high-rez PCM, so I can't say anything how the PB does with those sources. The deficiencies you hear from consumer-grade one-box SACD players (I have one myself) were not there.

There was a surprising and entirely unexpected improvement in 128fs DSD sources; remarkably, it sounded like the very best studio-grade PCM, or a direct mike-feed. 64fs DSD sounded more like a 1/4" 15ips 2nd-generation copy of a studio master tape ... very good, but not like a mike-feed. Both were a huge step up from consumer-grade SACD, which I don't find all that amazing.

This is what I've been saying for years about my PD MPS-5

 

[ack]

So what's the actual problem DSD (in its best incarnation) is trying to solve over hi-rez PCM?

 

[tailspn]

So what's the actual problem DSD (in its best incarnation) is trying to solve over hi-rez PCM?

Required decimation in order to convert to "hi-res" PCM. All current A/D converters have DSD front end Sigma-Delta modulators. The affect of decimation is reduced if the sampling rate ratios are kept small, but it's still easy to detect the difference from a raw DSD (2.8MHz) file, and one converted to DXD (352KHz) for processing.

 

[Ragnar]

So what's the actual problem DSD (in its best incarnation) is trying to solve over hi-rez PCM?

At the risk of sticking my neck out and perhaps going down in flames: I believe DSD was developed to create a problem (for DVDA), not solve one.

In the beginning, to many, DSD sounded better than Redbook CD, because very few companies were taking the steps necessary to improve d/a conversion sufficiently to bring out hi-rez PCM's sonic merrits (let alone Redbook CD), while Sony and Marantz were giving consumer SACD players their best shot. Over time, many brilliant engineers have worked to improve DSD performance, just as they have to hi-rez PCM. Even Redbook has gotten better (listen to a CD through a Berkeley Audio Design dac [shameless plug]) but few manufacturers have tackled the problem of hi-rez PCM for the reasons that Lynn has mentioned (cost.) And cost is the same issue for DSD to bring out its full potential.

As I paraphrased before: 'comes the resolution, all (hi-rez modes of) digital will sound the same.'

 

[Ragnar]

Required decimation in order to convert to "hi-res" PCM. All current A/D converters have DSD front end Sigma-Delta modulators. The affect of decimation is reduced if the sampling rate ratios are kept small, but it's still easy to detect the difference from a raw DSD (2.8MHz) file, and one converted to DXD (352KHz) for processing.

We might say all modern high performance sigma-delta A/D converters have DSD-wide front ends; conversion could go either way. It's just easier to process PCM, so why not choose that?

 

[ack]

I've always considered DSD a solution in need of a problem, but this decimation comment above... hmmm... can you elaborate, tailspn? and if PCM is easier to process as per ragnar, then how does decimation (in the DSD domain?) help again?

 

[Mike Lavigne]

This is what I've been saying for years about my PD MPS-5

+1.

 

[tailspn]

We might say all modern high performance sigma-delta A/D converters have DSD-wide front ends; conversion could go either way. It's just easier to process PCM, so why not choose that?

Because of the required decimation to convert to the lower sampling rate PCM. If the multiple Sigma-Delta converter loops that make it DSD-Wide are converted to 1-bit DSD at the same sampling rate, there's no decimation required. That's the way the E-Chips in a Sonoma operate. It's not the conversion at the same sampling rate that does the sonic damage, it's the decimation required to change the original sampling rate to a lower sampling rate. Decimation is simply a low pass filter, necessary so as to not exceed the Nyquist frequency of the new lower sampling rate. There are always losses when passing through a filter, digital or analog.

 

[tailspn]

I've always considered DSD a solution in need of a problem, but this decimation comment above... hmmm... can you elaborate, tailspn? and if PCM is easier to process as per ragnar, then how does decimation (in the DSD domain?) help again?

Point #2 under Figure 1. Decimation doesn't help. It's a necessary evil in the conversion process of any sampling rate to a lower sampling rate in either PCM or DSD, or one into the other.

 

[ack]

Point #2 under Figure 1. Decimation doesn't help. It's a necessary evil in the conversion process of any sampling rate to a lower sampling rate in either PCM or DSD, or one into the other.

I have a feeling you are talking about converting to RBCD? If so, I wasn't talking about that, I was talking about DSD vs hi-rez PCM, in which case we are not decimating the sampling rate, or are still... I am not sure I follow the decimation argument yet, and am not clear about which figure we are talking about either???

 

[tailspn]

I have a feeling you are talking about converting to RBCD? If so, I wasn't talking about that, I was talking about DSD vs hi-rez PCM, in which case we are not decimating the sampling rate, or are still... I am not sure I follow the decimation argument yet, and am not clear about which figure we are talking about either???

You'll have to do some research if you want to understand this. 64fs DSD has a sampling rate of 2.8MHz. DXD, the Pyramix PCM editing format has a sampling rate of 352KHz. That's a 8X reduction, and requires a filter to eliminate all energy above 176.4KHz so as not to be reflected back into the audio band due to Nyquist Theorem. RBCD sampling rate is 44.1KHz. That's a 64X reduction, and requires an even more drastic filter to eliminate all energy above 22KHz.

Decimation isn't an argument, it's the way this stuff works.

I'm in Boston also. Give me a call if you'd like to talk about it. I'm in the book. Tom Caulfield, Marshfield.