Not So Bit Perfect About DSD vs. PCM? Part III
Happy Fourth of July! I got up early and headed to the recreation area of my little city for the annual Will Rogers 5 and 10K race…with Charlie, the family dog (border collie). I haven’t been running much over the past 6 months due to the pneumonia thing and recent achy knee, but I figured what the heck. Charlie and I stayed at the back of the pack with the strollers and kids and just coasted through the 3.1 miles. It was a beautiful morning and Charlie does manage to make a bunch of friends.
This is Part III of a series of posts that corrects some inaccuracies found in the blog post at BitPerfect regarding click here“>DSD vs. PCM. Richard, the author of the lengthy post, was somewhat critical of my position that 96 kHz/24-bit PCM audio provides better fidelity than DSD 64 (we don’t even talk about the difficulties doing any work with 1-bit DSD). His technical argument seemed flawed at first read. He makes some statements that either aren’t relevant or simply wrong. It’s troubling that he spends 8000 words reinforcing his position with information and “facts” about digital conversion (both DSD and PCM) that when examined in great detail by an acknowledged expert prove problematic. Here’s the next installment:
BitPerfect: “If we confine ourselves purely to digital anti-aliasing filters used in a SDM-ADC, a slope of 300 dB/8ve inevitably requires an ‘elliptic’ filter. Whole books have been devoted to elliptic filters, so I shall confine myself to saying that these filters have rather ugly phase responses. In principle they also have a degree of pass-band ripple, but I am willing to stipulate to an argument that such ripple is practically inaudible. The phase argument is another matter, though. Although conventional wisdom has it that phase distortion is inaudible, there is an increasing body of anecdotal evidence that suggests the opposite is the case. One of the core pillars of Meridian’s recent MQA initiative is based on the assumed superiority of “minimum phase” filter architectures, for example.”
John Siau Wrong again! The slope is not 300dB per octave (for the reasons stated in yesterday’s post). Also the filter is almost always a linear-phase filter and not an elliptical filter. This claim is completely erroneous.
Dr. AIX There are filters involved in the conversions process in both DSD and PCM. The use of anti-aliasing filters in 96 kHz/24-bit PCM systems doesn’t require “brickwall” filters as stated in the blog or “elliptic” filter either. How or why could Richard include elements in his argument that are incorrect. Perhaps the knowledge base of those reading the post would simply take it on faith and thus his position could be taken as true. As for the recent MQA Meridian invention by Robert Stuart, he does advocate for higher sample rates to close the gap on timing differences but I’m not yet convinced that higher than 96 or 192 kHz really gets us anything…except much larger files.
BitPerfect: “By increasing the sample rate of PCM we can actually reduce the aggression required of our anti-aliasing filters. I have written a previous post on this subject, but the bottom line is that only at sample rates above the 8Fs family (352.8/384kHz) can anti-aliasing filters be implemented with sufficiently low phase distortion. And Dr. AIX poo-poohs even 24/352.8 (aka ‘DXD’) as a credible format for high-end audio. Here at BitPerfect we are persuaded by the notion that the sound of digital audio is actually the sound of the anti-aliasing filters that are necessary for its existence, and that the characteristic that predominantly governs this is their phase response.”
John Siau This phase-response argument is only applicable to analog filters used in front on non-oversampling A/D converters. It does not apply to oversampled (SDM) A/D converters. Ironically, it is DSD that has a phase-response issue.
Dr. AIX I can’t say I recall “pooh poohing” “DXD” otherwise known as 352.8 KHz/24-bit PCM without the anti-aliasing filters. I have questioned the misleading name because the DSD community didn’t want their “dirty little secret” – that 1-bit DSD requires the use of ultra high sample rate PCM – to pollute the purity of DSD encoding. With very few exceptions, the DSD files and SACD discs that are sold are has tasted PCM.
Just this morning, I received the new Blue Coast Music Group Newsletter. They announced a relationship with Mack Avenue Records and will be selling a variety of formats of some great new music. But they state, “Recorded at Avatar Studios in New York City by renown engineer, Joe Ferla, the project was initially recorded to 96 kHz/24-bit PCM WAV. It is our opinion these will offer the audiophile the best sonic experience. We have converted the files to DSD and FLAC for your convenience and performance on your listening system.”
They don’t mention that by converting and offering the source 96 kHz/24-bit PCM files to DSD (64 and 128), they’ve substantially increased the price. A FLAC version of the original 96/24 master is $15…converted to DSD 128, it will set you back $30. Stick with the PCM…get the “best sonic experience” and save some cash. Is having the DSD converted copy worth twice as much?
BitPerfect: “PCM requires an anti-aliasing filter, whereas DSD does not (actually, strictly speaking it does, but it is such a gentle filter that you could not with any kind of a straight face describe it as a ‘brick-wall’ filter). DSD has no inherent phase distortion resulting from a required filter. Instead, it has ultrasonic noise, and this is where Dr. AIX’s argument encounters difficulties. The simple solution is to filter it out. However, if we read his post, he seems to think that no such filtering is used…I quote: ‘It’s supposed to be out of the audio band but there is no ‘audio band’ for your playback equipment’. Seriously? All it calls for is a filter similar to PCM’s ‘anti-aliasing’ filter, except not nearly as rigorous in its requirements.”
John Siau The requirements of this filter are nearly identical to the requirements for the anti-aliasing filter in a 96 kHz PCM system. The difference is that the DSD post filter must be implemented in the analog domain and this means that the DSD post filter will cause phase distortion.
Dr. AIX If, as Richard claims, the ultrasonic noise (with no inherent phase distortion) is imply filtered out, then why is still present whenever I do a spectral analysis on DSD files. Theoretically, the noise shouldn’t be there…but it is. John’s response seems to provide the answer. I’m analyzing the digital files prior to the application of the analog filter…the one that causes phase distortion. I suppose there are circumstances where the ultrasonic frequencies are filtered out but in my experience and in talking to other engineers, there have been problems with the ultrasonic noise.
Another post that is way too long. I apologize…but I feel it’s important to establish who’s doing the spinning here.
Dr Waldrip, I agree with your comments, but be kinder to Blue Coast. That company believes that the original digital format is the best version. While Blue Coast greatly prefers DSD, if it sells an original PCM it says: “It is our opinion, these will offer the audiophile the best sonic experience.” This is consistent and is how Blue Coast sees provenance. Please don’t criticize them for that. Yes, Blue Coast doesn’t see that “tape to DSD” is an issue. So it goes.
I think Cookie produces and engineers great sounding recordings…she and I have very similar production techniques. However, she endorses special treatments for discs that I believe are “snake oil”, which troubles me and she believes in charging more for conversions to DSD than the original PCM…that’s the bump I had with the new additions to here download site.
As I read these assertions by bitperfect and the rebuttal arguments I am left feeling quite puzzled. This doesn’t seem to be simply a matter of opinion. It would appear that two audio engineers are capable of having diametrically opposite views of many of these issues. How can the mere lay person sort these things out? Is there no way to hold people accountable for mistatements of fact? And where do these “totally erroneous” claims written in completely confident and assured fashion come from? Many of these contradictions would seem to be easy to settle by those who are familiar with the field. It’s troubling.
You shouldn’t be puzzled. Some of the statements by Richard at BitPerfect are confusing (deliberately?), wrong, or uniformed. I very pleased that John Siau can bring clarity and accurate information to the current state-of-the-art in digital audio conversion.
There is no way that audio enthusiasts can hold individuals or companies accountable except by refusing to purchase their products…or by shedding light on the issues.
Of course Stewart should be puzzled! From the non-technical person’s point of view, this is like watching two debating teams debate two sides of a philosophical issue, when he hasn’t studied the issue. Both sides sound persuasive, and the viewer has no way or ability to ascertain the veracity of the points raised. The viewer ends up choosing between sides mainly on his inner emotional sway, or keeping count of the logical point-scoring score of each team (which is really a measure of debating skill more than a measure of truth).
If, like Stewart says, he wants to choose the technology that actually is best, then Watching Big Brains Debate is not getting him closer to his answer. He asks if there isn’t some way for him to reach finality.
About anti-aliasing filters: I thought these were used in the RECORDING process, that the filters used at the playback end to remove switching artifacts are not ‘anti-aliasing’ filters per se. Can you enlighten me on this?
There are filters involved at several stages in the digital audio production process. We call them “anti-aliasing” filters when used on the front end of the process to avoid “foldover” or “aliasing” and reconstruction filters in the D to A end of the process. You’re right the filters at the reconstruction end are not “anti-aliasing: filters. They are used to ensure that no frequencies above the Nyquist Frequency are produced to the analog outputs.
Remind me please: what *audible* consequence is there if we ignore the alias and leave it in the signal?
If frequencies higher than Fx/2 (the Nyqyuist frequency are encoded they appear in the audio band and are not harmonically related to the music. They might be quite low in level but they are to be avoided…thus the filters.
Hi Mark,
These extra long posts are the best ones! I always enjoy the extra juicy posts with John Siau chiming in with such effortless clarity, to the core arguments and hard facts.
Keep it up!
I bet BitPerfect must bit by bit perfectly debunked 😉
I wonder if you can trust anything, that this guy at BitPerfect writes ;-(
Richard is an excellent engineer and makes a solid playback engine. But as we’ve seen he sometimes misses some essential facts in the defense of DSD…or at least in his misrepresentation of PCM.
Thanks – I always though of BitPerfect as a nice piece of software.
Nothing in what I’ve been posting says anything about BitPerfect being anything less a nice piece of player software. I would opt for Amarra myself. The point is that a very knowledgeable person is spinning and being less than truthful with regards to the so called advantages of DSD over PCM. I sincerely hope he recognizes where he’s pulling the wool over people’s eyes.
I got the Blue Coast newsletter also, but when I followed the link for the Mack Avenue downloads, I found 2 statements that give me reason for concern about the FLAC downloads on the site.
The provenance statement for the Christian Mcbride Trio album states that the DSD and FLAC files are “second generation” and that a “Blue Coast conversion method” is used. Here is the full statement.
“Provenance: Recorded and Mixed to 96kHz, 24-bit WAV PCM. The 9624 WAV files (9624 is our shorthand for 96kHz, 24-bit encoding) are the original digital file generation received from the artist or label. The DSD and FLAC files are considered second generation and made from conversions using our Blue Coast conversion methods.”
Second generation seems right for the DSD conversion since there will be some differences in a WAV to DSD conversion, but the bitstream encoded in the FLAC conversion file should be bit-identical to the WAV file. This gave me concern about the “Blue Coast conversion method”. I couldn’t find any information about the conversion method used by Blue Coast, but I did find this statement on the site.
“Digital audio compressed by FLAC’s algorithm can typically be reduced to 50–60% of its original size, and decompressed into a near identical copy of the original audio data.”
“Near identical copy” is the concern here. Is it possible that the FLAC conversion used by Blue Coast is not actually lossless?
Due to this concern I am considering purchasing this album from HDTracks instead of Blue Coast, even though it costs more on HDTracks.
Mark…I can’t speak to the technical processes used at Blue Coast’s studio. However, I do know Cookie uses a lot of analog stages in her productions. She mixes from here multitrack source (analog tape or DSD 64) through an analog console, processes using PCM reverb, and mixes to a Korg DSD recorder (2.8 and 5.6 MHz). If the files from Mack Avenue were originally recorded at 96 kHz/24-bit PCM as they state, then the FLAC file by definition needs to decompress into exactly the same bits as the WAV original. If it doesn’t, they’re doing something wrong.
To charge more for the second generation DSD copies is hard for me to fathom. Buy the FLAC and decompress it, get the best sound, and save 50%.
I did buy the FLAC, but I got it from HDTracks. An Audition analysis shows that it is a real High-Def recording. I don’t bother to decompress FLAC files to WAV. Any DAC that has problems decoding FLAC in real time doesn’t deserve to exist in today’s world. It doesn’t take that much processing to decompress FLAC. In many cases real time decompression can be faster and more efficient than working with the decompressed data. I did spend a fair amount of time in the 70s and 80s working on compression and decompression algorithms.
I’m sure the 96 kHz/24-bit source recording is the best you can get.