ONLINE - ISSUE 25
Déjà vu All Over Again: PFO's Lynn
Olson Jousts with PFO Reader John Phelan On DSD, SACD, PCM, & All
Drawings by Bruce Walker
[PFO Senior Assistant Editor Lynn Olson recently had an exchange with PFO reader John Phelan on DSD, SACD, and Red Book PCM that I believe merits wider dissemination. PFO has always been a "creative forum for the audio arts," and this is precisely the sort of discourse I enjoy publishing. John's comments appear in standard font, Lynn's in italics.]
I should make it clear that I agree with Lynn Olson, and disagree quite completely with John on the subject of the quality of Red Book PCM, and especially with his characterization of DSD, which I view as flawed. Regarding PCM, I will say that, properly handled, Red Book can be made less unpleasant—but I would never put it in the pantheon of the true high fidelity formats: DSD/SACD, LP, and open reel tape. Nevertheless, John's points stirred up a fine interchange, as you'll see when you read it.
By the way, I should note that the text has been lightly edited for typos and continuity.
– Ye Olde Editor]
I have been perplexed in recent months by the number of people (you included) who still believe that DSD is superior to Red Book digital—it is not. But first, I shall critique a few of your thoughts.
"There were a lot of good recordings made from the early 1950s to the early 1980s". A very popular position among audiophiles, but on close inspection, not really true. First of all, these are OLD recordings. The problems of tape hiss, transfer losses and most of all, old microphones that introduced a severe resonant-peak in the lower treble. Then there is the miking technique. The Classics (mostly) used three symmetrically-placed microphones, which is NOT theoretically-correct miking. Three spaced mikes introduce phase anomalies which cannot be corrected later—any talented recording engineer knows this. These early miking techniques are well below that of Blumlein or its close (purist-miked) cousins.
"(Many) people despise Red Book". Who? Do these people know how to best play-it back? I doubt it. We have (read: many) newer recordings from Hyperion, Harmonia Mundi, Chesky, DG, EMI Classics and Nonesuch that will positively SMOKE those Old Master recordings (if transferred on to a CDR by "Reality Track" then played-back through a good Red Book player). If these things are done, you will finally hear what 16/44 can do—amazing!! As of this writing, a "good" player will still set you back a cool $6000-$10,000 (better, if you go higher—like the Reimyo player at 16K). In your piece, you say that "we will need ultra-low jitter players"—we have them!! In my view, Red Book is superior to SACD—when done right.
"Compressed audio is no good". Wrong on all counts—or at least a futile statement. Compressed audio, played-back in a lossless format (such as Apple's) are completely acceptable in the high-end circles. Even MP3's "replacement"—Ogg Vorbis sounds good. More importantly, we don't have to compress!! Many audiophiles are now storing their music collections on hard-disc in UNCOMPRESSED form. The very first audiophile hard drive was the Lynn Kivor—which (symbolically) offered uncompressed playback. And I agree that hard-disc can be superior to optical-based systems—it works for Red Book, can we say the same for SACD?
To learn more about the travesties of the DSD recording system, I highly recommend reading Peter Moncrieff's (long) treatises on this and DVD-A on his website, http://IAR-80.com. Scroll down to the bottom of the main page and you will see the sections on digital—read all of them (if you're serious about learning). If not Moncrieff, try Stereophile's Rich Report from 2000 or Robert Greene's essays in The Absolute Sound from 2001 for more info on why DSD is a bad idea.
Final thought. It's how a format sounds at home to the end user, not how it sounds to the recording engineer at the console is what really matters. It is a well known fact that DSD can sound good on Master but it falls FAR SHORT of expectations at home—even under ideal playback conditions. It's up to you to find out why (hint—it's not the lack of digital outs).
John, I've had the privilege of auditioning laboratory-grade, studio, and very advanced non-commercial prototypes for the last thirty-some years. To this day, the finest—and most realistic—sound I've ever heard was at BBC research labs in 1975, when I heard a first-generation quadraphonic mastertape recording of Beethoven's 9th played on a Studer 1/2" machine (without Dolby processing), and auditioned on custom BBC monitor speakers. I have never heard anything before or since that has come close to that. I have the somewhat contrarian stand that commercially available (audiophile) gear is far short of what exists in laboratories, high-end recording studios, and the homes of top-class DIY builders.
Before I worked for Tektronix in the Spectrum Analyzer business group (which sold almost entirely into the Mil-Spec market), I had a more cheerful view of the hifi industry. After working with truly world-class engineers, and seeing how genuinely state-of-the-art products are conceived, designed, debugged, and made ready for production, I came to see the hifi industry as the small-scale craft industry it really is. It was kind of a come-down, really, but it put the big egos, marketing schemes, and endless glossy magazine reviews in a different perspective.
Like every audio designer I know, I am forced to trust my own perceptions, and use analytical skills to unravel what I'm hearing. Sometimes I can do that, sometimes I can't. Sometimes measurements correlate, other times not. There's also the matter of subjective preference, much like preference in foods. If somebody can't stand the sound of vintage analog—either mastertapes or LP format—that's a matter of personal preference, not absolutes. Since I've been involved with spectrum analyzers and have a degree in perceptual psychology, I know there's a lot more to sound than Total Harmonic Distortion and S/N ratios, especially since different listeners tune into different aspects of sound. When it comes to audio, I'm not a fundamentalist. There are many paths to heaven, not just one. The folks who say otherwise are marketers with an agenda to push.
I find your spirited defense of a dated consumer-grade medium like Red Book a little odd—the entire recording industry disagrees with you.
You're going to have to look far and wide to find any recent recordings made at the 44.1/16 rate and bit depth. All it takes is a flip of a switch on a Meitner-grade unit to hear the difference between native-format 96/24 and the most carefully optimized reduction to 44.1/16.
Have you heard this for yourself? I have. The difference was more than a little obvious, and I personally would class it as the difference between professional quality and fair-to-middling consumer- grade. If the flip of the switch isn't audible, that says something about the resolution and quality of the playback system—at the minimum, it is not suitable for recording and mix down purposes.
Considering the severe limitations—which were heavily criticized at the time by many AES members, including myself—of the Red Book system, I didn't even see it as professional-grade even back in 1982, much less some twenty-four years later. It should be remembered the professional standard from 1975 to the early Eighties was the Tom Stockham 50/16 system, NOT the inferior Sony/Philips system. The AES membership was so antagonized by the Sony/Philips "my way or the highway" attitude they recommended a rival—and at the time incompatible—48/16 system for "professional" use. Their words, not mine. That, by the way, is why movies (and Dolby Digital) are made at 48kHz and multiples of that rate. Even at the dawn of the Red Book format, recording engineers didn't consider 44.1/16 a professional-quality system.
Remember, Philips' only previous contribution to the world of hifi was the Compact Cassette (not exactly a triumph of fidelity), and in 1982 Sony wasn't known as an audio innovator, either. In fact, Philips wanted to choose 14-bit depth (!), and it was Sony that pushed and pushed for 16 bits. Let's not give them retrospective glory for what was essentially a marketing triumph that leveraged the dying Laserdisc technology into a new and far more profitable market.
I find it interesting we saw no development of a higher-resolution consumer medium until the Philips and Sony patent pool expired, and the extremely profitable royalties from every CD and every CD player came to an end.
Your defense of compression I find completely baffling. What exactly is sonically desirable about compression? I'm serious. Are we running out of disk space? Is the Internet getting slower? Are processors becoming less powerful? Is the storage capacity of iPods too small—maybe 5000 tunes isn't enough, now we need 500,000 instead? I admit I don't follow this line of logic at all. How on Earth does compression improve an already flawed signal? Aren't we talking about high-end audio in this forum, instead of clever ways to degrade quality and hope nobody notices? I'd say that's the antithesis of quality—removing it bit by bit (literally) and hoping nobody notices.
I do find one point of agreement—DSD in particular suffers from very poor consumer implementations—and I include players like the Sony SCD-1 in this characterization. High resolution PCM, and especially DSD, create tremendous amounts of out-of-band noise, otherwise known as RFI. A Tektronix friend and I measured a noise spectrum that was flat to 20 MHz—coming off the unfiltered output of DAC converter chip playing Red Book content! I calculated the minimum speed to avoid slewing in the analog electronics was 800V/uSec—and yet I see $15,000 players, complete with glowing reviews in the glossy magazines, that use 5532 op-amps that have a slewing rate of 13V/uSec. I don't care how famous the designer is, slewing ain't hifi, sorry.
I expect the requirements to avoid slewing in a high-res PCM or DSD player to be in 2000V/uSec region, essentially HD video territory.
This, to me, is an example of famous audio designers missing the obvious, probably because they don't have any RF spectrum analyzers in the lab, and aren't familiar with RF design techniques. But solving the problem isn't rocket science—a studio-grade 1:1 isolation transformer, with electrostatic screening, basically filters off all RFI, and breaks the ground connection as well, where quite a bit of RFI floats around.
Senior Assistant Editor, Positive Feedback Online
[John Phelan responded to Lynn Olson as follows:]
Lynn, I appreciate your honesty with your belief that there are multiple ways to achieve sonic perfection. In my view, DSD is not one of them.
Why not? Well, it's mostly pure noise and a huge waste of space!! Moreover, its sampling speed is four times slower/worse than Phillip's own Bitstream system of the 1980s. Because of this and other reasons, DSD offers no more than 6 bits of intrinsic resolution—Red Book offers (and delivers) 16, more than twice as much. Now, I'm not saying that the CD standard is perfect; it's just that it's FAR better than DSD-SACD. The technical problems that faced Red Book (average calculating, ringing patterns, band limiting, box-car shaping) can largely be overcome by sampling the waveforms at much higher rates (i.e., 176 kHz). I'm sure you know this. This, along with
much longer filter tap lengths at home (to better flesh-out the transients), are a great one-two punch that knocks out the "Red Book Scare" that still inflicts so many—including you.
Do try to listen to a few CDR-burned Red Book recordings by the labels I mentioned in my last note (on a good player—I like the Reimyo) and tell me what you think. I wouldn't have sold my Clearaudio Master Reference Turntable had CD not sounded so good. Note that I'm using a cutting-edge power conditioner—the new generation units work FAR better than the ones of years past. They're now preventing cross-component contamination that was a problem not known about before (via the AC line—not the inter-chassis ground). I also use a good three-point equipment rack, it helps drain-out the microphony. These two tweaks really help!!!
As for compression, I agree—it's not in our best interest. Hence my mentioning of uncompressed digital—I use it every day.
Thanks for the IAR references; I was just plowing my way through the assortment of articles in the Internet magazine. Umm, where to start...
I see no valid reference for the assertion that DSD has only 6 bits of effective resolution. Without a coherent explanation and derivation of the number, it looks to me like it's pulled out of thin air. The explanation of how noise shaping works—and it is used in all modern hybrid delta-sigma PCM converters as well as DSD—frankly makes no sense at all. I worked for nine years as a technical writer documenting the most arcane things imaginable, and my editors would have demanded a complete rewrite of that section.
The hypothetical "computer" in noise-shaping systems doesn't try and create idealized waveforms, and then repeat them ad infinitum. That's completely ridiculous. They are NOT computers, synthesizers, or even spectrum analyzers. Noise-shaping systems have no understanding of what they're doing at all—in fact, they are notorious for going unstable with certain combinations of input signals. That's where the famous "idle tones" come from—instability in a high-order feedback system.
1) Noise-shaping is a technique that uses digital feedback to linearize an inaccurate low-bit converter.
2) Like all feedback systems, it uses a process of successive approximation to linearize the device under feedback control, and as dictated by feedback theory, must operate many times faster than the signal of interest.
3) Like all feedback systems, stability is the most serious engineering problem to be solved. This typically involves a tradeoff between stability and precision, paying particular attention to recovery from transient-overshoot conditions. The difficulty of the solution is in direct proportion to the order of the feedback. Some systems are so complex that they cannot be accurately simulated.
THAT'S IT. There. That didn't take five pages, did it?
These days, noise-shaping algorithms are quite high-order, ranging from 5th- to 9th-order. They essentially boil down to nested feedback loops, and like all feedback systems with poorly defined elements to be linearized, can enter into odd states, and even latch up and stay that way until re-set. Malcolm Hawksford has written the most powerful critiques of noise-shaping, and I suspect IAR read the Hawksford papers (in the AES Journal) and attempted to rewrite them in more popular language. If you're curious, skip IAR, and read the Hawksford AES papers in the original. There's a bit of math, but Hawksford is an excellent writer, and an internationally renowned digital engineer.
To repeat, noise-shaping is used in ALL modern high-resolution converters (PCM and DSD), which are typically 4- to 8-bit converters running very fast and using high-order noise-shaping to attain their published specs—which apply to sine waves only. I suspect the demise of single-bit converters, and their replacement by hybrid technology, had more than a little bit to do with Hawksford's papers.
The Old School alternative to delta-sigma hybrids are the pure ladder converters, the last examples being the ten-year-old Burr-Brown PCM-63 (true 20 bits at 384kHz) and eight-year-old Burr-Brown PCM-1704 (also 20 bits at 384kHz). Being true ladder converters, they require phenomenally accurate laser-trimming, ultra-performance reference voltage supplies, and pretty sophisticated engineering in the (outboard) I/V conversion. This was the converter where my Tek friend and I measured a flat comb spectrum going from 20kHz to 20MHz, then declining into the noise around 50MHz. A sobering sight to see, I can tell you. It gave me a very healthy respect for the severe demands placed on the audio section, at least the parts responsible for buffering and low-pass filtering.
I like the PCM-63 and PCM-1704 converters myself, and find myself in some agreement with Hawksford. Not only am I unsure of the mathematical soundness of the DSD algorithm, I am equally doubtful of noise-shaping (again, used in all modern DVD-A and SACD players). As IAR mentions elsewhere, the underlying premise of dither breaks down when you move away from sine waves measured over long time intervals. The same caveat applies to noise-shaping, where the greatest benefits are attained over long time intervals with sine wave stimulus. Over brief time intervals, the benefits of dither and noise-shaping greatly decrease.
I surmise the sonic difference between the old school ladder converters and the modern delta-sigma hybrids is due to the difference in the way they handle transient conditions—the old school uses no feedback at all, and the new ones use lots and lots. This kind of thing is usually audible. Feedback has a long history of fooling the spectrum analyzer but not the ear—it takes rather exotic test signals to trip up a feedback systems, but it can be done.
As for the assertion of blurring above some arbitrary frequency (again, a number chosen with no apparent derivation or reference to anything else), this is a subjective experience backed up by a pseudo-technical argument. Again, the numbers for 44.1/16 "decreasing resolution" above 2kHz, or DSD above 8kHz, appear to be pulled out of thin air. I don't buy it, sorry. Back in Tektronix we called this a hand-waving argument—a lot of noise and opinion, but no data. I certainly don't see anything like this on a spectrum analyzer, either swept-frequency or FFT. Subjectively, now that's a different story, and there may be something there.
Sorry to rain on the IAR parade, but that's how I see it. Subjective opinions—you bet, why not, we all hear things different—but this mishmash of misunderstood and badly explained digital engineering, combined with an appallingly uninformed and bitter attack on DSD (which does have legitimate technical criticisms aimed at it), genuinely disappointed me. The repeated pitches for IAR-designed "tweaks" mixed in with the reviews undermined a great deal of the subjective comments.
Yes, I'm one of those guys that considers most tweaks a waste of time—I'd rather work on horn and dipole-radiation theory, resonances in drivers, the harmonic spectra of different devices, transient stability in power supplies, and the technology that underlies recording and playback process. That's way more fun than buying commercial stuff—where's the thrill in that? Commercial stuff is boring—these guys mostly copy each other, sad to say. As you can see from my Web pages (http://www.nutshellhifi.com/), I like the really weird stuff, things you can't buy anywhere in the world at any price. Now THAT's my kind of fun.
Senior Assistant Editor, Positive Feedback Online
Okay—I believe you. But DSD STILL contains too much noise for my liking—I prefer a more substantiated approach. With SACD out of the picture, we're left with Code Red.
Thanks for spending so much time on this discussion—but I hope we're not finished yet. If you would, give me your thoughts on what http://mother-of-tone.com says. This is a funny name for what is really Altmann Audio of Germany. For the record, I don't plan on giving you an endless array of websites to plough through but this one looks VERY credible.
Dick Olsher (of EnjoytheMusic.com) recently reviewed their Attraction DAC and said it was the best digital he has ever heard—at $1500!! Let me know what you think—just click on the site's side bars and read away (relief—the articles here are shorter than IAR's).
Actually, John, I want to thank you for a thought-provoking discussion. I try and tread a fine line between heavy-handed debunking and keeping an open mind about unexplained phenomena—which extends to really out-there stuff like Roswell, JFK, 9/11 and other tinfoil-hat favorites. The tools I use are: trusting my own perceptions, and setting the perceptions of others (unless I know them personally, and have their systems for myself with music I'm familiar with) pretty much to zero. This is because I've been to the homes of quite a few well-known reviewers, and I thought their systems were really awful—especially the systems of the tweaker brigade.
This is why I think reviews from big-name reviewer XYZ tell you nothing about how the gear sounds (unless you know for a fact your tastes and perceptions are the same). For that reason, subjective reviews are not something I pay a lot of attention to, except to advise people to not set any great weight to any magazine review.
What does annoy me are content-free reviews pretending to be authoritative. I've designed enough audio gear to know that the price- tag and international fame of the designer mean nothing. The sound of all hifi gear is a function of the devices used, the overall philosophy of the design, and the skill of implementation. That's it.
Really, there's no magic involved. It's like cooking: good ingredients, good recipe, and skill on the part of the cook..
Case in point: an IAR review of a $20,000 two-way speaker with a 6" bass-mid driver and a dome tweeter. In this review, we are told nothing about the cone material, magnet material, crossover frequency, crossover slope, parts quality of crossover (including the brand of the caps, very important), and cabinet material. I don't care how famous the designer is or how elevated the price-point, take it from somebody that's designed a lot of speakers, the parameters I listed above set a hard upper limit upper boundary of quality. Really, truly. Polypropylene cones sound like plastic; paper sounds like paper; metal sounds like metal, and carbon-fiber and Kevlar can sound pretty damn bad, thanks to severe HF breakup modes.
Similarly, for CD players, DACs, and DVD players, the rock-bottom minimum specification is the brand and type of DAC, and the choice of analog electronics, including a general description of the circuit if it uses discrete transistors or tubes. This controls the overall sound of the DAC, believe me. Ladder DACs alwayss sound different than delta-sigma, and op-amps sound different than discrete circuits. Always always always. For the reviewer to omit the most important specs is grossly negligent.
Now, assuming we have made XYZ choices in a speaker—or DAC—or power amplifier—we've set an upper limit to the sound quality. Now all the designer can do is screw it up some more, which is very common in the high-end biz. Case in point: using cone materials with a very rough HF breakup region, and 6dB/octave crossover that lets the crud and crap right through. That is a design error that goes unnoticed by almost every reviewer, since they don't know what cone breakup sounds like. Once you do, you hear it in the first few seconds in many expensive commercial systems.
Another case in point: analog electronics in DACs that have inadequate slew rate. This is almost ubiquitous except for studio-grade equipment, where you finally see competent engineering. I wish the audiophile DAC designers would shell out the money for a 100 MHz RF spectrum analyzer and learn how to use it.
This is one area where I hold reviewers responsible: if they don't publish the key parameters of the device they're reviewing, the review can be actively misleading, in persuading great readers in the Great Man theory of design—that anything touched by the Guru is always the pinnacle of human achievement. This is nothing more than thinly-disguised cultism. There are world-class designers out there, but none of them are the ones you read about in magazines. They are very obscure DIY'ers that have worked for high-tech companies like Tektronix, Intel, and Hewlett-Packard. These guys are really good.
P.S. Yes, will take a look at mother-of-tone—always interested in new stuff!!
Senior Assistant Editor, Positive Feedback Online
[After checking out the mother-of-tone.com site, Lynn added the following commentary:]
John, I'm in almost complete agreement with everything on the MOT site. As a loudspeaker designer you make the painful discovery there's going to be a residue of coloration no matter what you do—this is why I get mad when I see reviews saying the "Great Man" has finally gotten rid of all coloration. Wrong. I don't care what technology is used, there's always some added coloration that's not in the original recording..
Rather than fight it with 300-lb. composite enclosures and $3000 power conditions, the smart—and artistic—choice is to shift around the coloration so it is musically consonant. Making things out of lacquered spruce wood makes a whole lot of sense to me—the pix of guitars and guitar amps say it all. Guitar players know tone—they make their living from it!
Based on this site alone, I'll probably be making my next speaker enclosures out of lacquered spruce wood, or at least comparing it to an identical enclosure made out of Baltic Birch (also a great material). MDF by comparison is "grayish" or dull sounding, and composites I suspect are unnatural sounding. People like wood for a reason—it sounds good. I suspect that wood and brass work well together—musicians have been using them for a very long time.
I'm not a big fan of the TDA1541 Philips converter, which I've found on the coarse side (although extremely vivid with a high fun-factor).
The PCM-63, though, is the absolute pinnacle of ladder technology, along with the PCM-1704 (which is a bit less flexible). I suspect the airy, analog-ish, but not-that-involving sound of modern 192/24 players is because they are delta-sigma (digital feedback) DACs. Ladder DACs make the 16- or 20-bit conversion all in one go, sample by sample, with no memory between samples.
More good reads, along similar lines::
Senior Assistant Editor, Positive Feedback Online