What do audio systems do? At the most basic level, they are Illusion Engines. I'll repeat that: a hi-fi system is really an Illusion Engine, a type of mechanical contrivance that hypnotizes the audience into thinking musicians are somehow present (or at least nearby). If the contrivance fails in this, it fails utterly, just as a magic trick entertains or it doesn't.
Re-framed in this context, "accuracy" simply doesn't apply. It's like rating a dream, or a hallucination, on an imaginary "accuracy" scale, although the mainstreamers have come up with a series of incantations, body postures (jaw-dropping, pants-flapping, cojones, etc) and audiophile-approved test discs that purport to accomplish this feat.
Although watching the highly trained Positive Feedback review staff going through the audiophile motions is pretty entertaining, the hi-fi systems that seem to produce the most agitation from them only rarely produce an illusion of musicians for me. From this experience I conclude that different kinds of people enjoy different kinds of magic tricks. Some people like THC, some like psylocybin, some like a dry martini that is shaken not stirred, some like Night Train, and others never indulge in any form of self-amusement lest the Wrath of God (or Science, take your pick) smite them down.
Having described audio systems as a special class of Illusion Engine (like THC, psylocybin, ethyl alcohol, Vision Quest, etc), there's one group for whom the word "accuracy" actually conveys meaning: recording and broadcast organizations with ready access to professional musicians. For the last 30 years, BBC monitor speakers have been subjectively assessed by rapidly walking between the control booth and the performing hall. The short trip is a quick reality check, and the BBC and other national broadcasting organizations command the resources to have both large groups of professional musicians and the engineering talent to design high-quality monitoring loudspeakers. This is why I give special weight to the writings of D.E.L. Shorter of the BBC, who pioneered the techniques of modern crossover design, cumulative-decay measurements, and swept distortion measurements as techniques to chase out subjective coloration.
Let's descend from the clouds and visit the more earthly realm of the hi-fi magazine reviewer, hi-fi retailer, and consumer (AKA "the food chain"). By and large, most of these folks are technically challenged, and incapable of building or designing their own (unlike the inhabitants of the levels above). They are forced by their limitations to approach audio in a building-block fashion, picking and choosing amps, speakers, cables, CD players, power cords, MPingo blocks, Vibraplanes, etc etc. As a group, they habitually fasten on simple one-dimensional numbers, such as cost, watts, THD, freq response, the -3dB bass limit, or if especially clueless, the 2-bit universe of "Stereophile" Class A, B, C, and D ratings, or at the lowest level of all, the 1-bit binary logic of "THX-approved".
This may seem a harsh view of the industry, but it is "accurate" in terms of which products sell and which ones don't. Without a 'Pile rating or the "THX" logo, most retailers won't give a new vendor the time of day. They won't even let you in their door. That's just how it is. So in one sense, "accuracy" is nothing more than market acceptance. By this logic, Microsoft defines accuracy in software and McDonald's defines accuracy in food. Think about it.
Viewing the audio chain in terms of convenient marketing "chunks" of CD players, power amps, and speakers, is to cater to the ignorance of the mass market. Test data is simple to measure at this "chunk" size, but correlation to sonics is low to nonexistent. Correlation to sales pitches aimed at reviewers, dealers, and customers, though, is excellent.
The device level is where things get a lot more interesting for the serious designer and enthusiast. For one thing, the "chunk" size is about right for getting at least a modest degree of correlation between what you hear and what you measure. For speaker drivers, roughness in the 3D cumulative-decay waterfall is a good index of potential coloration; however, the 3D waterfall loses most of its meaning when applied to the entire speaker, since the crossover neatly hides most of the gremlins from sight (although not from the ear of a skilled listener).
The device-level approach also works for electronics; the audiophile-approved 6DJ8/6922 loses much of its "low-distortion" charm when its abundance of 3rd and higher harmonics appears on the spectrum analyzer. Similarly, the audiophile favorite of the 12AU7 loses some of its luster when a dirt-cheap surplus NOS 6SN7 has three times lower distortion and three times the drive capability. These things are audible; when you work at the device level, repeatable correlations between sonics and appropriate measurements begin to dawn. Unfortunately, as long as magazine reviewers are incapable of appreciating the functional difference between a mu-follower and a SRPP, they will never penetrate this level of insight and understanding.
This is the right level for the skilled illusionist to work their magic. The adventurous triode builders, unlike the armchair generals of the mainstream magazines, have actually built mu-followers, SRPP's, RC-coupled circuits, transformer-coupled circuits, and explored the wonders of parallel-feed. Some of the geekier folks have turned on their spectrum analyzers and peered at the little wiggles down in the -120dB region ... and lo and behold, found useful correlations. By doing so, they are treading in the footsteps of Norman Crowhust, D.E.L. Shorter, and other pioneers of the Fifties.
Norman Crowhurst wrote a fascinating analysis of feedback multiplying the order of harmonics, which has been reprinted in Glass Audio, Vol 7-6, pp. 20 through 30. Mr. Crowhurst starts with one tube generating only 2nd harmonic, adds a second tube in series (resulting in 2nd, 3rd, and 4th), and then makes the whole thing push-pull (resulting in 3rd, 5th, 7th, and 9th), and last but not least, adds feedback to the circuit, which creates a series of harmonics out to the 81st. All of this complexity arises from theoretically-perfect tubes that only create pure 2nd harmonics!
With real devices there are even more harmonics. Then there's the effect of reactive loads, which adds a frequency dependency to the harmonic structure! (With reactive loads, additional harmonics appear due to the elliptical loadline seen by the power tubes. The elliptical load-line dips into the very nonlinear low-current region, resulting in an instantaneous increase in upper harmonics. This spectral "roughening" is most audible with strong low frequency program material and hard-to-drive horn or vented bass drivers.)
Feedback has no ability to "improve" the proportion of the distortion harmonics in a raw amplifier; it reduces all of the harmonics found in an amplifier in direct proportion to the feedback ratio, then adds new harmonics of its own thanks to the summing action of the feedback node. In practice, this means replacing large percentages of 2nd and 3rd harmonic with very small percentages of very many high-order terms. THD meters simply add all the terms together, paying no attention at all to the order of the distortion.
Crowhurst's mathematical derivation of the harmonic multiplication properties of feedback is impressively clear and direct, and D.E.L. Shorter's 1950's Wireless World analysis of how IM distortion dominates HD with real-world musical sources is equally clear. Put the two together and it is apparent that conventional feedback circuits create extremely large numbers of low-level sum-and-difference IM products. This adds a dynamic program-correlated distortion floor to the truly random noise from the electronics and the recording environment. Random noise by itself is perceptively benign, since the ear can hear "into" the noise, and discerns echoes as much as 10 to 20 dB deep into the noise by using a process of cross-correlation in the brain.
Echo (space) discrimination is an ancient survival skill that directly ties to the emotional limbic center in the brain; defeat this process by cluttering the echoes with high-order distortion, and emotional perception of the music is profoundly altered, perhaps even removed entirely.
This is why digital compression is so destructive to the emotional content of music; the algorithm efficiently removes all of the "unnecessary" low-level information, leaving behind the bare bones and sinews of the dominant harmonics. In a sense, digitally compressed audio is the culmination of the destruction that began with high feedback, then Class AB transistors, then low-resolution 44.1/16 PCM digital, and finally a process that boldly discards 90 to 95% of an already coarse-grained undersampled signal as "unnecessary."
No wonder that recorded music is on a downhill slide since these "advances" were foisted on the public; the transmission medium itself is hostile to subtlety and emotion. Want an extreme comparison? Compare a good vintage LP on an all-tube system to Dolby Digital or MP3 playback on an all-transistor system.
The monopoly power that created the 44.1/16 PCM Compact Disc is mirrored in the US high-end publishing industry, and the results have been equally destructive. There's been a gradual coarsening of discussion and dumbing-down of magazine reviewers, dealers, and consumers over the last two decades. This is a direct outcome of the increasing concentration of power towards two publishers with the power to act as a make-or-break gatekeepers for the entire high-end industry.
Students of economic history are well versed in this phenomenon, which results in the slowdown or absence of innovation in a given industry once power is concentrated in the monopoly-holder. (In economic theory is there is little functional difference between a monopoly and an oligopoly, where monopoly power is shared in an informal alliance between the Big Two or Big Three. The social, technological, and economic result is the same - a slowdown or cessation of technological innovation, price stabilization, and the deliberate creation of powerful barriers to market entry. The business-magazine euphemism for this is a "mature market" - one where competition is a formalized ritual between the Big Two or Big Three. I leave political parallels as an exercise to the reader.)
Returning to audio, this is why the mainstream hi-fi industry is at a technological dead end with declining sales and a gradual slippage into the home-theatre mass-market ... this is the natural result of the expansion of monopoly power at the expense of other stakeholders in the industry. Stakeholders - hmm, who are they? Take a look in the mirror. Anyone who loves music is a stakeholder, not just those with a financial interest in the industry.
Since the mainstream manufacturers, reviewers, retailers, and consumers were locked up, innovation and a re-think had to come from outside the closed circle. And by the early Nineties it did, right on schedule, thanks to desktop publishing and the Internet. Do yourself a favor and forget the review magazines, and build a library of back issues of Vacuum Tube Valley, VALVE, Glass Audio, and if you're really lucky, some out-of-print copies of Sound Practices magazine. On the Net, you can visit the denizens of the Tube DIY Asylum, or visit the folks at DIY Forums instead - and this is just the beginning of places to visit, discuss, and learn.
And it isn't just DIY where you can find good things to listen to. Yes, there are wonderful products made by manufacturers you never heard of before - and if you read the slick magazines with glossy four-color illustrations, you never will. This is why shows like the Rocky Mountain Audio Fest exist, along with other regional enthusiast shows, like the gone-but-not-forgotten VSAC. There are small enthusiast shows all over the world; visit them and prepare to be surprised.
Accuracy? Forgetaboutit! How does "High Fidelity" sound instead? It was good enough for Major Armstrong (inventor of FM), Alan Blumlein (inventor of stereo), D.E.L. Shorter and Norman Crowhurst. Maybe it's time to reclaim it and call it our own.
© Lynn Olson 1998, 2005. All Rights Reserved.
First published in 1998 on the Joelist by Lynn Olson, revised and updated in 2005.