In an offhand comment on Facebook I mentioned that putting music on an LP entails some necessary sonic compromises for engineering reasons, and a friend replied, “Are you saying the sonic compromises are still being made for today’s vinyl, or for CDs and other digital formats too?”
This post is an edited version of my reply, with some elaboration and detail added, though I admit I’ve still tried not to be too long-winded. I’m leaving things out almost everywhere.
Short answer to my friend’s question: The sonic compromises I referred to are specific to vinyl, or at least the impetus for them is. But you will sometimes hear those compromises on CDs and other digital formats. Read on.
The perennial subtext for questions about music media: Which sounds best? CD, LP, or Streaming?
When CDs were new and record companies were in a rush to fill stores with them, the companies digitized their existing master tapes as quickly as they could, and of course the lion’s share of those tapes had been prepared for the purpose of driving LP cutting lathes, so they had the necessary vinyl compromises built in. They had even more compromises built in because the LPs of the day were selling to everybody, with every level of equipment quality; those LPs were not only LPs but they were mostly least-common-denominator LPs. Made for playing on inexpensive home record players, those LPs had lots of compromises, and for the CDs of that time, those LP compromises were printed to the CDs, too, because it was clear that people would buy the CDs just because of their immunity to scratches and dirt, which are among the most obvious problems with vinyl, but certainly not the only ones.
What were the LP compromises I’m talking about? Lots of things, but some of the biggest are:
- Low bass largely absent, and middle bass strongly attenuated. This was to avoid making the record skip. On inexpensive systems the stylus cantilever wouldn’t allow large bass excursions. Even on most high-quality systems there was very little damping in the mechanical system made up of the stylus cantilever and the tonearm mass (until the advent of dynamic stabilizers on Shure’s V15 type IV cartridges in 1978), so the signal had to avoid exciting resonances in that system.
- Not too much treble and sibilance drastically reduced. Signals that would require very sudden movement of the stylus tip had to be limited because they would cause the stylus to mistrack — lose contact with the walls of the groove — and this almost always damages the groove and accelerates stylus wear. Equalization and multiband dynamic range compression were used to limit the amount of treble signal, and a combination of those two effects was so common for taming sibilance that it has a special name: de-essing. You don’t have to listen very carefully to hear that lots of recorded 20th-century vocals have almost no “S” sounds in them at all, and part of the reason is the de-essing applied to limit stylus mistracking. Modest de-essing is still often used even when the signal isn’t going to be put on LP because it makes the vocals, especially close-miked vocals, sound better.
- Bass in the center. Remember above where I said the stylus cantilever and tonearm formed a resonant system with little damping on most turntables? The most effective mitigation was to tune the resonance of that system below the audio frequency range, usually somewhere between 4Hz and 10Hz if I remember right. But this tuning was tricky, and even done right it addressed the resonance for horizontal movement only. To support the tracking weight, the suspension for the stylus cantilever often had to be stiffer in the vertical direction, so a system properly tuned for tonearm resonance in the horizontal plane would often have vertical resonance in the bass audio frequency range. To avoid mistracking by exciting this vertical resonance too much, LPs had to have very little difference between channels in the bass frequency range.
A few years after CDs became common, a time came when different master recordings were prepared for CD and LP. This was made easier by the fact that more and more production (recording, mixing, mastering) were done in the digital domain anyway.
With big money flowing into the record companies from strong sales of CDs at inflated prices, they could afford, and often did choose, to make separate masters for LPs and CDs, with the LP masters, of course, having more compromises for the medium and therefore the sonic signature of those compromises laid over the music. The record companies soon internalized the idea that CDs were dramatically more profitable than LPs because CDs were cheaper to produce and were nonetheless selling for more than LPs. Despite the industry’s promises to pass along the savings in production costs brought by CDs, the public was willing to pay more for CDs, mostly because CDs delivered the music better. This was true even though the industry’s lack of experience with digital media meant that Mistakes Were Made on a regular basis in putting music out on CDs, and those mistakes would later bolster the shoddy “digital sounds bad” part of the foundation for today’s resurgence of vinyl.
Given those pressures, it’s no surprise that the LP master tapes were not prepared with a lot of care, and many of the pressings were abysmally awful as well, with dust particles and other contamination routinely allowed into the process to cite just one more way that mainstream LPs were an unnecessarily inferior product in the early and middle ’90s. The record companies had little to gain by extending the life of that format, and it’s no accident that they didn’t make and sell the best quality LPs they could in this period where most of their money was coming from CDs and the only things driving LP sales were lower price, bigger cover art, and utility to folks who hadn’t popped for a CD player yet.
So LPs essentially died.
The death of LPs and the fact that streaming hadn’t taken off yet were two of the important factors that fueled the Loudness Wars, whose greatest toxicity was felt roughly from the middle ’90s through the early 201x decade. I won’t say much more about that because it’s another topic, but this, too, contributed to the mistaken impression that CDs somehow sound inferior to LPs. The reality is, no, CDs merely allow a kind of inferior sound that’s harder to fit on a functioning LP. (That and other details of the Loudness War are a possible topic for a future blog post.)
Fast forward a bit to today, and we have LPs outselling or nearly outselling any other physical media, including CDs, partly on the basis of misunderstandings about what each of the formats can and cannot do. Record companies and their funding figure in a smaller and smaller number of the decisions in music production, with independent artists and small labels having taken over much of that role. Lacking the capital that the record companies had in the heyday of CDs, today’s mostly independent decision makers once again must choose whether to distribute product on LPs, on CDs, or on both. If you distribute at all today, you distribute via digital streaming and (to a smaller extent) downloads; that is a given. But less and less attention gets paid to physical media. CDs don’t really require a master recording that’s sonically different from what streaming needs, so the same production process works well for both streaming and CD. But if you want to sell music on LPs — and many people do — you’re going to face today’s version of the old question: Do I use my LP master recording for streaming and CDs, or do I pay for additional work in mastering to take the best advantage of both kinds of media? In most mastering workflows, preparing the material for LP amounts to added steps at the end of the process, so the cost difference often isn’t that big because the CD+streaming master nearly gets created as a necessary by-product of the LP master creation anyway.
But there is a cost difference, and so that question is getting answered both ways pretty often; I’m not aware of a clear trend.
And here we finally arrive at the answer to my friend’s question: While it’s absolutely true that today’s LPs require sonic compromises just as LPs always have, those compromises might or might not appear on other media that carry the same material. And importantly, LPs are much more of a niche today than in their own heyday, selling to listeners with better home stereos and more money than when LPs were mainstream. So greater care is typically taken in today’s LP production than ever before, with the result that yes, despite all the flaws of the medium, a modern LP can (and almost always does) really sound amazingly good. It does still include compromises, but it won’t have all the bass removed like some old LPs did because they had to play on extremely cheap systems. Today’s vinyl won’t usually have all the treble removed to prevent mistracking by cheap styli. It won’t have the imprint of dirt particles on its surface. And so on and so forth. It’s still true that an LP cannot sound as good as a CD at its best, but the gap in practice is smaller than ever, regardless of whether separate master recordings are used for each medium.
The header above mentions streaming, so I’ll just say quickly that nobody thinks typical streaming sounds better than CDs. There are a few streaming services that serve lossless 24-bit streams, and those can sound better than CDs if you have the trained ears, the high-quality electronics, and the super quiet listening environment to help you detect any difference at all.
Afterword: What’s Wrong with CDs?
Like any format, CDs have some built-in limitations. Layered on top of those limitations there have been artifacts of the engineering tools for coping with the limitations, and more importantly there have been large-scale failures to make the best possible use of the medium. It’s natural that it takes time for an entire industry to shift its practices to new ones that are better optimized for a new format, and that’s exactly what happened during the ’80s when CDs first appeared. To a lesser extent that evolution in practices continues even today.
To understand that landscape, you first have to understand some basics of how digital audio works, so I’ll quickly cover a bit of that.
Limitations and Challenges of the Format: Sampling and Quantizing
I’m going to try to explain these ideas accurately without mathematics. Hold my beer.
All digital audio representations are built on streams of samples, and sampling, even sampling in the analog domain, introduces two practical engineering challenges: Suppression of frequency mirroring is the first, and clock jitter is the second. Note that neither of these phenomena has anything to do with representing the signal digitally; these challenges are purely the result of the decision to represent a signal as a stream of samples, no matter how the samples are recorded. Keep in mind that if we could filter perfectly and clock our sampling perfectly, sampling would not be a source of signal distortion at all.
In a very fundamental way, a system’s ability to correctly reconstruct a signal from a stream of samples taken at regular intervals depends on sampling the signal often enough to catch even the highest frequencies that it contains. If we can’t be sure the sampling frequency is high enough, or equivalently, that the frequency content of the sampled signal is limited enough, we end up with multiple “correct” ways to interpret any given stream of samples. To be a bit more specific, the only way to get an unambiguous stream of audio samples is to take the samples from a signal that has no frequency content at or above half the sampling frequency. (To learn more about this — with mathematics — read about Nyquist’s Theorem.)
Engineers who design audio systems do two things to ensure that the sampling frequency and the highest frequency in the signal are far enough apart: They make the sampling frequency high enough to catch the full audio frequency range, and they apply filters to the signal before sampling to make sure frequencies at or above half the sampling frequency are eliminated as completely as possible. If those filters don’t work perfectly, the sample stream can get corrupted with signal information that won’t be interpreted correctly on playback. The result is called “aliasing” or “frequency mirroring” and it sounds TERRIBLE.
A lot of engineering goes into those anti-aliasing filters. Above I mentioned one way these filters can be imperfect: They can let high frequencies through that they should filter out. There’s another way they can be imperfect: They can distort the signal in the frequency range that they’re supposed to leave untouched. These distortions usually take the form of a frequency response that isn’t flat in magnitude, phase, or both.
Just to be explicit about one very important point that I implied above, I offer this reminder: Anti-alias filtering has to happen at the source, before any sampling occurs. This means in the recording studio, the mixing studio, and/or the mastering studio. Nothing in our home playback systems can correct for aliasing or anti-aliasing problems that occur upstream.
Sampling: Clock Jitter
In electronic and electromechanical systems, precise control of timing is an engineering challenge. For analog media like LPs, this challenge shows up as wow and flutter: speeds that should be constant aren’t, quite. In systems that represent a signal using a sequence of samples, the analogous challenge is called clock jitter: the sampling interval that’s supposed to be exactly regular isn’t, quite. You can explore a clock jitter listening test on this page. Like aliasing, jitter is not a “digital” problem per se. It’s a “sampling” problem, whether the samples are represented digitally or any other way.
A lot of engineering goes into stable, highly accurate clocks and other design elements to make sampling as accurate as possible. As with aliasing, errors in the timing of samples are permanent; nothing downstream can compensate for them. In some professional digital recording studios, the most expensive single thing in the building is the clock.
Unlike sampling, quantizing is a specifically digital step in the process, and it inherently adds distortion no matter how accurately our system does the job. Quantizing means using a limited amount of digital information to represent each sample. The more limited our information budget for each sample, the more information we throw away about the signal when we quantize each sample, and that information is irretrievably lost to the playback system. The change in signal introduced by throwing information away in this step is called “quantization noise.”
CDs represent each audio sample as a 16-bit value. The number of bits determines how much louder than the quantization noise the stored signal can be. In CD playback the 16 bits of each sample are interpreted in the most obvious way: you can think of each sample as a 16-bit binary number. With that interpretation, the stored signal can always be at least 96dB louder than the quantization noise. This 96dB value of CD dynamic range is HUGE compared to the dynamic range available on an LP! I’ve digitized numerous LPs and on the very best ones, the quietest silence (usually found in the pure surface noise between tracks) is about 45dB below the loudest signals, which gives a gargantuan 50dB margin of superiority to 16-bit CD audio.
But under some conditions quantization noise can sound really, REALLY bad, more unmusical than any of the compromises required to make a practical LP. When the recorded signal on a CD is loud enough (and it almost always is), that kind of pathological quantization noise pattern doesn’t happen, but in especially quiet passages it can be a problem if nothing is done to address it. And things can be done; see the dithering discussion below.
So those are the four big challenges inherent in all digital audio, including CDs: aliasing artifacts, anti-aliasing filter artifacts, clock jitter, and quantization noise. Those four things combine to make up the price you pay for avoiding the hiss of analog tape, the surface noise and pops and clicks and stylus / groove tracking challenges of LPs, and the wow, flutter, and frequency response issues of every consumer analog format.
Mistakes Were Made? What Mistakes?
Well, conscious compromises were made, too. Often it costs money to do better, and budgets are limited. As electronics have gotten smaller and cheaper over the last 40+ years, the industry has been able to improve a great deal on some of the compromises of early CDs.
Here are some of the compromises that affected early CDs: Anti-alias filters weren’t always great; they are far better now for many technical reasons. Clocks also weren’t always great, and some studio clocking systems had subtle bugs in them that took the industry a while to notice and fix. Like anti-alias filters, clocks are generally much better now than in the early consumer digital audio era, but there is always the possibility of bugs and design errors.
On top of the conscious compromises, there were legitimate mistakes. Mastering engineers left “headroom” rather than use full scale of the 16-bit word available to them on CDs, causing a loss of dynamic range and increased quantization noise problems.
Speaking of quantization noise problems, remember when I mentioned that there are hacks for coping with the worst aspects of quantization noise? The most important such hack is called dithering, and it amounts to adding carefully controlled noise to the signal right before quantizing. (In this context, quantizing can be converting from analog to digital like we’ve already discussed, or it can be a word-length reduction step like rounding 24-bit audio samples to 16-bit ones. Either way, the idea is the same.) With the right amount of added noise, even a signal whose amplitude is below the size of the least significant bit can be made clearly audible! Check out this awesome dithering demo page for more info.
Early in the consumer digital audio era, most audio engineers didn’t have a good handle on noise-shaped dithering of digitally recorded, mixed, and mastered audio; many didn’t even know what dithering was! Why? Because at that time dithering often wasn’t needed. With sources that were originally recorded on analog tape, or that were mixed or mastered to analog tape after being recorded digitally, the hiss of the analog tape provided dithering for free, at least at the stage of preparing final 16-bit sample streams. It might have been missing from earlier quantization stages in the case of digitally mixed audio, it might not have been ideally noise-shaped, and it might not have been at the ideal level, but it prevented the worst artifacts of quantizing low-level signals. But as more and more material was released with no analog tape stage in the process (remember those “AAD” and “DDD” designations on CD packages?), the industry’s use of good dithering lagged sadly behind the need for it for some time. This mistake was a big contributor to “digital sounds bad” perceptions because it meant quantization noise disproportionately affected audio that was the most purely digital in its preparation.
Briefly, a few other problems were unleashed or amplified by the demise of LPs and the new preeminence of digital audio: Loudness war escalation was enabled by CDs, or rather, by not needing to fit signals on LPs. With the new freedom to keep bass and treble open, some engineers overdid it and fatigued listeners’ ears. Cheap home playback systems introduced the possibility of inter-sample clipping that wasn’t accounted for in early preparation of digital audio.
So that’s about it. I claim that if you use best practices around clocking, dithering, leveling to avoid inter-sample clipping — basically apply all the ideas that the industry has learned about digital audio over the last 40 years — you can get a CD whose sound quality clearly exceeds the best possible LP. You might like large cover art, you might enjoy the experience of holding and smelling a 12-inch vinyl disc, you might appreciate using your beautiful turntable and cartridge, and you might like the sound of the compromises that enable the music to be delivered on LP in the first place. I like those things, too, a lot of the time. And you might find CDs that aren’t made the best they can be. But when it comes to absolute sound fidelity alone, I don’t think you’ll find that LPs win out over CDs when both are made to be their best.
Way-Afterword: What’s Wrong with Lossy Compression and Streaming?
This is another topic and my post is already too long. In the mean time, please just don’t be like Neil Young and confuse dynamic range compression with lossy data compression.