Debunking Digital-Audio Myths

Get the lowdown on common digital-audio fallacies. The Information Age is a wonderful time, isn't it? With global media and the Internet, you can find

Get the lowdown on common digital-audio fallacies.

The Information Age is a wonderful time, isn't it? With globalmedia and the Internet, you can find data on just about any topic.Unfortunately, a lot of conflicting information is floating aroundout there, and it's often hard to tell fact from fiction. Thisarticle attempts to clear the air by addressing some commonmisconceptions about digital audio.

Myth No. 1: Copies of files aren't always perfect

Dubs between analog tape decks aren't perfect; every time youmake an analog copy, the signal degrades. It's therefore natural toassume that all copying methods share that characteristic. Copyingan audio file on a computer, however, is completely different frommaking an analog copy.

When you copy a file on a computer — whether it's an audiofile, a Microsoft Word document, or a shareware program— the operating system has to ensure that every byte of datacopies correctly. If one byte in a Word document goesastray, you might get spelling errors, formatting problems, orworse. If one byte in a copy of a shareware program goes south, thesoftware might not run at all.

Because of this situation, accurate copies of any file type arecrucial, and digital-audio files are no exception. To preventproblems, the operating system uses a verification scheme toestablish that all copies are byte-for-byte perfect. In theunlikely event that an error appears in the copy, the computer letsyou know.

So when you copy an audio file from one hard drive to another orback up data to a tape drive or CD-R drive, rest assured thatyou're creating a perfect duplicate.

Myth No. 2: All file compression degrades audio

Compressed audio formats, such as MP3, have truly changed theface of recorded media by letting music be exchanged easily overthe internet. The MP3 format shrinks audio files using“lossy” compression, which means that not all of themusical data is actually stored in the MP3 file. The more importantdata is maintained while less important data is thrown away. Theaudio file is then reconstructed on playback with varying resultsin audio quality (see Fig. 1). In any event, MP3 audioquality is degraded somewhat with respect to the original file.

Because MP3 is one of the most widely known audio-compressionformats, many people assume that all methods of compressing audiofiles work the same lossy way. However, not all of them do. Someprograms, such as Emagic's Zap and Waves'TrackPac, are specifically designed for lossless audiocompression (see Fig. 2). Those programs can't shrinkfiles as much as MP3 does, but they do retain all data whilecompressing files to about 50 percent of their original sizes.

Also lossless by design are general-purpose compression programssuch as PKWare's PKZIP, WinZip Computing'sWinZip, and Aladdin Systems' StuffIt. To theseprograms, an audio file is just like a Microsoft Exceldocument; every byte of data must be retained. Again, the file-sizereduction isn't as dramatic as with MP3 compression (and it's oftenless effective than audio-specific compression programs), but youcan be sure that the quality of any zipped or stuffed audio file iscompletely unaffected by the compression.

Myth No. 3: CD quality

What the heck does “CD quality” mean, anyway? Mycumulative annoyance at the misuse of this phrase leaves me feelinglike a cranky old curmudgeon when I hear it. Sure, I'll accept thedescription for any device that operates at 16 bits and 44.1 kHz— a CD player, for example — as long as its real-worldperformance measures up to the potential implied by thosespecs.

Unfortunately, the term is often used to describe almostanything that can spit out a tune. I've seen a $30 sound card witha 65 dB signal-to-noise ratio boast CD quality, even though 16 bitsshould offer a signal-to-noise ratio closer to 90 dB. Moreover,I've seen MP3 and MiniDisc players claiming CD quality, thoughthose devices start with CD-quality audio and then shrink it usinglossy compression, reducing both file size and fidelity. Soon we'llhave 4-bit digital toasters claiming their beeps are CD quality.Give me a break!

Even worse is the phrase “near CD quality.” Forthose unfamiliar with marketing doublespeak, “near” isthe same as “virtually.” In plain English, both wordstranslate to “not.” So what was once a technical termis now simply advertising gibberish.

Finally, I have to ask: is CD quality still supposed to be agood thing? At one time, 16-bit, 44.1 kHz audio was synonymous withstate-of-the-art digital technology. But that was then. In today's24-bit world (with 96 kHz sampling rates gaining in popularity),those CD specs are looking a bit long in the tooth. Maybe insteadof CD quality, the industry can agree on a more appropriate termlike “real old-fashioned CD goodness.” It's just athought.

Myth No. 4: 24-bit is 24-bit is 24-bit

Resolution is an easy way to specify a digital device's quality.Unfortunately, it is not a reliable benchmark. I remember a meetingwith a representative from a major digital-audio-chip manufacturerin which two of the manufacturer's models of 20-bit D/A chips wereevaluated. When asked why one of the chips was abnormally noisy andperforming more like a 14-bit D/A than its 20-bit spec suggested,the representative responded that it was “20 bits —with 6 bits of marketing.”

So what's the moral of the story? Just because two devices areboth “24-bit” does not mean they exhibit the same audioquality. In fact, fidelity can vary so widely that a well-designed16-bit device may sound better than a poorly designed 24-bitinstrument.

One variable is the quality of the D/A or A/D chip. The majormanufacturers of these chips may have a line of parts with the samegeneral specifications (such as 24-bit, 44.1 to 48 kHz) but withwidely diverging noise amounts and differing prices. Theclock-circuit quality is also important for minimizing jitter. (Formore on jitter, see Myth No. 5.) In fact, several high-end A/D/Amanufacturers specifically cite the their clocks' stability as animportant selling point (see Fig. 3).

Finally, remember that the A in D/A stands for“analog.” You know that there are good and bad soundinganalog mixers, preamps, and other gear, so it should come as nosurprise that a “digital” device's analog parts canmake a real difference in its overall sound quality. High-qualityanalog parts and clever analog design are absolutely essential fora digital device to realize its true potential.

Myth No. 5: Jitter is recorded during digital dubs

In a perfect world, each digital-audio sample is recorded andplayed back at exact, even intervals derived precisely from eachtick of the digital-audio word clock. For instance, a 44.1 KHzsystem should sample the incoming audio exactly 44,100 times persecond. Real-world clocks aren't quite perfect, however, and eachtick of the clock may be slightly behind or slightly ahead of whereit's supposed to be. That difference between the ideal timing andthe actual timing is called jitter.

Jitter causes distortion in digital audio, but it's differentfrom what you generally think of as distortion. Instead ofdistortion in amplitude, such as overdrive in a guitar amp, jitteris distortion in time that causes slight variations in the audiowaveform's shape. In a sine wave, for example, varying eachsample's timing causes the waveform to bulge out and cave in atdifferent points, as opposed to following the ideal smooth curve(see Fig. 4).

Every digital-audio device produces some amount of jitter, butsome devices exhibit much more than others do. Jitter can also becumulative: as a signal passes through multiple signal processors,mixers, and so on, the jitter may get progressively worse. Jitterbecomes “frozen” when you record an analog source witha digital system. In other words, every time you play back theaudio, you hear the effect of the jitter that was present duringthe recording. You also hear the jitter produced by thedigital-to-analog converter.

However, recording from one digital device to another isdifferent; as long as the data stays in the digital domain, jitteris not recorded. The only thing actually captured is a sequence ofamplitude values; digital media simply have no provisions forstoring information about individual sample timing. The timing isbased implicitly on the sampling rate and is freshly re-created bythe digital-to-analog converter's clock every time the audio isplayed back.

Even digital-audio tape systems don't play audio directly fromthe tape. Instead, they pass the data through a RAM buffer fromwhich a clock pulls individual samples and sends them to theoutputs. As a result, variations in tape speed or data spacingaren't reflected in the output data.

Although jitter causes distortion on playback — and cancertainly generate unalterable distortion during the A/D processwhen recording from an analog source — it is not recordedwhen making a digital dub or when recording between digitaldevices. On the other hand, if the jitter of one device involved ina digital dub is bad, it can cause problems of a different sort, asI will discuss in the next myth.

Myth No. 6: Digital dubs are perfect copies

If you copy audio between DAT decks, CDs, or modular digitalmultitrack (MDM) tape machines, you might expect the copy to beperfect. After all, digital audio is just ones and zeros, right? AsI described previously, copies of audio files on computers areindeed perfect. However, when you use tape and CD-audio media,those zeros and ones are being assisted, and sometimes created, byerror correction.

Here's the problem: If an error occurs when a computer reads afile from a hard drive, the computer can go back and try again. Abackup tape drive can do the same thing, rewinding the tape asnecessary. But with an audio DAT and other digital media, the tapeor disc must always keep moving; otherwise, you hear a pause inplayback. When errors occur, and they always do, going back andtrying again simply isn't an option. Instead, the DAT and CDformats include several methods for real-time error correction.

The first and most effective error correction is recorded ontothe tape in the form of Reed-Solomon error-correction codes. Thesecodes take up more than 25 percent of the data on a DAT tape or CD,and they allow most errors to be completely corrected, yieldingdata that is byte-for-byte perfect.

Occasionally, even this sophisticated error-correction mechanismcan't recover the data, resulting in one or more completely blanksamples. In that case, the second level of error correction comesinto play. This technique, called interpolation, considersthe data before and after the blank sample or samples and thenmakes a guess as to what value might have been in that blankspace.

Say you have the following sequence of samples: 2, 6,, 14, 15. Simple interpolation might draw a straightline between the samples just before and just after the error sothat the sequence becomes 2, 6, (10), 14, 15. More sophisticatedinterpolation constructs a curve based on two or more samples oneither side of the error. This procedure recognizes that the twosamples after the error have closer values than the two before theerror, and it tries to reflect that curvature in the correcteddata. Its reconstruction of the data might look more like 2, 6,(12), 14, 15 (see Fig. 5). Generally, the more points youlook at during interpolation, the more accurate of a guess you canmake as to the error's original value, though it's still just aguess.

Audio is almost always a continuous waveform, so the results ofinterpolation are good enough for musical use; it's nearlyimpossible to hear a single error. However, if a tape is in not inperfect condition or the deck's heads haven't been cleanedrecently, you might have enough errors to cause a generaldegradation of audio quality, which passes on to any copies youmake — digital or analog. Some DAT machines and MDMs displaythe current error rate. If yours does, monitor it from time totime.

So far I've only covered minor imperfections in digital dubs.Some problems can be more severe. For instance, if a DAT tape hastoo consecutive many errors, the error correction may not work atall, and you'll hear digital noise instead of your recording.

Jitter is another issue. As I discussed previously, it shouldn'taffect purely digital connections. AES standards specify maximumallowable jitter on output and minimum jitter tolerance on input,which defines the greatest amount of jitter that the input signalcan include and still be received properly. The current standardspecifies a minimum jitter tolerance of several times the maximumoutput jitter to allow for chains of devices and maximalinteroperability. Devices that conform to these standards shouldn'thave jitter-related difficulties. Not all devices meet thenecessary specs, however, and some older devices may have beenbuilt prior to the adoption of the current standards.

If a device has extremely high jitter at its digital output orif the device receiving the data can't handle much jitter at itsinput, their communication can fail, causing pops, clicks, andother artifacts. That holds true for connections between alldigital devices, including digital audio workstations (DAWs), soundcards, and effects processors.

Likewise, digital cables can fall prey to most of the samegremlins as their analog counterparts, including loose connections,defects, and impedance mismatches. You can always get errors inDAWs from disk-related problems (such as buffers set too small), orprocessor spikes caused by brief overloads in system activity.

Finally, many digital-audio glitches — clicks, pops,low-level “fizz,” and the like — can be caused bya simple problem: improper word-clock settings. Two devices thatare connected digitally but are not in agreement on the same wordclock form a recipe for real trouble. In my experience, this is thesource of many erroneous complaints that digital transfers areerror prone. If you merely ensure that all connected devices agreeon a single word-clock master, you can eliminate this commonheadache.

In summary, to ensure premium digital dubs, keep tape headsclean, preserve media (such as DAT tapes) in safe storageenvironments, set word clocks correctly, pay attention to errorrates, take the same care with digital cables as you would withanalog cables, make sure that your equipment has reasonable jittercharacteristics, and set your DAW buffer sizes conservatively. Aslong as you take those precautionary steps, you can expect digitaldubs to work well.

One other point you should keep in mind is to always monitorwhile you dub, just as you would with analog tape. Your ears shouldbe your final reference. Remember, the recording you save may beyour own.

Myth No. 7: All digital synths and effects sound the same

This myth is also known as: All digital EQs sound the same, Allvirtual analog synths sound the same, and All digital compressorsstink. This is as true for digital gear as it is for analog gear,which is to say, not at all.

With analog devices, you have great-sounding EQs andlousy-sounding EQs. What's more, a couple of great-sounding EQs (orcompressors or synthesizers) may sound totally different from eachother. It should be no shock that the same situation exists forhardware and software in the digital domain.

What makes a good or bad digital EQ, compressor, filter, oroscillator? Many issues of digital-audio quality arise from onesource, and it happens to be one of the major differences betweenanalog and digital audio: frequency range. Analog audio has atheoretically infinite frequency range, whereas digital audio(software and hardware) has a hard limit on high frequencies, asdetermined by the sample rate.

Many analog processes take “infinite” frequencyrange for granted, but they can't do the same in the digitaldomain. For example, a standard analog peaking EQ has a bell curvethat is symmetrical around the center frequency; one side slopes to0 Hz, and the other slopes toward infinite Hz.

You can implement the same EQ in the digital domain, butinfinity is suddenly much closer. In fact, what was infinite Hz isnow the Nyquist frequency (half the sampling rate) or 22.05 kHz ata sampling rate of 44.1 kHz. This difference in the proximity ofinfinity results in an EQ curve with a dramatically lopsided shape(see Fig. 6).

Things can get stranger as the EQ's center frequency approachesthe Nyquist frequency. At those high frequencies, I've seen digitalEQs that started to take on weird globular shapes and even go downin actual frequency as I turned up the frequency knob. I've evenencountered a peaking EQ that looked much more like a resonanthighpass filter.

Those problems can be avoided or at least minimized by cleverprogramming. The degree to which the programmer is successfuldefines, to a great extent, the differences between good and baddigital EQs.

Besides the creation of a more-or-less correct EQ curve, thereare also matters of taste and personality, just as in analog EQs.The way a programmer chooses to approach these infinite-frequencyquandaries affects the overall sound. Moreover, some productsemulate the more esoteric sonic distinctions among classic analogEQs, such as slope and overshoot characteristics.

Compressors and limiters also have frequency-related issues.You're probably familiar with aliasing — it causes audioartifacts when sampled audio contains frequencies higher than theNyquist frequency. Aliasing doesn't occur only during sampling,however; it can also happen entirely within the digital domain.

For instance, compression and limiting work by modulating oneaudio-rate signal (the input) with another audio-rate signal (thecompressor or limiter's automatic gain control, which operates inthe audio range when the attack and release envelope times arefast). When you modulate one audio-rate signal with another, it hasthe effect of adding the two signals' frequencies; if the totalexceeds the Nyquist frequency, you'll get some aliasing.

A full-bandwidth audio signal processed with a limiter or acompressor with fast attack or decay times falls into thatcategory; the faster the attack or release and the greater thecompression or limiting amount, the more aliasing you hear. That isthe cause of the crunchiness many people hear in digital-dynamicsprocessors. Again, clever programming, especially oversampling, canminimize these aliasing artifacts.

You'll find similar predicaments in synths. Resonant filterssuffer from the same infinity-is-much-too-close syndrome as digitalEQs, and various synths differ widely in their success ataddressing the problem. For instance, standard Chamberlin digitalfilters (the most common type) only work correctly to aboutone-sixth of the sampling rate. For a synthesizer running at 44.1kHz, that means the resonance tops out at about 7 kHz.

Oscillators have problems similar to compressors. For example, asquare wave at, say, 4 kHz is actually generating frequencies wellabove the Nyquist limit, because of the waveform's sharp edges.Untamed, that can cause excruciating aliasing, especially towardthe top of the keyboard (as you can hear in some popular products).Similar aliasing can also happen when samples are transposed abovetheir original pitches. Techniques for dealing with thesecomplications vary and account for some of the sonic differencesamong synthesizers.

Finally, it's worth noting that some solutions are commonknowledge and in the public domain, whereas many others areprotected by patents or kept close to the vest as trade secrets. Inshort, different products use different techniques for dealing withfrequency-related obstacles, and some are simply more successfuland pleasant sounding than others.

Myth No. 8: Hardware sounds better than software

Defining an audio process as a mathematical equation isessentially what digital hardware and software is all about.Whether you're using a synth or effects processor with dedicatedhardware, an algorithm running on a DSP chip, or a plug-in on aPentium or PowerPC, it's still just a numbers game.

The thing that creates sound quality in a digital synth oreffect is the math itself, or the algorithm. As long as the mathstays the same, it can run on custom hardware, off-the-shelf DSPs,Pentiums, or PowerPCs and still produce the same output. To look atit another way, the method that converts the math to a form you canuse (a software plug-in or a hardware box, for example) isunimportant; only the math matters (see Fig. 7).

So hardware should sound the same as software, right? Ingeneral, yes, though several factors complicate the matter. Someissues are technical; others are simply practical.

On the technical side, you may not always be able to run exactlythe same math on different hardware. You have two common approachesto performing math on computers: one is called fixedpoint, and the other is called floating point.Without getting overly specific, I'll just say that the samemathematical operation, such as adding two numbers, may produceslightly different results depending on which method you use.

Many popular DSPs, such as the Motorola 56000 series, performonly fixed-point math; others, such as the SHARC from AnalogDevices, practice floating-point math. Desktop processors, such asthe PowerPC and Pentium, can handle both calculation types but, insome cases, may do one better than the other. As a result, it maynot always be practical — or even possible — to performthe same math on two different machines. In that case, thealgorithm designer must write the algorithm specifically for eachprocessor. With careful work, the algorithm may sound exactly alikeon each machine, even to a golden ear. In some cases, however,slight differences may remain.

Technical issues aside, there are also practical reasons whyhardware and software products may sound different. The underlyingalgorithms of two products may be very different, for example. Ifyou compare a hardware product from one manufacturer with asoftware product from another, chances are that the algorithms willnot be the same.

The factory presets are important factors in synthesizers' andeffects processors' overall sound. Without good sound design, eventhe best algorithm may not sing as sweetly as it could. Conversely,talented sound designers can sometimes make a mediocre algorithmsound surprisingly good. So even if the two products' underlyingalgorithms are similar, the talent of the factory sound designerscan make all the difference in the world, and that varies from onefactory to another.

There is no theoretical reason why hardware should sounddifferent from software. Any differences you encounter are mostlikely the results of comparing apples and oranges, because fewproducts are offered in identical hardware and software forms.

Dan Phillips is a Bay Area-based composer and producer, andhe is product manager at Korg R&D. Check out his Web site Thanks toAndy Leary, Rudy Trubitt, and Benny Rietveld for their assistancewith this article.