No onecan deny that today’s digital technology allows the discerninghome recordist to make very high-quality recordings. Affordable MDMsand digital mixers are helping to level the playing field for homestudios and small, independent facilities. Why is it, then, that somestudios consistently crank out mixes that sound just a tad clearer, alittle smoother, and a bit bigger than those produced by the masses?You might chalk it up to better analog gear, such as mics and preamps.Maybe the studio’s acoustics have something to do with it, orperhaps someone with really good ears is mixing the projects. Thesefactors can all have a profound influence on the sound of a recording.But what if you have already got great gear, experienced ears, and afine-sounding room, and you’re still not capturing that goldensound? What else will deliver that last bit of quality to your digitalrecordings? If you want the very best quality possible, there arecertain practices that you must follow.
Higher is Better
The cardinal rule for digital recording is to deliver the highestpossible audio levels to your A/D and D/A converters without clipping.Converters are the single greatest source of distortion in digitalaudio. (By distortion, I don’t mean the familiar-soundingoverdrive effect but rather any aberration from the original analogwaveform.)
To keep distortion at an absolute minimum, your levels should be ashot as possible. This uses the full bit resolution (see sidebar,"Digital Definitions") of the converters for the smoothest and mostaccurate translation of the analog waveform into the digital domain. Bythe way, this rule applies to the converters in digital reverbs,delays, and other outboard gear the same as it does to the convertersin MDMs and digital mixers.
For the same reason, it’s often a good idea to compress a trackbefore it hits an A/D converter rather than afterward in the digitaldomain. By compressing a track and applying make-up gain to raise itsoverall level, you are presenting a hotter average signal to the A/Dconverter and using the full bit resolution the converter can offer.This should give you smoother-sounding, more detailed tracks.
This assumes that you know ahead of time that the track will needcompression at some point. Of course, whether or not (and when) youcompress a signal should always be a musical decision, not a technicalone. Obviously, if your analog compressor isn’t very good and thedigital compressors in your workstation or digital mixer sound great,you’ll want to use the latter.
Once inside the digital domain, precious bits can be lost due totruncation. For example, when the 20-bit converters on a digital mixer,such as the Yamaha 02R, feed a 16-bit recorder, such as an ADAT orDA-88, the "bottom" four bits are lopped off. Similarly, when amixer’s 24-bit digital output feeds a 16-bit R-DAT’s digitalinputs, eight bits are lost. In both cases, the audio suffers aphenomenon called requantizing distortion—the stuff that leadsnaysayers to proclaim that digital is cold and edgy.
To preserve some of the audio detail contained in the bits thatwould have otherwise been thrown away and to avoid harsh-soundingrequantizing distortion, digital mixers, converters, and other devicesoften allow you to add dither to the signal. Dither, in its most basicform, is low-level broadband noise that prevents truncation-deriveddistortion and is designed to preserve detail at low recording levels.In return, it adds a very tiny amount of noise that sounds like hiss.(Because digital signals typically have dither added during the A/Dconversion process, adding dither to a digital recording is oftenreferred to as redithering.) Not all dither is broadband, however;there are many different types of dither, each having its own spectralcontent. (For a more complete discussion of dither, see "Square One:Dithering Heights" in the December 1996 issue of EM.)
When you add dither to a 20-bit signal, before it is recorded to a16-bit MDM or DAT deck, the effect is heard mostly on signals below -40dBFS (40 dB below "Full Scale," or digital zero), for example, on fadeins and fade outs. Reverb tails fade smoothly rather than cutting off,and percussive sounds, like drum hits and finger-picked guitar, mightbe more clearly defined.
I say "might be" because, in reality, sometimes a reditheredrecording will sound less clearly defined than a truncatedversion—especially if the quality of the dithering is poor.I’m actually not a big fan of dithering; I prefer Apogee’sUV22 process instead. (I must admit, though, that I haven’t heardevery type of noise-shaped dither currently available. Some people, forexample, are big fans of Sony’s Super Bit Mapping.) As withdithering, UV22 adds noise to the signal. However, the noise in UV22 isconfined to a frequency band centered around 22 kHz (hence the nameUV22); therefore it is essentially inaudible.
About eighteen months ago, for my own personal edification, I set upa 3-way test in which I compared various recordings made at -40 dBFS. Ilistened to how the recordings sounded when processed with UV22, withdither added, and with simple truncation. The UV22 process made reverbssound more airy. Acoustic guitars and cymbals sounded cleaner andclearer. The stereo positioning of all the elements in the mix wasalso, by far, the most solid. The truncated recording sounded chokedand fuzzy in comparison. The redithered recording, significantlyobscured by hiss, was the least pristine of all.
Keep in mind that these tests were done at very low recordinglevels. At higher levels (peaks up around 0 dBFS), the dither imparteda subtle veil to the mix; truncation made vocals, sax, and harmonica atad edgy; and the UV22 process added both an airy clarity and ananalog-like smoothness to the overall sound.
Unfortunately, UV22 processing is currently offered only onexpensive recording gear, such as the Apogee AD-1000 and AD-8000 A/Dconverters, the Millennia Media HV-3C stereo mic preamp/converter, andthe Z-Systems Z-Q1 digital equalizer. Pro Tools users can takeadvantage of UV22 with Apogee’s MasterTools TDM plug-in (see Fig.1).
UV22’s effect, like that of dither, is quite subtle and wouldbe lost on most untrained ears, especially when heard on a crummy homestereo system. You should try to audition UV22 for yourself to see ifyou can justify the expense. Many mastering houses can provide UV22processing for your project if you don’t have the bucks to buy thegear yourself.
Whether you use dither or UV22 when recording to an MDM or R-DAT,make sure you match the source (e.g., digital mixer) and thedestination word lengths to avoid truncation. For example, the Yamaha02R allows you to select the word length of the digital data to be sentout from its tape buses. If you’re recording digitally to mostMDMs or DAT machines, you should choose a 16-bit word length. On theother hand, if you’re recording to the new Alesis M20, you shouldchoose the 20-bit word length for the tape buses.
Avoiding the Jitters
There’s a common myth in our industry that making a digitalcopy of a track or DAT master will always result in an exact replicawith zero degradation. Contrary to popular opinion, however, the wayyou move data around inside the digital domain can have a noticeableeffect on the way your music sounds. This is due to a phenomenon knownas jitter.
To understand jitter, it’s helpful to take a look at theimportance of synchronizing digital audio bitstreams. When one piece ofdigital gear sends data to another piece of digital gear, the twopieces of equipment must share the same stable clock. The clock setsthe sampling frequency of both devices to be exactly the same. Althoughtwo devices may be ostensibly set to 44.1 kHz, their samplingfrequencies will drift and become slightly different with respect toeach other if they are not synchronized to the same clock. Because thepotential for clock drift is compounded when a signal is routed throughmultiple digital devices, it’s best to slave all digital devicesin your studio to a single external master clock.
If the clock is not rock solid, its timing inaccuracies will causesome audio bits to arrive early or late at the receiving device,introducing audible artifacts into your tracks. This is not because theactual values of the bits change; rather, it is because the arrivaltiming of those bits—the quantizing intervals—drifts. This isloosely analogous to a MIDI sequencer failing to snap notes exactly toa grid when quantizing a track at 100 percent strength—the notevalues remain the same, but the timing isn’t totally locked in.When this happens in digital audio, mild distortion occurs.
Although the clock-recovery circuits in high-end D/A converters cancorrect these timing anomalies, many of the converters offered asstandard fare on cost-effective digital gear just can’t cope withthe problem. For most personal-studio owners, jitter is an unfortunatefact of life.
What does jitter sound like? That depends on how jittery the signalis (that is, how wide the timing variations are). If the jitter is verylow, the effect is virtually inaudible. When jitter is audible, itmanifests itself in a number of subtle ways depending on the spectra ofthe jitter itself. The frequency components of jitter can vary widelyand can, therefore, modulate the incoming signal in different ways,causing a variety of subtle effects. Because of its chameleonic nature,jitter is something you must train your ears to recognize. Here aresome things to listen for.
Jitter is most obvious on stereo tracks (including mixes), wherephase anomalies are heard more readily. In most cases, the high-enddetail of your mix will suffer. For example, the "ping" of a cymbal hitwill be less defined and will lose some of its silvery sweetness.Flatpicking on an acoustic guitar will sound duller, harsher, orlacking in complex overtones.
Clarity in the low midrange often suffers; reverbs become moreflattened or 2-dimensional, and you can’t hear as far into themix. The mix will sound more like it’s coming from two speakers ona flat plane rather than occupying a 3-dimensional space. Subtlesweetener parts that are tucked back in the mix will be a tad harder tohear due to masking. Soundstage localization (the exact pan position ofeach element within the stereo field) will become a little more vague:the lead vocal might sound somewhat nebulous instead of smack dab inthe middle of the speakers.
If the jitter is severe enough, the mix will actually collapseinward slightly from the speakers, resulting in a narrower stereoimage. Sometimes, jitter will even rob a mix of a little bottom-endwarmth, causing guitars and drums to sound slightly glassy or harsh.The bass guitar and kick drum might not sound as tight and focused asthey should.
How serious a problem is jitter? Some gear is more jittery thanothers. The higher the quality of your digital audio equipment, theless jitter it will introduce into the bitstream. Although the effectsof jitter are usually quite subtle, even with budget gear, there’sno amount of EQ, panning, or effects processing that can prevent orundo the damage. Fortunately, there are easy ways to keep jitter to aminimum, so why not get the best out of your gear? Most of thefollowing tips won’t cost you a dime.
Your first line of defense against jitter is to use the most stableclock available as your word-clock master for the entire system. Ahigh-end studio might slave its digital mixer, converters, DAW, andMDMs to a dedicated master audio-sync box, such as the AardvarkAardSync II. Some outboard converters, such as the Apogee AD-1000, arenoted for having extremely low jitter and work well as a masterword-clock source for your other gear. But you don’t necessarilyneed to buy any expensive toys to improve your synchronization in amodest setup.
Every piece of digital audio gear, regardless of the price tag, hasits own internal clock. If your equipment has word-clock I/O, trysynchronizing your system first from one piece of gear (e.g., using themixer as the word-clock master) and then the other (using your MDM asthe word-clock master). See whether one setup sounds better than theother one does.
It is usually a good idea to synchronize your system using aword-clock feed that is independent of the digital audio bitstream. Forexample, I typically slave my digital mixer to the word-clock output ofmy Alesis BRC rather than to the clock embedded in the fiber-opticoutput of my master ADAT. The theory is that more jitter will beintroduced if you force the receiving device’s clock-recoverycircuitry to extract the clock from a bitstream full of audio data,which it sees as noise. By feeding a master clock to all slaved devicesvia a dedicated line, you can theoretically keep jitter to a minimum.However, some devices put out horribly noisy word clock, so you shouldalways try synchronizing your system in all possible ways to see whatsounds the best.
The length of your digital audio cables also influences jitter. Thelonger the cable, the higher the jitter. Using a 1-meter cable willmake your signal sound better—typically, a tad warmer, smoother,and more detailed—than using a 5-meter cable. Generally speaking,anything longer than five meters should be avoided.
Notice that I said "digital audio cables." Using standard microphonecables for AES/EBU lines or standard coax cables for S/PDIF lines willgive you inferior results for two reasons: First, both the AES/EBU andthe S/PDIF spec require cabling to have a specific impedance. The wrongimpedance will increase jitter. Second, analog audio (and the cables ituses) has a bandwidth in the thousands of hertz. Digital audio, withits clock signals, has a bandwidth in the millions of hertz. If youdon’t want to screw up the data going from here to there in adigital system, use cables, such as Apogee’s Wyde Eye cables, thathave the necessary bandwidth and impedance for digital audio.
While we’re on the subject of cables, make absolutely sure thatany delicate fiber-optic cables in your studio are well protected. Abreak, kink, or even a sharp bend in a fiber-optic cable spells dataloss, dropouts, and distortion. To protect my six 5-meter Alesisfiber-optic cables on their journey between three ADATs and a mixer, Itie them together very (and I mean very) loosely with twisty ties andsheath the entire bundle in Snakeskin from American RecorderTechnologies. Snakeskin is a flexible, smooth, springy, tube-shapedmaterial that feels a lot like, well, snake skin. You can unroll it tolay cables inside, and it springs back to a tube shape when released.It resists impact, won’t snag on other gear, can be cut to length,and comes in different diameters. It’s a bit pricey, but hey, soare damaged fiber-optic cables!
One last tip on avoiding data corruption: when copying a master tapefrom one R-DAT to another, use the shortest length digital cablepossible, and go straight from one deck to the other (see Fig. 2). Ifboth decks are patched to a digital mixer, avoid the convenience ofrunning the audio through the mixer. The benefit of using a short"straight wire" path between R-DATs is extremely subtle, but it isperceptible. In my own personal blindfold tests, clones made with adirect connection via a 1-meter Wyde Eye cable sounded slightly warmerand smoother, with silkier highs and tighter stereo imaging, comparedto clones made with 5-meter Wyde Eye cables routing the signal throughmy Yamaha 02R digital console.
The exact explanation for these results is hard to pin down. Someindustry experts claim that jitter can affect a digital-domain DATrecording by being incorporated onto the control track, which serves asa clock for the audio samples. Others say this is rubbish and the realculprit is that many digital devices actually change the data they aresupposed to pass through unaltered. Whatever the cause, it makes nosense to route your precious music through anything unnecessary. Bitethe bullet and repatch for the shortest, most direct signal path.
Know Thy Converters!
Most of the time, the tracks on an MDM are recorded at their optimal48 kHz sample rate. That poses a problem at mixdown when a digitalmixer must deliver CD-compatible 44.1 kHz audio to a mixdown deck, suchas a DAT recorder. In that case, you have two choices: use asample-rate converter to convert from 48 kHz to 44.1 kHz (therebystaying in the digital domain), or go through D/A/D conversion (mixeranalog outputs to DAT analog inputs). Which is better? That depends onthe converters you have on hand.
For example, when I routed a mix through the 02R stereo bus’excellent 20-bit D/A converters and then back into the digital domainusing the Panasonic SV-3700’s A/D converters, I heard a small butsignificant decrease in the stereo width, midrange clarity, andhigh-end detail of the mix. Routing the same (automated) mix through aZ-Systems Z-Link+ sample-rate converter, I lost only about half as muchwidth, clarity, and detail as going through the D/A/Dconversions—a major improvement. On the other hand, using the02R’s D/A converters in conjunction with Apogee AD-1000 A/Dconverters sounded the best of all. The lesson? Quality, not function,should be the main criterion for which gear you use. (Don’t assumethat a 20-bit converter will automatically sound better than an 18-bitconverter, either. Specs can lie.) Compare the sound of all theconverters at your disposal, and choose the most flattering signalpath.
Of course, you can avoid sample-rate conversion by recording at 44.1kHz throughout the system, including on your MDM. But unfortunately, ifyou are using an Alesis BRC or other synchronizer whose time-code framerate is sample-locked to 48 kHz (meaning that the synchronizer willspit out one time code frame for every x number of samples), recordingat 44.1 kHz on the MDM will cause the real-time frame rate to slow downby a corresponding amount (approximately 8 percent) from the rateindicated by its format (e.g., 30 fps time code will actually slow downto 27.56 frames per second). This can pose problems for the automationsystems on some mixers. The Yamaha 02R mixer, for example, isespecially finicky about the rate at which it receives time-code framesand can freak out if the frame rate is more than a few percent sloweror faster than its time-code format indicates. In this type of system,you are better off recording at 48 kHz on the MDM and living withsample-rate conversion rather than creating time-code problems for theautomation system at mixdown.
The Path to Perfection
The key to making high-quality digital audio recordings is to getinto the digital domain as soon as possible (allowing for artisticanalog preprocessing, such as compression), keep your levels just belowclipping, use the shortest digital audio cables available, and avoidrouting your signal through unnecessary equipment.
Whether you use redithering or UV22 is a personal choice. In eithercase, remember to choose the most stable clock source to synchronizeall of the digital gear in your studio, and use a dedicated, high-gradeword-clock I/O whenever available. If you can afford the expense, usinga dedicated low-jitter master clock (such as the AardSync II) to runyour entire studio will yield the best results. This assumes that allyour digital gear can lock to an external clock, which is an importantfeature to keep in mind when considering a new purchase.
Aside from feeding your converters healthy levels, which can make ahuge difference in audio quality, most of the suggestions in thisarticle, taken individually, will produce only subtle improvements.Analog techniques, such as proper microphone choice and placement, willtypically have a far greater impact on your projects. But little thingsdo add up. If you’re serious about digital recording, you cansqueeze a little more quality from the gear you already own.
Michael Cooper is a producer, engineer, and owner of MichaelCooper Recording in Eugene, Oregon. Special thanks to Erik Lovell ofAardvark, Andy Moorer of Sonic Solutions, Richard Elen of ApogeeElectronics, Gary Hall of d-House, and Mike Rockwell of Digidesign forsharing their vast knowledge on the subject of digital audio.
A/D converter Analog-to-digital converter, a device thatdigitizes an analog waveform.
bit depth See bit resolution.
bit resolution The number of bits per sample that a digitaldevice (such as an A/D converter, or a multitrack recorder) uses toconvert or store data. The greater the number of bits in a digitalsample, the more accurate the digitized description of theinstantaneous value of the audio waveform. Also called bit depth orword length.
D/A converter Digital-to-analog converter, a device thatconverts digital data into an analog waveform.
dither Random data, or noise, added to a digital signal forthe purpose of moving data from the least-significant bits to "higher"bits in a digital word. Dither is added to preserve high-resolutiondetail and to reduce requantizing distortion. Virtually all modern A/Dconverters add dither to the analog signals they process; therefore,when you add dither to a digital signal, you are actually redithering,or dithering again.
jitter Timing inaccuracies in the transmission and receptionof a digital bitstream, which cause distortion.
redithering See dither.
requantizing distortion Distortion caused by shortening theword length of a sample as a result of truncation. Distortion occursbecause the intervals at which the waveform are quantized becomelarger, changing the original waveform.
truncation The shortening of a digital word, wherein databelonging to the least significant bits is lost.
UV22 A proprietary process developed by Apogee Electronicsthat adds a high-frequency, narrow-band signal just below half thesampling frequency (the Nyquist frequency): around 22 kHz for 44.1 kHzsystems such as DAT and CD.
word clock The timing signal that is used in a multidevicedigital-audio system to synchronize the sampling frequency at which thesystem’s component devices operate. To avoid data loss anddistortion, all digital devices in a system must be slaved to a singleword-clock master so that their sampling frequencies will be exactlythe same.
word length See bit resolution.