Introduction to Mastering
An overview of workflow and concepts

Mastering is an essential part of the production process in ensuring that what is delivered to your
audience is of the highest quality possible. It is also one of the more misunderstood parts of audio production.
This section of my site offers basic information on the mastering process and workflow for clients who are interested in how the process works, those new to audio production, and those interested in learning to master their own material.

Getting A Mix Ready for Mastering

Before discussing the mastering process I want to touch on a few tips for getting a mix ready for mastering. These are very basic as this is meant as an overview, so if you are a client with a project I will be mastering, please contact me for specifics in advance of sending your project for mastering.

Resolution - Send your stereo mix in the same resolution as your mix session. So if you mixed the project at 96/24, then send the stereo mix at 96/24, do not down sample or truncate to 44.1/16 CD standard, as this will be the final step in mastering. Whenever possible, it is best to record and mix in 24 bit, as 16bit gives much less headroom and resolution to work with.

Headroom - A stereo mix should not be sent to mastering at its final volume. Ideally there should be nothing on your master channel boosting the volume of your mix, and it should definitely not be hitting a limiter as this gives little to work with in mastering. Try to have peak levels at no higher than -3dbfs (preferably more) to give enough headroom for mastering.

Beginning and Endings of Tracks - Leave plenty of space at the beginning and endings of tracks, if there is a whole minute of random noise before and after the track that is fine. Like headroom this gives more flexibility, as a track that goes from digital black straight into the music can be jarring. Also if there are buzzing amps or other noises, do not mute these in the moments before the music starts, if using noise removal there needs to be a clean sample of the noise to make an accurate profile for removal.

Live Recordings - For live recordings, if possible send the entire program whole, without taking out applause or dead time between selections. These moments can be important in creating a realistic ambiance for the album.

Use Saturation and Effects in Moderation - While getting just the right tone to your mix is important, keep in mind that mastering can amplify these effects since the overall dynamic range will be reduced. So if you are saturating almost to the point of distortion or have crazy wet reverb, mastering may push the effect over the edge into what does not work, meaning a master may not be able to have the level as high as you may desire without negative side effects. You can also add a limiter to your master channel to reduce the dynamic range to get an idea of how some of the more exaggerated elements will translate.

Listen In Mono - A great mix will sound big and full in mono. With most playback applications being in stereo the art of mixing for mono seems to be often overlooked, but is an important part of good audio fundamentals. If you hit the mono button and parts of your mix disappear or the volume drops substantially, then you likely have phase issues that will hurt the overall clarity and quality of the sound. If you are not used to working in mono, try it out, often if you are having trouble getting the balance just right, try mono and you may be surprised how big it will sound in stereo if it is well balanced in mono.

How To Learn Mastering

When leaning about audio production there is an abundance of information on recording and mixing, but when
searching for information on mastering the most common response I've seen is "just have a pro do it." So why
the exclusivity for mastering? Is it something only the elite few can achieve? And what makes it so unapproachable to those who love making music but do not do it as a career? Mastering is like any other skill,
and can certainly be achieved with enough practice and study. And this opportunity to get enough practice is where the difficulty ultimately lies. Learning the aesthetics and technical aspects of audio mastering takes a great deal of time and practice in combination with having a highly developed ear that can discern the many small details in program material as well as understanding the sonic standards of many diverse genres of music.
So for an individual working on just their own material, having the volume of work necessary to build this experience is going to be difficult.
Personally before leaning mastering I had many years of formal ear training as part of my background as a classically trained musician, had studied thousands of recordings, and then leaned mastering from an engineer with decades of experience. Even with this training it was several years and hundreds of projects before I felt that I could act intuitively when mastering and know that I can approach any genre with confidence.
So for those wishing to learn about mastering, my advice is to keep working and be patient with yourself. Mastering can very easily do more damage than good, so begin conservatively. Do not try to do everything at once, work with one element at a time, and learn how each process effects the sound. Compare your work to finished albums you like, and keep working to get closer to them, first in tonal balance, and then later in level.

Pre Mastering or Mastering?

The process that has become known as 'mastering' is more accurately called pre-mastering. The audio process
begins with the recording, this is then given to a mix engineer who shapes the many individual channels of the recording into a two channel stereo mix. The stereo mix is then given to a mastering engineer who makes the final adjustments to the audio and creates the pre-master. The pre-master is then delivered to the manufacturer or digital distributor. This final process of creating the physical master, or encoding the digital master, is technically mastering, where the actual master for distribution is created. For the purpose of this
article, pre-mastering will be referred to as mastering.

Metering

When mastering you will often be required to deliver material that meets exacting standards for broadcast, distribution, video, client request, etc. Additionally to ensure your material plays back at safe and appropriate levels without distortion you will need to understand proper metering as well as various broadcast standards. Below is an overview of different meters I use when mastering and how they are applied.

Decibel - Definition -Before beginning the section on meters we should first establish what exactly a decibel is since every sound meter will most likely be in decibels. A decibel (deci-bel, 1/10th of a bel) is a logarithmic ratio used to define the the ratio between two power quantities, like intensity or acoustic energy. So for example an increase of 3db will be roughly 2X the value of the source, while an increase of 10db will be 10X the value. Since a decibel is a ratio, when using it in general terms (when you see something expressed just as db) the ratio will be from a given origin point. Decibels can also be used to describe absolute values when they are attached to a fixed reference point; you will see the reference after the db, such as dbfs, dbu, db-spl, etc. When working with digital sound we will often see dbfs or decibels relative to full scale, where the reference point of odbfs is the maximum peak amplitude level in a PCM digital system. Because this scale is referencing a maximum, its values are given as a negative below this maximum point. When working with analog gear you will often see dbu which references 0dbu = 0.775 volts unloaded. If we are looking at loudspeaker output it may be in db-spl which is referencing a sound pressure level of odb-spl = 0.0002 dynes/cm2, or the threshold of human hearing at 1khz. So you can see the decibel can be used to describe a large array of different and not always compatible values, so while this is a basic concept, it is very important in being able to understand what our meters are telling us.

Peak and True Peak/Inter sample Peak Meter - This measures the peak level of your material. For mastering it is important to use an over sampled meter that reads the True Peak or also called Inter Sample Peak levels. Inter Sample peaks occur between the samples, and can cause distortion if clipping occurs in a digital to analog converter - aka - playback device. For CD my max allowable peak level is -0.3dbfs, with a true peak max of -0.2dbfs. For broadcast and Mastered for iTunes material the true peak level should not exceed -1.0dbfs.

RMS Meter - RMS (Root Mean Square) is basically the average level. Its slightly more complicated than that, but for our purposes this will suffice. In mastering since our peak level will be near maximum most of the time, it does not give us much useful information of the loudness of our material, so using RMS levels gives us a much more meaningful figure. This can also be used to determine dynamic range or crest factor, which is the difference between RMS and peak level. If our peak level is -1dbfs and the RMS is -20dbfs, then we have a large dynamic range of 19db; while a peak of -1dbfs and RMS of -6dbfs would be a very small dynamic range of 5db and would sound much louder. Basically the higher the RMS level, the less dynamic range in the material, and the louder the material will sound. Before using your RMS meter, check to be sure that it is compliant with AES specs, as many meters out there will give readings that are off by as much as 3 or 4db.

K Meter - This meter was devised by respected mastering guru Bob Katz, and is a combination of peak and RMS meters with different zones specifying target ranges. There are three K Meter scales: K-12 with a 12db dynamic range meant for pop and broadcast material; K-14 with a 14db dynamic range meant for rock type material; and K-20 with a 20db dynamic range meant for very dynamic material, theatrical material, and classical music. In addition to mastering levels the K system also includes specifications for standardized monitoring levels. I find this a useful reference and utilize many of Mr. Katz ideas in my mastering setup, however I do not master strictly to this scale as I find it a bit too conservative for the current mastering climate, as these scales will produce masters that fall below the current expected levels and put my monitor volume outside what I find comfortable.

Loudness Unit Meter - LUs are close to RMS levels in that they are both an average and will look similar on a meter, but the LU is weighted using filtering to make it more useful for broadcast material. An LU meter will generally have three different readings: momentary, short term, and integrated. I tend to focus more on the integrated level, as this will give an overall average for the level of the entire song. For a rock album I generally have a target integrated level of between -12lufs and -10lufs, depending on the material. The standard for broadcast is -23lufs.

VU Meter - A VU meter on your output will give you a musical average and the needle ballistics can give you valuable information about the actual level of your music. If you are having trouble with balance, watch and listen for which element is making the needle jump. For mastering rock and popular music styles I calibrate my VU to 0VU = -10dbfs. For classical and acoustic styles it will be lower, depending on the genre. Generally the meter should be hovering around 0VU on the chorus or loudest sections.

Frequency Spectrum Analyzer - Using either an FFT or RTA analyzer will show the frequency balance of the material. While it can be tempting to watch this meter when EQing your master, trust your ears and use the analyzer to help identify problem spots.

Stereo Analyzer / Stereo Oscilloscope - Used to reveal phase errors. Phase problems are especially an issue for material to be broadcast which should have a high degree of mono compatibility. If phase errors are extensive and the material is highly out of phase, I may ask if a remix is possible.

Spectrogram - This shows a visual representation of audio in duration pitch and intensity. I generally only use a spectrogram when working to identify problems, such as feedback or intermittent noise, from the audio.

While meters are great tools, in the end its about how it sounds. While I will often have targets to shoot for,
I still must individually evaluate the material. If I am shooting for an integrated loudness of -10LU but
the material is very dynamic, I will need to reevaluate and aim for whatever in the end produces the best result.

* For more information on Loudness and metering I recommend the materials available through the EBU Website

Mastering Workflow

This is a step by step general walk through of my mastering process. This is by no means The Correct way to approach a master, but it is a workflow that I have found successful and it may give you a starting point if you are working to learn. My workflow is of course not limited to this, but this is how most projects generally proceed. I will speak about concepts in general terms to keep this to a reasonable length, but feel free to contact me with any questions about specifics.

Analyze the Material - Part of mastering is about quality control, so when beginning the process the first step is to both look and listen through the material and take note of problems. The problems may include: phase issues, especially if the master is in stems (separate stereo tracks for each of the groups - ie: vocals, drums, guitars, etc); destructive noise like clipping, buzz, pops, clicks, etc; and in the case of live recordings dropouts and intermittent noises.

Once these problem areas have been identified I will make the necessary corrections before moving on toward addressing the musical issues.

Editing - Any cuts and edits should be made at the beginning of the process, as they can alter the level of the recording whenever a cross fade is used. In general it is best to make edits in the mix phase, as editing a multitrack source can be done more easily and transparently than a stereo mix since each track can be spliced separately to ensure the cut is optimally placed. However, transparent cuts can certainly be made in the the mastering phase, but they must be approached with care.

Listen and Take Detailed Notes - Listen through the material a couple times, focusing on specific issues on each listen through. Taking accurate notes is important, as the human capability to recall sound in detail is not as accurate as we would like to believe. If the mix is overall in good shape and well balanced, I may not need to write much as the process may be fairly straight forward, but being thorough is important. This must be repeated throughout the process.

Stereo Balance - For most projects the goal is to have a recording that sounds centered and balanced across the stereo image. This may include adjustments to the left right balance, proportions of the left and right side in the stereo image, and centering of the focus point. If the material is off center, while it may even sound ok on speakers, it will often sound much more exaggerated and lopsided when listening on headphones. This is especially an issue with classical recordings, where microphone and ensemble positioning may make the ensemble weighted to one side.

EQ - The most important concept for EQ in mastering is BALANCE. It is very easy to overdo the EQ, cranking up the bass or treble to get the sound really in your face, but when working to get a great sounding master restraint is often the wiser course of action. Start by listening for what is being obscured, and gently cutting elements that are covering them up. I always begin by cutting, as boosting runs the additional problem of increasing the overall level, and taking away much needed headroom. The most common problem I encounter is too much bass, especially sub bass, in the mix. This is often due to a less than ideal monitoring environment or incorrectly calibrated monitors at the time of mixing. Or lets face it, some people just want to crank up the bass. The louder you want your master to be, the more balanced the mix will need to be.
In mastering automating the EQ is often necessary and can be a more transparent way of working rather than picking a setting that generally works for the entire song or project. An example would be a chorus that needs to be livened up, using a mid-side EQ with a high shelf to boost the high frequencies on the side signal in just the chorus can help open it up and give that extra sparkle needed to add excitement to the moment.
In mastering I use EQs with both Finite Impulse Response (FIR) and Infinite Impulse Response (IIR) filters. These are more commonly referred to as Linear Phase and Minimal Latency EQs. Each has its pros and cons and uses in the mastering environment. A linear phase EQ will not change the phase relationships of the frequencies being altered, but introduces a pre ring distortion. The minimal latency EQ may cause phase relationship issues, especially on large changes, but its post ring distortion may be less noticeable, making it a better choice for some applications. Until you learn which EQ works best for you in varying situations, try both, and choose whichever sounds best to you.

Dynamic Control - aka - Compression
Contrary to popular belief, mastering is not about taking a mix and crushing it with a compressor. While many mixes may have fallen victim to this in the pursuit of the loudness war, many mastering engineers are thankfully backing off this idea that the most hyper compressed lifeless screaming loud mix wins. In reality I often use little or no compression in mastering, and when I do it is to correct an issue, or enhance the mix in a way that gives it more punch or impact. Automatically compressing a mix can drastically alter its balance, and compromise the musical intent of the mix engineer. Below I will go over some of the basic types of compression I employ in mastering, and their uses:

Downward Compression - This is your basic compression. I often will use very little compression at a fairly low ratio (around 2:1), compressing only 1/2 to 1db. This is done if the mix overall is solid, but feels just a little bit loose. A very small amount of compression can help solidify the mix and soften a few peaks before leveling. If there is an element causing harsh transients, like a snare that is too loud for example, I will use a very fast compressor to reduce and even out the peaks. If the groove seems a little weak and needs to be strengthened a slower compressor can be used, emphasizing the the transients and giving the track more bite. When setting the attack time listen for how the compressor is effecting the transient, or attack of each note. It often helps to over compress to exaggerate the effect letting you better hear the adjustments, then remove the compression entirely and slowly bring the threshold back down until you get your desired effect. When setting the release you also need to pay attention to the transients. If the release is too long the compressor will not return to zero gain reduction before the next beat, dulling the beat overall. If the release is too short it can result in a pumping sound, or even in distortion. There are some genres where pumping is desirable, such as dance or club music, but even in these settings it should be used carefully as the effect can be distracting and fatiguing to the ear when overdone.

Upward Compression - Parallel Compression - New York Compression - You can find all these terms basically meaning the same thing, taking a copy of the mix and heavily compressing it, and then mixing it with an uncompressed copy. Because all of the transient information is retained in the uncompressed copy of the mix, this can be a transparent way to add compression while not sounding compressed. There are basically two ways I approach parallel compression:
The first is the more transparent way of employing the technique. Set the compressor to a very fast attack and med to slow release with a fairly low ratio, around 2:1, and compress the mix heavily until it is basically a uniform dynamic. Mix this in fairly conservatively with the original, just until you hear the overall sound take on more body and thickness. This technique is especially useful with acoustic material where the softer elements need to be strengthened. Overusing this technique can cause the sound to become muddy, and can introduce distortion in the heavily mixed signal if clamped down too much.
The second approach is similar to using a compressor to add bite to the groove using just downward compression. Using a slower attack and higher ratio, generally closer to 4:1, compress one copy of the mix until you are getting a clear increase in its transient clarity. You will want to compress a fair bit, but not until the mix is completely squashed. Then mix the compressed copy with the original, again be conservative, a little can go a long way and it is easy to muddy up the sound if overused.

Multiband Compression - Whether or not to use a multiband compressor in the mastering environment is an often debated concept among mastering engineers. I do not see any reason to exclude a tool that may be useful, as long as I am aware of its limitations and potential drawbacks. When used subtly it can help solve balance issues or bring out an element without sounding boosted. A common example would be a wind band recording where the mics were too close to the flutes and pics, using a multiband to tame the high frequencies without altering the rest of the balance. Be careful when using a multiband compressor, as it can drastically alter the balance of the mix, so I find it best to be used sparingly and only when necessary. Also it is recommended to use a linear phase multiband to reduce crossover distortion when working with program material.

Expansion - Using an expander can open up the mix if it lacks dynamic contrast. When using an expander I find it most useful to automate it, using it in select passages such as opening up the dynamic range of a chorus. In general if you feel the mix is too compressed when it is received in mastering, an expander can help add a little life back into the mix, but it cannot work miracles.

Riding The Fader - By far the most transparent approach to controlling your dynamics, automate the master fader. In general I prefer not to use compression on acoustic and classical music. If the recording is balanced then all that should be required is to make adjustments with the volume fader, but if the recording is not properly balanced compression may be used correctively (ie - if the brass are overbalancing the strings a compressor can correct this imbalance by not allowing the brass to get as loud). But in properly balanced recordings turning up the volume is soft sections and then down in loud sections can control the dynamics in a way that makes the music listenable while sounding completely transparent and natural. My general rule with classical and acoustic recordings is that you should be able to listen comfortably on high quality headphones without having to adjust the volume throughout the piece. Riding the fader is also used to enhance popular styles, often boosting the chorus by as little as 1db can add a big sense of impact, and bringing down a raucous outtro a little can keep it from being pummeling and over the top.

Sweetening - In mastering sometimes the sound or tone of a mix will need to be enhanced. If a recording lacks warmth or depth, adding some saturation, generally tube or tape, can smooth the edges and round out the sound. If a mix is too dry and stark reverb can be applied to give a little more depth. Reverb should be added carefully and subtly, often using just early reflections can open up the sound in a transparent way. Often reverb applied to an entire mix will have to be aggressively shaped using EQ, and even compressors and gates in some cases, as the various elements will not always respond in favorable ways.

Fades, Crossfades, and Other Creative Changes - All creative volume changes, such as fade ins and outs, crossfading between songs, fader rides, etc., should be made prior to the leveling stage.

Leveling - The moment we've all been waiting for - how do we make this song LOUD!
With so many new tools available and more powerful computers, its never been easier to destroy a good album
by crushing it into submission with compressors and limiting to make it loud enough to cause your speakers
to explode. Lucky the trend of hyper loud masters and the quest for ever reduced dynamic range is starting to
reverse course. With more and more music going to a file based format and being consumed through streaming services and online broadcasts, as well as mobile hardware manufactures trying to offer a better listening experience, the future is headed toward loudness normalization. With loudness normalization, songs played back to back will be lined up to playback at the same level using their integrated loudness levels. So a hyper compressed master will not sound any louder than a more moderately loud master with more dynamic range. In this scenario the song with greater dynamics will most likely sound bigger and dare I say... louder. Loudness normalization will not alter the internal dynamics of a song, so if the hyper compressed sound is something you desire, it will not take that away.
So with this in mind, the following are ways to level your audio, with some guidelines for how loud they should be. While this article was written as a basic workflow guideline, when it comes time to level the material many adjustments must often be made, so the earlier steps will be revisited throughout the
leveling process to be sure the master remains balanced.

Limiting - I am going to focus the discussion of leveling with brickwall limiting, because it is the more widely accessible method for raising the level of your music. Most DAWs will come with a basic brickwall limiter, for example Pro Tools' Maxim, but be aware that many of these offerings will not provide the same performance as the more expensive tools aimed at professional mastering, so listen carefully as it can be easy to push them past their usable limits.
Start by setting up the output ceiling of your limiter. As stated earlier, for a CD my output will be -0.3. Using a true peak / inter sample peak meter you can be certain the max peak level does not exceeded this, but if you do not have one and are unfamiliar with the output characteristics of your limiter, ie - how high will the inter sample peaks be for this type of material, err on the side of caution and set the output to a lower level, like -1db. Some limiters have many other settings, and for brevity I will not go into other settings for limiting, but recommend you take the time to carefully read the manual to get a better understanding of the various settings and how they will effect your audio.
Start adding gain to the material (or bringing down the limiter threshold, depending on the setup) while keeping an eye on your meters and targets. At this stage I find the VU meter most helpful, as I get an idea of where the level stands musically. I will want the needle hovering around 0VU (this will depend on the calibration of the VU meter, which for me is generally 0VU=-10dbfs for rock and around -18dbfs for classical) in loud sections, and not getting too far into the red in the climax moments. I begin by watching the meter because it can be easy to push too far if not hitting the distortion point. The goal is not to push it as far as it will go, but to reach the target level or where we will get the best compromise of dynamics and level. Listen first for distortion, if you are getting a lot of distortion at fairly low limiting levels there is probably a buildup in the mid and lower frequencies that may be addressed by going back to the EQ. Fine tuning the settings of your limiter may also be required where applicable, as there is no one size fits all setting. When using limiters I have several different models that I use depending on the application and genre. Next listen for the dynamics, more specifically the beat or groove. A limiter works by removing the tops of the transients, so the big downside is that limiting can easily suck the energy and life out of a mix. At this point it becomes a balancing act of EQ, compressors, and limiting to find the optimum compromise where we are getting an exciting product at a reasonable level. At this point I recommend using level matching to compare the processed and unprocessed sound. Lower the output ceiling so the output sounds the same when the limiter is switched on and off. This way we can truly evaluate the final product, we are easily deceived by the louder material, as our ear will generally think it is better. It can be shocking to hear how bad an over limited master can be when level matched with the original source.

For classical and most acoustic material, the peaks should not hit the limiter with the exception of loud percussive sounds or hard transients, a trombone rip in a brass quintet for example. The level in a classical master should be controlled almost entirely by riding the fader and careful compression, if your recording is very quiet, go back and reevaluate choices in these stages and do not push the material into the limiter so that the musical peaks are being actively effected. The leveling stage should generally be just about adding gain to bring the material up to the proper level.

There are other techniques used in leveling to increase either the average level or perceived level of the audio, such as clipping (using a digital clipper or clipping a high end A/D converter) and saturation. These are often genre specific, and I do not recommend attempting them until you have a firm grasp of the basics. And hopefully as we move back towards more exciting and dynamic masters, these techniques will be required less often.

Peak Normalizing - Peak normalizing measures the distance from the highest peak to the ceiling (odbfs or a user specified ceiling), and then increases the gain for the entire file by that amount. I do not use peak normalizing and do not recommend it because the peak levels often tell us little about the actual loudness of the material, as our average level gives the most accurate representation of loudness. If we peak normalize we may end up with masters that are way off from the desired target volume. A good example would be a recording of a harpsichord. The harpsichord has a fairly limited dynamic range, when close miced generally the difference between pp-ff in the low register is around 5db, while the difference in the upper register is 1-3db. The harpsichord also has a fairly low crest factor or peak to average ratio. So if we were to normalize a recording of a harpsichord made at a close distance to the instrument, the result would likely be in line with a hot rock album. Just because we have available headroom does not mean we necessarily need to use it all, as the material needs to be presented in a way that falls within the normal range for its genre.

Dither - Dither is randomized noise (basically shaped hiss) added at very low levels, ~-96dbfs for 16bit dither, to the signal which modulates the least significant bit when truncating to a lower bit depth. The result is increased resolution in the lower levels. The max signal to noise ratio remains at 96db for a 16bit word length but the actual dynamic range could be as much as 110-115dbs as musical elements can be heard below the noise floor. This also removes quantization distortion. A good way to visualize this is by thinking of a sine wave. Without dither in a digital system this wave would be blocky and square, like a sine wave made up of stair steps, and by adding dither it will be smoothed out and rounded. Dither is a rather complicated mathematical process, but it is very important in digital audio and I recommend further reading on the topic. The important thing to understand is that dithering should be applied in the mastering stage as the final process before truncating to a lower bit depth, for example if your master is 24bit and you are reducing it to a 16bit CD format.

CD Layout - Once the master is ready to be completed, if you are making a CD you will need to lay out the track markers, gaps, and CD text. If you are using a DAW designed for mastering this process will likely be able to be completed within the same program you have been working. If you are using a more generalized production DAW, like Pro Tools for example, you can either create the master by using a third party software plugin, or by bouncing the file out and importing it into a stand alone CD authoring program. iTunes or the burning software that comes bundled with your computer are generally not suited for this task. To deliver a master you will need software capable of CD text, ISRC, and UPC codes as well as being able to create redbook compliant masters and DDP images with MD5 checksums and CUE files.

CD Delivery - To deliver your finished product meant for CD to the client or CD manufacturer the two most common delivery types are by a physical CD master or a DDP (disk description protocol) Image, which is a digital file that contains all the information for printing the CD. A DDP image is the safest and most convenient way to deliver a CD, as the digital file includes data security checksums to ensure it is recreated properly and it can be uploaded directly to the client. A physical CD runs the risk of burning errors, so after burning each copy it is important to run a diagnostic to be sure any errors are within acceptable limits.

Encoding - Because of music is now mostly consumed on devices that store music in compressed files, proper encoding is an essential step. Most large online distributors, like iTunes, will do the encoding for you so you will need to deliver the audio in PCM uncompressed format that is at least CD quality (44.1/16), and some will accept high resolution masters (48/24, 88.2/24, 96/24 and up). For iTunes they also have their preferred specifications which they refer to as Mastered for iTunes (for more info: http://www.apple.com/itunes/mastered-for-itunes/).
If you are doing the encoding yourself it is recommended to use software designed for mastering applications that will allow you to identify masking at different bit rates and see where overs are occurring. Whenever possible it is best to use the highest bitrate possible when making Mp3s or other compressed files. While all compressed lossy file types will be lower in resolution than a master, this can be controlled and reduced with careful processing. When lowering the bitrate to try to reduce file size, the higher frequencies will be effected first. If you absolutely must deliver small files, this can be compensated for in the mastering phase with software that lets you audition the sound of different codecs and bitrates. The biggest overall issue with encoding is overs, or clipping. Encoding will generally raise the peak level of your material, so you will need to lower it before encoding to be sure that clipping does not occur. An Mp3 or ACC file that is clipping will distort on playback, and can ruin a recording.
Encoding to lossless formats, like FLAC or ALAC, is much simpler, as the quality of the master is retained.

Broadcast - Mastering a project for release on the radio can be tricky, as all stations will not process your material for the airwaves in the same way and broadcast processing in general can be fairly extreme. I have mastered both classical and rock material for radio, internet streaming radio, including NPR, and streaming video and in my experience have found that backing off compression, concentrating on a clean frequency balance, paying special attention to be sure the bottom end is not overly hyped, as well as enhancing mono compatibility will yield the best results. For classical projects I will use more manual gain correction with the fader to keep the dynamic range more uniform, but will not lean on a compressor. For rock projects I find that leaving the dynamic range more open, especially if the CD master is hot, will make it more yielding to broadcast processing. This may initially seem counter intuitive, since broadcast is working with a limited dynamic range and needs to ensure their material is listenable in higher noise situations. But this can be equated to taking a hot compressed mix, and then trying to compress it further; in my experience I have found better results being more conservative and letting the broadcast station crush it to their liking. When working with material that is being sent for broadcast I also leave additional headroom to give them room for conversion and processing. For broadcast I send material with a max peak level of -1.0dbfs.

So What Gear Should I Use?? - I have purposely left specific gear out of this article, as the most important tools for mastering are your ears and what lies between them. A good engineer can coax a great sound out of whatever they have to work with, and our audio forefathers made incredible sounds with a fraction of the tools we have today. So if you are new to audio production and are interested in mastering, start with what you have and are familiar with, even the basic bundled plugins that come with your DAW can produce fine results. As you progress and learn, you will start to see what elements are holding your master back, and then you can start investing in what you know you need, rather than what some article, blog, or forum said you should need. Just like learning a musical instrument, the first step is a firm foundation in fundamentals. You could spend a fortune on mastering gear and absolutely pristine converters, but without an understand of the basics they will not do the work for you. All that being said, if you would like to know what tools I am using or what I recommend, feel free to contact me, I like to talk gear just as much as the next guy.

I hope you have found this informative and useful. I was lucky enough to have mentors who were very supportive of me and who answered my endless questions when I was first starting out, so now try to return the favor for anyone interested in learning about audio production.
So please feel free to contact me with any questions you have.