Limiters and compressors

October 15, 2016
 by Paul McGowan

Recording studios have a wide variety of choices when it comes to squeezing music’s dynamics. This squeezing can be for a variety of reasons including, increased loudness, and insurance against clipping.

To perform these tasks two pieces of technology are typically employed, limiters and compressors. I find both terms to be a bit confusing, because limiters can compress, and compressors can limit. Let’s see if we can unravel what’s going on.

The basic idea behind a limiter is to reduce peak volume levels, while a compressor’s function is the opposite, turning the volume up on soft passages. Depending on how they are used, recording engineers can get very different results. Let’s take the limiter as our subject for today.

A simple peak limiter is an automatic volume control that senses when fast moving audio peaks occur and turns down their level. In this use case the limiter is compressing the maximum dynamic range into a smaller space, so we might say the limiter is compressing, though that doesn’t mean we’d hear it doing so.

Here’s the thing. Properly used and set, music run through a limiter won’t sound compressed. That is because the ear won’t detect overall level changes when they apply only to occasional peaks. But how does a limiter distinguish peaks from steady state music? Timing.

Most limiters are built with timing smarts that help them sound natural: fast transients are reduced without affecting overall level because the volume adjustment is applied only to the transient. You don’t “hear” a well setup peak limiter. Change timing settings incorrectly and listeners experience “pumping”. You might have heard some radio stations do this where a plosive or a loud transient turns the overall level noticeably down, then the sound slowly comes back up, as if a hand is slowly turning up the level.

True compression can easily happen with limiters in the wrong hands. An engineer wishing to increase overall loudness of a track or album, need only turn the level up so the softest passages are loud. Now, when the loud passage in the music comes, the volume is automatically adjusted down to match that of the softer levels and you get this obnoxious compression ruining the original dynamics.

Tomorrow, compressors.

Subscribe to Paul's Posts

14 comments on “Limiters and compressors”

  1. Low powered tube amps have a similar effect by clipping the peaks when a loud transient comes along. Usually, this goes unnoticed because they are said to clip in a benign manner. But when a complex orchestral assault comes along, it can be quite noticeable. This is a case where careful amp and speaker matching can be very important.

  2. By this time there’d usually be at least half a dozen comments but today there are none. Is it because it’s Saturday? Maybe? Maybe not.

    Circuits that perform dynamic compression, expansion, peak limiting, automatic gain control are very useful not just in hi fi technology but in many other applications as well. So today I’m going to post about the essence of the concept related to an information window (not Microsoft Windows) in a simplified form…without any math (I hope because the full blown explanation has a lot of it and it’s hard) and several months of hard won knowledge will be compressed into a few (I hope) short paragraphs. This subject deals with among other things deliberately non linear systems and is a senior level course for EEs if not a course taught at a higher level (my schooling was as tough as they could make it.)

    The concept starts with a window. The height of the window is dynamic range. The width is frequency. There may be a grid like a screen in the window. This grid defines the allowable detectable changes to parameters within the grid and is what is typically known as resolution. Digital systems invariably have them, analog systems theoretically don’t but don’t always count on it. The window is not necessarily rectangular, in fact it can have any shape.

    At the top of the window is saturation. An increase to a signal above that point will not make it through the window. The bottom of the window is the noise floor, the point at which the signal is obliterated by inherent noise in the system.

    Information that falls within the window’s borders will pass through it, those that don’t will be blocked.

    One window you can think about is the window of your own hearing. Now how is that window determined? By listening tests and it is not the same for everyone. In fact not the same for the same person at different times. The size and shape of the window is the outer envelope of the Fletcher Munson curves. The top curve is the amplitude of sound where it is no longer sound alone but pain. The low end of the curve is the threshold of hearing. Because of its usual purpose, the extreme parts of the graph may not intersect but trust me, the full curve always will. This is the window that defines what you can and can’t hear. Audiophile bullshit aside, this is the real deal Another window is the envelope containing the sound of music. This can lie inside the first window or partially outside it such as harmonics above the highest frequency you can hear. Whatever physiological effects ultrasound can have on your body (it can injure or even kill you if it is loud enough) it is of no use in listening to music. In fact systems capable of ultrasound can degrade performance within the audible window. Sounds near the top of the window while audible as sound can also temporarily or permanently injure you hearing especially if exposure is sustained or repeated. In fact a single audio shock wave can deafen you for life.

    Now for the window of the recording playback system. I’ll only talk about electrical signals here. The elements of a recording playback sound system consists of many windows through which the electrical signal must pass. Each window is a different size and shape starting with the microphone electrical window and ending with the speaker crossover network. Microphones and speakers have their own mechanical windows but that would make this discussion too complex.

    The windows in a recording/playback system each have a different size and shape as I said. In aggregate you are hearing through all of them one after another after another. The total window is restricted by the smallest area covered by all of them stacked together. Pointless to record loud deep bass on a vinyl record if every phonograph cartridge has a window too restricted to track it. The problem is to make the electrical signal suitable to pass through ALL of the windows. To do that it must be deformed, distorted, and then when it is through a restrictive window it must be inversely distorted if it can be. Clearly the manner and sometimes the level calibration of the inverse distortion must be precise, especially if the distortion is non linear. In the case of linear distortion, usually the loudness range of the bass is too great and would overload the system, the loudness of the treble too low and would be obliterated by noise. This is true for vinyl records, magnetic tape, and FM broadcasts.

    One use of a limiter is to prevent overmodulation of a radio signal. As the audio signal gets louder, the deviation of the broadcast frequency from the nominal carrier frequency gets greater. This could interfere with adjacent radio stations creating what is called “heterodyne distortion. It is illegal and the FCC imposes fines for doing it. So even with digital pop recordings, producers want their product to be the loudest sound on the radio. By compressing dynamics and overloading the system creating harmonic distortion the limiter makes the signal as loud as is legally possible.

    In hi fi systems there are other problems getting the signal to fit inside the overall window or parts of it. One big problem is with tape recorders especially when recording music with inherent great dynamic range. Notice I didn’t say loudness, I said dynamic range, the difference between the softest and loudest sounds. In addition to peak limiting to avoid saturating the tape, other methods are used. One is gain riding where the recording engineers twiddles the loudness control on the recording amplifier manually. Compression generally makes the softer signals louder to keep them from being obliterated by tape hiss or at least minimizing tape hiss while at the same time compression limits the loudest signal by reducing gain above a certain level so that the signal doesn’t get clipped by saturating the tape. The limiter is there as a backup. The ultimate equalizer/compressor/expander for hi fi audio systems may be Dolby A Professional. It chops the signals into four frequency bands, compresses them each separately where the degree of compression depends on the instantaneous signal strength in each band, is optimized for the part of the spectrum where tape hiss is most objectionable, reassembles them and then records them. It breaks them down into four bands again on tape playback, precisely applies inverse expansion, and then combines them again. If properly calibrated and used correctly the system works very well to reduce tape hiss. If not….it can be a disaster.

    In wow and flutter the horizontal part of the scale, the frequency axis is vibrating back and forth. This means the frequency you get out of the window is not necessary the frequency you put in. It warbles.

    So, for all of you audiophiles who believe in the purity of your vinyl recordings and systems, welcome to the Vinyl Record Sausage Manufacturing Plant. Welcome to the real world. Too bad we can’t show you the full plant, there’s a lot more to making these sausages. What isn’t processed here? Digital Compact discs. That window is inherently large enough, stable enough, and has sufficient resolution to not need any sausage making machines. However, in noisy areas, even there dynamic compression may be desirable. So all of you audiophiles, you’ve been listening to compressed, limited, equalized, sausage sound all of your audiophile lives. That is your reference. And that’s only part of it. The sound itself undergoes enormous geometric compression and distortion by the microphone and speakers but that is for another day. So why do I think audiophiles don’t know the real deal when they hear it? Because almost all of what they do hear is “CANNED SOUND!!!” And of those that do know the difference an laud this expensive equipment anyway, IMO they are knowingly lying, usually to make money.

  3. The question I have for everyone here regarding compression has to do with the differences between how we perceive a piece of music with large dynamics performed live vs. the perception of that dynamic range when played back on your stereo systems. I’m sure there are a number of us who have systems that can “handle” truly high dynamic range – Paul’s likely being one of them.

    When I listen to some “uncompressed” recordings – classical recordings in particular, which are often taken as the test cases for this sort of discussion, due to their being (potentially) unamplified in any way at the time of performance, and untreated in post with compressors – I often find that if I set a “comfortable level” for the loudest passages, the ppp bits are not naturalistically loud.

    That is, when I am at an orchestral performance (assuming a decent seat), I NEVER feel like it’s either too loud or soft – I don’t wish for the orchestral to play louder or softer.

    But when I play a recording of same or similar, I often want to turn up the quiet parts to hear the detail/texture…then the fff passages are just too dang “loud” and I’m reaching for the volume control.

    Anyone with big, dynamically-capable systems please weigh in. Is this just because my system (and most others I have heard) are dynamically compromised? Or the rooms they are in…

    1. The sound is not just compressed in loudness, it is compressed in space and time as well. This is a result of a flawed concept.

      In the real world, musical sound comes mostly at you from every direction unless you sit right next to the performers the way microphones do. In listening to recordings sound comes at you mostly from two, three, four, or five point sources, especially at high frequencies. You brain can easily tell the difference. The second way, the recording is perceived as blaring, the first way it is not. Sound also naturally decays with time, loud sounds typically about taking twice as long at mid frequencies compared to high frequencies in very large rooms. Sound decays at least ten to twenty times faster in your room. These are among the distortions that matter and nobody making equipment or recordings for this industry has figured it out yet. This is why recordings sound like canned music to those who hear the real thing and are familiar with it… is canned music tightly squeezed and packed. To those who are not familiar with it, it could hardly matter less. To them their two dimensional world of sound is all that exists.

  4. Because the other side of this coin/question is that there are certainly classical recordings that have been mastered “judiciously” with someone with an ear for natural-sounding dynamics as they are/can be reproduced on the majority of systems. And when it’s done right, you would be unaware that compression was used.

  5. The idea that humans can’t hear fast peaks and also don’t hear their removal is false. I know there is a mountain of data confirming this premise. I have sat in rooms full of audiophiles at “High End” shows and at the Connecticut Audio Society; rooms full of audio engineers at meetings and conventions; and concert halls full of lifetime subscribers who did not notice the presence or absence of fast peaks – but I do.

    I am not alone either, like a deluded crackpot with a vivid imagination. The problem is you need to hear real fast peaks regularly, like daily for a period of conscious listening; and also NOT hear music with fast peaks removed as often in order to remain sensitized. So is all recorded music limited? Not quite, but very few people can make the choice to only listen to uncompressed music.

    How can you tell if the music is not limited, given that you have a lifetime of listening to limited recordings? The most obvious sign is the average level of the recording is 15-20dB lower. You start the disk and it sounds like you have to rotate your volume control 45 degrees clockwise – and you probably have to go further than this to achieve realistic levels, by which time something in your system is clipping.

    What about “Reference Recordings” or vintage Decca and Red Seal? Surely they didn’t use a limiter? And what about the people who inherited their seat in Carnegie Hall from their grandparents and have gone to symphony since they were children? They are all subject to acoustic compression by short, medium and long range effects.

    Sound waves launch as either pressure or velocity waves. Solid surfaces like metal cymbals and cowbells, wooden sounding boards and marimba keys, even the skin of drum heads and the cones and domes of speakers are much denser than air. They compress the air by acceleration so they are predominantly pressure waves on launch. Aerophones like woodwinds and human voice are predominantly velocity waves at the point of origin, which is a hole that lets the resonant column of air leak energy into the room.

    Either way, the wave equalizes as it spreads until the pressure and velocity vectors are equal in energy. Pressure and velocity transients are diffused by this action, and so reduce their “crest factor” (ratio of peak to average level) within a few meters of the sound source.

    Transient peaks are phase aligned half waves at all frequencies (look up”Dirac Impulse Function” to see the Fourier de-composition). As sound travels through air, the high frequencies are both attenuated and phase shifted with respect to the midrange and bass frequencies. By the time you get 10 to 15 meters from a sound source, peaks have lost as much as 6dB because the frequencies don’t pile up as high when they are at slightly different times and lose highs to the low pass filter of air.

    The long range effect is the natural room reverb. The acoustic compression of reverb is related to the Schroeder limit, the distance where the reverberant sound energy is equal to the direct energy. The monstrous steel framed halls starting around 1890 have a Schroeder limit near the first row! This is where you start to lose articulation and the musical consonants (how the notes start and stop) get muddled. The room buries transient peaks in the echoes of previous sounds, unless there are a few seconds of silence before the peak. Again, the peak to average ratio goes down just like the action of a limiter.

    Typical microphone positions are at the tenth row, which is beyond the Schroeder limit for all symphonic halls, with the possible exception of the few original wood frame “shoebox” halls still standing, like Boston Symphony and die Musikverein in Vienna. This is always beyond the pressure/velocity equalization and far enough for transient smearing by air.

    I often record piano and string quartets from a meter or two way, which is a musician’s perspective. It is more like playing in the ensemble than being in the audience. My HDR has electronic meters fed directly off the A2D, which is running at 2.8MHz. These peak-hold displays are faster than human hearing, and reveal that pianos and string quartets have peaks as loud as a wailing lead guitar!

    This brings us to the ultimate reason why nobody hears real peaks: 1″ dome tweeters can’t reproduce them. I only listen to AMT tweeters with 5x to 40x the surface area of dome tweeters, higher efficiency and lower inductance. The “Le” of the speaker voice coil is the limit to speed, and there are no domes that are fast enough and loud enough to come close to a violin. Note that crossovers also reduce transient peaks by phase shifting the frequency components, so anything sharper than second order Bessel is robbing you of peaks.

    So, music professionals who hear the pluck of picks and the snap of sticks at close range daily know what transients sound like, but anyone who listens to speakers, including the recording, mixing and mastering engineers, does not. What the engineers know is that electronic limiters sound better than amplifier clipping or dome tweeters clipping when they hit their excursion limits – and a lot better than blown tweeters that tried to reproduce fast transients. If the studio monitors use domes, including horns with 1″ dome diaphragms, then they are simulating the lack of peak capability in home speakers.

    So as you might guess, I avoid Fisher/Geffen Hall, the Met Opera and Carnegie/Stern Auditorium. When I am forced to, we get seats in the first few rows. This is also a reason why I stopped listening to commercial recordings. The last of fast peaks is annoying.

    1. I agree, great post. Perhaps my statement that if you only affect the peaks you don’t hear it is too sweeping and simplistic.

      My hope was to explain a limiter’s function and actions, not to excuse their use – use which I don’t condone. If I were making a recording I would not use a limiter or a compressor.

      Perhaps I need to write of that?

      Reference Recordings surely does not use either.

      1. NO, but symphonic recordings are either miked at a distance generating substantial ACOUSTIC compression/limiting; or a forest of spot mics feeding a 48-96 track digital recorder, a compressor per channel and the dreaded digital reverb.

        Latter 19th to 20th Century symphonic and operatic music like Mahler, Verdi, Wagner and Bernstein was written for this acoustic – the problem is applying it as “one size fits all” as if acoustics could be quantified on a scale of 1 to 10. It is ludicrous to have chamber music in symphonic halls.

        Baroque chambers, fin de siecle salons and other personal spaces where the intended acoustics of string quartets, fortepiano and violin sonatas, art songs, etc. These were sumptuously furnished with upholstered seating, heavy drapes over the windows and tapestries on the walls, coffered ceilings, columns and other diffusive architectural details that have been abandoned in contemporary residential architecture.

        I re-created this acoustic in Manhattan, tuned for piano and string quartet, where you can sit on a sofa ten feet from a Steinway Model D for the most immersive experience in modern times. The only compression is that of the wooden sound boards, which have non-linearly increasing spring force with displacement.

    2. Great post indeed ! I printed it and reread it several times to try to understand what you meant. I think I got it.
      This explains my experience with a new local concert hall that visually had all the trappings of an acoustically well designed venue. I sat in the fourth row at my first concert and the sound was illuminating. “Articulation and the musical consonants ” were sharp and satisfying. For the second concert, my group was forced to sit in the last row of the orchestra section. Despite world class artists, I was barely able to sit through the program. The sound was smeared, lacking any definition. My two friends with me that night, who are educated listeners, suffered equally. Our wives couldn’t tell the difference and were annoyed at us for making such a fuss. So be it !
      Thanks for your post Acuvox.

  6. Thanks gcsakakini, for answering my question. Anyone else? The reason I ask, is I think it gets to the issue of the use of compression in recordings. There is a camp for whom it is anathema, simply on principle. But that does not mean that not using it at all automatically gives you a more natural sounding recorded representation of the performance given current technology and on most systems.

    And yes, we know everything that has anything to do with this entire process is flawed. Recording, mixing, mastering, stereo systems, rooms, setup, ears, brains, etc. Any chance we could give all that a rest?

    1. Your amp/speakers and room limit your dynamic range to 30-40dB less than live acoustic music. As I mentioned above, 1″ dome tweeters fall ate least 20dB short of real transient peaks. Most home speakers are 88dB or less efficient, and orchestras peak out around 120dB. If you used a 1,500 Watt amp to get to this level, your speakers would be flaming or at least smoking – and that is assuming your ears are a meter from the speakers. At six feet, make that 3,000 Watts.

      On the other end, concert halls are quieter than living rooms. Residential noise floors give up 10dB or more.

      If you want to relax into the music, with sound that is optimal for the health of your ears and mind, buy some tickets! Get a season subscription! Sign up for house concerts!

      Stop trying to find nirvana in machines.

      p.s. with acoustic music in the room, you can hear much further into the noise floor. Human hearing is much better at separating sounds than microphone pickup.

      1. Avox – I’ve been recording most of my life, so I get all that. No one is “trying to find nirvana in machines”! Well, except when I’m listening to Nirvana. But the dynamic range on their recordings disqualifies them from the discussion. Which is ironic, come to think of it, since a big part of their musical innovation at the time was playing very softly……then REALLY LOUD!!

Leave a Reply

© 2023 PS Audio, Inc.

linkedin facebook pinterest youtube rss twitter instagram facebook-blank rss-blank linkedin-blank pinterest youtube twitter instagram