Forget the advertisements just this once!

You should normally not see this.

HdK Articles

USING COMPRESSION WISELY AT HOUSE de KRIS

by Kris

10/11/99

INTRO

House de Kris (HdK) strives to provide good equipment for visiting artists to use in the creation of their works to produce a quality product. But any equipment that is not adjusted or used properly will yield inferior and/or disappointing results. While HdK cannot provide the absolute pinnacle in the state-of-the-art equipment, the electronics supplied should be of sufficient quality so as to not be the limiting factor in a quality driven artist.

For the purpose of this article, we shall assume that our quality mix starts with a good DJ who has thoughtfully sequenced a mix, executed the mixing as best as can be expected, and that the CD players and mixer have not degraded the music signal to a point of rendering the mix unacceptable. In other words, we are assuming the signal to the recorder is fantastic and a good starting point for a mix CD. Currently, HdK mixes are done in the analog domain then digitized promptly in the signal path. This file is a picture worth about 1k of words. This picture shows that as soon as the freshly created analog signal (output from CD player) enters the mixer and gets mixed, it is routed out to the recorder either via or bypassing a compressor. It is this compressor that this article deals with. Please note, this article is intended to be a rather general guide, but it will make use of specific examples of HdK equipment to drive the points home. Currently, the House de Kris equipment of interest for this article is a dbx 266A compressor, dbx DDPw/DIO, Tascam DA-30mkII DAT recorder, and Marantz CDR-630 CD-R recorder.

DIGITAL/ANALOG DIFFERENCES

Let's face it, we ARE in the digital age. Digital has brought many benefits when it was applied to audio. It has also caused us to do some past practices differently to accommodate the new style. For example, in analog the maximum signal amplitude spec is usually pretty vague. It really depends on how much distortion we are willing to live with. One person may define it as the 3% THD point and come up with some number representing the maximum signal allowable. Then, someone else decides they can live with a bit more distortion and push it to, perhaps, a level 5dB greater than the first guy. This is fine and is done all the time. It also illustrates that there is no absolute top limit in the analog world. Compare this to digital with a finite number of possible coded levels based on word width. With digital, there is no way to code a signal that is larger than the permissible max word size, no matter how much distortion you are willing to put up with. OK, let's restate that, trying to code a signal larger than possible as defined by word width will create horrendous distortion, but the signal amplitude will still be limited to the maximum amplitude of the digital system.

Since digital does not have a moving target for its maximum signal amplitude and is fully know, all other signals in a digital system are usually measured with respect to this maximum signal level. This maximum signal level is call the 0dBFS. This is to say that the maximum signal is 0dB below Full Scale. Full Scale is as big as she gets. You can't get fuller than full.

The other end of the spectrum, minimum signal, is not quite as cut and dry. A noise floor spec can be derived for analog, but recording below the noise floor is very possible. This has to do with the spectral density of the noise and of the signal itself, and the desire of the brain to "pick out" the signal of interest in the noise. Likewise, a minimum theoretical signal can be determined for digital as well. But, just as in analog, signals smaller than the noise floor can be coded. This is done by using noise shaping techniques.

Other merit indicators could be discussed here, such as frequency response, but they would not be directly related to the topic at hand - dynamics compression. I find it interesting that specifications can be written to favor one method (analog or digital) over the other. This is called specmanship and can be quite misleading to whoever reads the specs. Then again, that is the intention.

DIGITAL'S PROBLEM CHILD

As previously mentioned, driving analog harder and harder usually only results in increased distortion beyond the expected results. The 'expected results' being a larger signal. In digital, overdriving gets the same amplitude signal as non-overdriven (0dBFS), but with TONS more distortion. Digital clipping distortion is an irritating sound that most people wish to avoid at all costs. The nerve grating artifacts generated by digital clipping are enough to render an otherwise beautiful piece of music into something you have no desire to hear a second time. It is because of this severity of the distortion that most recordists strive to stay far away from 0dBFS to ensure never crossing the digital clipping threshold. This then leads to making recordings at perhaps levels lower than what is necessary to prevent digital clipping. The downside of this technique is that the signal is now closer to the system's noise floor. This is perceived as an apparent increase in the noise floor.

Now, contrary to all the I have just written about the horrors of digital clipping, that fact that a digital clip has occurred does not imply it is audible. If a single cycle of a percussive sound has one sample that exceeded the 0dBFS level, odds are very high it will go unnoticed. Even with several cycles on the attack portion of a percussive event with perhaps a couple samples per cycle, this still may not be noticeable in music. But, if a cyclic signal has many milliseconds of clipping with clipping occurring in every cycle (or at least cyclically) then this will create the condition of having people clap their hands over their ears and head for the door. This is not a case of mild overdrive, as in the percussion hit with a clipped sample. No, this is blatant continuous overdriving. Although this may sound as if I am endorsing minor infractions of the digital coding laws, HdK still does not feel 'good' about a recording unless there are NO digital clipping occurrences in a recording.

DIGITAL METERING PART I

It would seem that in order to avoid this digital clipping, the recordists should stay away from 0dBFS. This is true. But, recordists also know that in order to maximize coding resolution, they must stay as close to 0dBFS as possible. Many use recording methodologies for digital that they used for analog. Unfortunately, this doesn't quite cut it. Instead of just carrying over past techniques, some minor tweaking of said techniques are in order. The most common modification to recording techniques comes in where to keep the activity of the recorder's meters. Analog times had most recordist keeping the meters "around" the '0' mark. This could either be the typical average reading VU meter or the less common peak reading dB meter. Either way it didn't matter. The manufacture had set the '0' point arbitrarily to a point which yielded good sounding recordings if the average of meter activity was located "about" at the '0' mark. This gave plenty of room for short overloads from drums or other peak producing sounds.

There was really no standardization of consumer recording equipment as to the actual signal level of a '0' indication. The only standard was the 'double D' symbol on equipment supporting Dolby noise reduction. This was supposed to equate to a recorded signal resulting in a magnetic density of 250nW/m. Most equipment had the Dolby mark at '+3' on the meter. Some manufactures who felt their machines where of better quality than others put the double-Dolby at the '0' mark. Recordings made in the same fashion on this machine would sound about 3dB hotter than on the typical machines. House de Kris has no experience with professional analog recording equipment, otherwise I wouldn't have had to use the 'consumer' disclaimer at the start of this paragraph.

Now comes digital. The norm for manufacturers is to use metering that is indicative of how signals are referred to in the digital domain. In other words, with respect to 0dBFS since it is a well know maximum possible signal. Meters range typically on digital machines from -50 to 0 and -20 to +3 or +6 for analog recorders. For some reason, human nature makes us want to equate the analog '0' to digital '0' when working with digital recorders. This then has the digital recordist setting levels to make a majority of the music hover around '0' leaving absolutely no room for peaks in the music with out digital clipping. But, if the recordist were to equate the digital meter's '0' to analog's '+6' then the majority of the music should hover around '-6' on the digital meter. This would then leave some room for peaks without overload. Digital clipping could still occur, but it will be for (hopefully) only peaks and will be much more tolerable.

The above picture illustrates an example of both world's meters, analog and digital. In this view, it is easy to see how recordists who hover about '0' on analog machines should keep the meter activity at about '-6' on a digital machine.

ENTER THE COMPRESSOR

Although digital has the capability to exceed the dynamic needs of most music and environments, there is still a need to add compression to the signal. This stems from the tendency to digital clip brought on by the urge to maximize resolution by driving the recorder hard. The higher than necessary signal levels encourage clipping, which require compression to prevent. For the recordist that wants to hover around '0' on their digital recorder, they would like a compressor that prevents overload in the last quarter dB or so. Such severe compression, or limiting in this case, would sound almost as bad as letting the recorder enter into digital clipping. A quite workable compromise is to dedicate the last couple to few dBs of usable digital recording space (from -2dBFS to 0dBFS) to compressed signals. Still, the recordist will have to learn to have the music hover around -4dBFS or so, to prevent sending excessive amounts of the signal into the compression zone. For example, if a compressor is set up to provide a compression ratio of 10:1 (a ten dB change on the input results in a single dB change on the output) and its threshold is at -2dBFS, then the system can be driven to effectively +18dBFS before digital clipping occurs. It should be noted that this is severe compression and throwing away 18dB of signal may be noticeable on all music other than kick drums. Below is a diagram of this example.

PROBLEMS OF ANALOG COMPRESSORS WITH DIGITAL RECORDERS

The detector path of the typical compressor has many time constants associated with it that prevent it from effectively preventing some types of peak signals from being acted upon. Most importantly, digital overloads of a single sample are usually missed by analog compressors. One contributor to this is that many compressors (especially those from dbx) use an RMS detector of the input signal. Since digital is sensitive to peak values as opposed to RMS values, RMS detectors are slow to respond to the needs of digital. RMS detectors are chosen in compressors to give the most unobtrusive and acceptable compression. Peak detectors are available, but other things get in the way of them working to the digital needs. This problem is the next stage in compressor control voltage generation which is the envelope generator. To prevent annoying fast changes in gain, compressors usually have a certain amount of attack time associated with them. Again, this is done to make them tolerable to listeners, but it completely misses the boat for digital clipping prevention on peaks. In addition, there is often effective spectral shaping to the detector voltage. This will make the compressor less of an ideal agent for alleviating digital overloads.

The first two complaints may not be fatal in digital recording. Both of these (RMS detection and slower than lightning attacks) contribute to the compressor missing fast transient overloads. Per the above section, "Digital's Problem Child," occasional single cycle overloads are probably not noticeable. But, again, HdK would rather strive for perfection than shooting for mediocrity.

SPECIFIC SETUPS AT HdK

Referring to the mixer picture again, House de Kris tries to have the signal in the analog domain for as short a path as possible. Immediately after the buffer which follows the crossfader, the signal exits the mixer and passes through a "Y" cable with the destinations being the compressor and the unbalanced inputs of the DAT recorder. The balanced outputs of the compressor are tied directly to the balanced inputs of the DAT recorder. This affords an easy way of selecting if the recording will have a compressor in the signal path or not.

The compressor itself is setup to have a fairly high amount of compression, about 8:1, with a relatively high threshold. Also, the attack is quite fast with a pretty quick release. dbx's famous "Over-Easy" threshold knee is bypassed to prevent any processing for signals below the threshold. This configuration of the compressor is essentially a hard limiter much like the pictured example in the "Enter the Compressor" section above.

Using the metering system on the mixer usually has no correlation to reality, and is not recommended for setting levels for the recorder. They are pretty to watch though. The best way to ensure a mix that has minimum effects of the compressor is to monitor the Gain Reduction indicator on the compressor itself. With none of these lights turning on, there is no compression happening to the signal. If the Gain Reduction meter has two or three lights that flash only momentarily, this will probably not be noticeable for most music. Anytime more compression is used, or if lights are staying on for sustained periods, the effects of compression will most likely be audible on the finished product.

If these simple guidelines are followed, the resulting mix session should have very good subjective audio quality. This will typically yield a recording that has about 1dB of margin. Very safe and not sorry. However, some artists may WANT compression to show itself throughout a recording. In this case, drive the compressor hard and watch all them pretty lights illuminate. This will then require the recordist to monitor the DAT's margin indicator to ensure digital clipping still does not happen. Some attenuation of the compressor's output may be required in this case.

Although the previous paragraphs are written with analog compression in mind, the same basic concept holds true when using the more recently acquired dbx DDP. Slightly different practices again will be required, but still, manageable.

DIGITAL METERING PART II

Interestingly enough, all digital meters are not created equal. The same music passing from one piece of equipment to another may appear to have different dynamic characteristics depending upon what meter is being observed. For example, the Tascam DA-30mkII will never show an 'over' with a digital input. This makes sense, since no number of larger magnitude than +/-FS can be represented. But, the Marantz CDR-630 can, and does, indicate an 'over' condition while receiving digital audio. In the case of the Marantz, if several successive occurrences of either + or -FS happen, then an 'over' condition is assumed. Well, this makes sense, too.

Also, the point at which the 'over' light comes on with respect to audible distortion is different on different machines. Using a 1kHz sine wave in each channel, the DDP's clipping indicator lights at just the audible point, while the Tascam's 'over' light comes on just a fraction of a dB before. Likewise, the Marantz's 'over' light comes on almost 2dB after audibility. When having the Marantz meter the output from the DDP, the Marantz 'over' light came on 1.5dB after the DDP's 'clip' light when stepping by 0.5dB. With music, distortion is not usually noticeable with the Tascam in 'over' condition often.

As a comparison of three different metering systems, HdK set up and recorded the data using painstaking manual methods. The results are basically the response characteristics of dbx, Tascam, and Marantz digital meters over a 26dB range with 0.5dB steps. The first column 'Level' is arbitrary while the 'DDP' column is the output meter on the digital stream. The Tascam was next in line and then passed the digital data on to the Marantz. Due to the help of a cat half way through data collecting, the exact levels used the second time through are perhaps not the most well balanced or best choice. But, I didn't want to spend even more time on this project.

COMMERCIAL CLIPPING

Believe it or not, digital clipping is a very common occurrence in today's commercial releases of popular music. I've noticed this to be especially true on 'loud' music. One recent "punk" CD I checked last month had the 'over' lights on the Marantz CDR-630 flash about once a second. This was not a noticeable overload at any time. But then, I wasn't listening to it too closely. Other commercial pop CDs cause the Marantz's 'over' lights to flash, but not nearly as often.