Musikmesse Frankfurt
Horse Bar revisited
Audio Days Paris
Vive la France
AES New York
The return to NYC
NAMM Moscow
Go East!
Audio Days Paris
Encore une fois
Musikmesse Frankfurt
Lucky Number Sevin ;-)
NAMM Anaheim
Visiting Mickey
Musikmesse Frankfurt
Old friends, new friends.
Audio Days Paris
Meet the makers

Fifth anniversary
Time to party!
Musikmesse Frankfurt
Less trouble, more fun.
Unicorn or time travel?
The Joel Hamilton interview
AES New York
Welcome back.

Question upon question : )~
The pan60 interview
Musikmesse Frankfurt
Five already?
Forward or backward 
On compressor topologies
AES San Francisco
Ghandi & the Giants

Musikmesse Frankfurt
Meet the family
High-End Gear Hamburg
Small yet Ooh-La-Laah
AES New York
Jazz, Viruses & Liberty

Musikmesse Frankfurt
One, two, three...
Demotour London
London calling elysia!
AES San Francisco

Direct Metal Mastering
On visit at Stockfisch-Records
Musikmesse Frankfurt
Second sighting, twice as good
AES Vienna
Compressor Boys versus Kaiserschmarrn
Chemical Brothers
The brothers gonna work it out!
AES New York
We love New York!
High-End Gear Cologne
Meeting of the pro audio freaks

How everything started
The alpha compressor and how it entered ...
Musikmesse Frankfurt
Premiere in Frankfurt/Main
AES Paris
Baguette and class-A
Emergency Compressor
First aid for dynamic disasters 
AES San Francisco
elysia tours the United States

Forward or backward

Basic principles and topologies of analog compressors
by Dieter Kahlen & Ruben Tilgner
Studio Magazin (Germany) – Analog Special 2010

Dynamic processors belong to the most important tools for processing audio without any doubt. What makes them unique is the fact that there is no real alternative for what they do. On the face of it this may sound banal, but it is not always the case with other basic signal processing tools. In contrast to a compressor, the effects of an equalizer can also be achieved with alternative means under certain preconditions. Changing the recording environment, the type of microphone used or the position of the mic itself are examples for this. In many cases an equalizer is even obsolete when there is the possibility to play with these parameters during recording.

Dynamics, however, cannot be changed with 'natural' measures like these, therefore there is always demand for adequate technology to take care of this. The most important type of dynamics processor is certainly the compressor. Its inalienability and omnipresence in all kinds of musical formats does not need to be specially emphasized here – without it, large parts of modern sound aesthetics couldn't simply be achieved.

The natural dynamics of the human voice and its 'incompatibility' for most modern types of music caused by its low density is one of the classic applications for a compressor. Together with processing broadcast signals of radio stations, this has been one of the most important incitement for its technological progress. Embedding lead vocals with a dynamic range of 20 dB and more in front of the microphone into a dense pop mix without a compressor is very likely to be a losing game. But these days the compressor has more duties than just correcting signals.

In many cases it is used in much more obvious and audible ways to achieve certain creative effects in order to add a special character to a production. A typical example for this is modifying drums. For example, a  brutal metal drumset based on heavy compression is pretty far away from the sound of natural drums.

Feed Forward and Feedback
The crucial components of an analog compressor are its gain reduction cell and the sidechain which generates the control voltage the gain reduction cell needs to do its job. Regardless of the specific gain reduction element, there are two basic types of circuit – the feed forward and the feedback topology. Most of the older compressors are based on feedback designs which tap their control signal behind the gain reduction cell and thus from the already processed signal. The advantage here is a non-linear characteristic curve of the control element (e.g. a FET) and component tolerances are stabilized by this kind of circuit.

The control voltage responsible for processing the gain already contains possible non-linearities and compensates them. The price of this, however, is a reduced range of possible compression settings in many cases. Especially FETs can have high tolerances from one piece to the next, which are usually kept under control by the feedback arrangement. By the way, this is also a reason why classic FET compressors are rarely available as stereo units: a synchronous type of control which is an absolute must for stereo operation is complicated to create with FETs and needs special effort like temperate-coupled transistor pairs in a common housing.

Feed forward compressors generate their control voltage before the signal has passed the gain cell, and they have notedly higher requirements regarding the linearity of their gain reduction cell. A FET compressor with a feed forward topology would be quite a challenge, because it would require an exact individual matching of the specific characteristic curve of each single FET component used. For this reason almost all feed forward controlled compressors are based on a VCA. The benefits of well designed feed forward compressors are the availability of very short attack times as well as broad ranges for the threshold and ratio parameters.

During the last few decades we have seen quite a lot of different compressor topologies with their gain reduction cells as the decisive difference. Of course it is important to have a look at when the specific models have been designed because of the components that were available at the time – a modern VCA like it is built today could not even have been thought about in the 50ies. In this chapter we want to have a closer look at the most important topologies while trying to stick to the historic order of their development. In the beginning there was the vacuum tube, of course, with the variable-mu circuit design as the ancestor of all compressors.

These designs are based on certain types of tubes where the amplification can be changed with the grid voltage easily. In principle, the amplification of any tube type can be varied like this to a certain degree, but usually the linear range  would just not be large enough to make sense for a compressor design. Some of the tubes which have been used for circuits like this have become pretty rare and hard to get these days, for example the 6386 type which is used in the Fairchild 670. Simple Vari-Mu compressors have been offered by smaller US manufacturers already in the 50ies and they have mainly been used to make the sound of local radio broadcasting stations more attractive.

The famous Fairchild 660/670, which is a well known representative of this topology, has been developed a little bit later, and in opposite to the early Vari-Mu compressors it uses an already quite complex sidechain architecture. Compressors of this kind are almost always fed back and stand out – when they are designed well – because of an especially fat and homogenous sound character which is loved by many users. The domain of this type of compressor are smooth and unobtrusive corrections of the program dynamics without drastic interventions, but with very nice sonic qualities which can complete a production very well.

The opto compressor has its name from using one or more photoconductive cells as its control element. These components reduce their resistance at rising exposure to light, and in compressors they are used as a kind of voltage divider which is responsible to control the gain according to the structure of the input signal. Originally the photoconductive cell was triggered by a light bulb which in later designs has often been replaced with LEDs or EL elements that emit their light according to the control voltage generated by the sidechain.

As a general rule, the design of an opto compressor is quite simple and its time response is quite slow – especially when the luminous source is a light bulb, which has an RMS-like characteristic as an enjoyable side effect. On top of this, the photoconductive cell causes some kind of 'hold' effect, because the change in resistance does not happen straight away. The special time response of the opto compressor incapacitates it for certain tasks, but on the other hand it can sound very musical. The release progress of this gain reduction cell is coupled to the program material, as the release time decreases when the amount of gain reduction rises.

This effect and the especially mellow soft knee characteristic as well as the strong non-linearity of the photoconductive cell have been deliberately used to achieve certain sound characteristics – some kind of casual 'accident' which could hardly have been achieved in any other way at that time. Today, similar characteristic curves can be implemented in other compressor designs, too, but this requires quite some effort. The sound of a typical opto compressor is fat, unobtrusive and has low distortion.

However, the type of control offers only few variation possibilities. On the other hand, an opto compressor behaves extremely smoothly; it adapts well to the characteristic of the input signal and can also be used for more complex mixes (if one can live with the limited access to the material). There is not much the user can set wrong here. A typical classic of this genre is the Teletronix/Universal Audio LA-2A.

FET compressors take advantage on the circumstance that a field-effect transistor (FET) can be used as a controllable resistor in a certain area of its characteristic curve. However, this only works in a quite small voltage range and it requires comparatively high amounts of makeup gain as the field-effect transistor can process only low signal levels at low distortion. FET compressors are very flexible in regard to their time constants.

They allow very short attack settings, while the ratio is not so flexible – in many cases only a few switchable values are offered. As mentioned earlier, because of the weak linearity of a FET control element it is almost always embedded in a feedback design. Maximum neutrality is not one of its strong points, therefore FET compressors are often used in applications which allow some abrasive behavior.

Because of the limited dynamic range and the troubles in creating a useable link mode, this type of compressor is not really the first recommendation when it comes to discerning mastering applications. A typical domain of the FET compressor is the spectacular processing of drums. The most popular archetype of this topology is certainly the UREI 1176LN.

The most modern and flexible control element for compressors these days is doubtlessly the voltage controlled amplifier (VCA), and it is also the most commonly used compressor topology these days. Not before the VCA had been introduced the feed forward variant became a real option, because the characteristic curve of this control element is linear and distinct and therefore reproductible as well. Good VCA compressors can be much more flexible and universal in use than the other topologies explained above, but they require more effort in terms of their development, too.

Besides coping with classic compression tasks, the VCA also allows more extreme effects like overcompression and negative ratios. But if needed, VCA compressors deliver a neutral and clean type of control the older topologies can hardly offer. However, in practical use there are huge differences regarding the quality of the specific VCA components as well in terms of the circuitry they are embedded into. This seems to be the reason why VCAs do not have a good reputation for some puristic users – which I cannot agree with.

The construction of a good VCA compressor is a demanding task, and not every developer will have the ambition and the leisure to really max this technology out. If a cheap VCA is used as the basis of a circuit, the result just cannot please the demanding ear. In my opinion, the argument that the VCA has sonic drawbacks by design is completely wrong. Here at elysia we build VCAs from scratch from discrete components of our choice in order to have the greatest flexibility in terms of transistors used and their technical parameters.

All in all, VCAs are the most flexible control elements for compressors, because almost any thinkable characteristic can be implemented with them. The individual sonic results depend on the design of the sidechain and the choice of the time parameters. Soft or hard knee, subtle or effect-oriented compression – the VCA makes it all possible. On the other hand, the quality range of VCA compressors in the market goes from hardly spectacular units for newbies up to high end products well suited for mastering applications.

A rare animal is the pulse width modulation (PWM) compressor. It is based on a complex and hard to handle principle which generates a digital control signal with a variable impulse width by modulating the pulse width. The classic of this technology is the EMT 156 which had originally been designed for broadcasting use. Today, Crane Song is an example of a manufacturer using this technology: their compressor topology is a combination of a FET as gain reduction cell controlled by a PWM signal.

The control characteristic of a compressor does not exclusively depend on its gain cell, but also on the way its control voltage is generated. Usually the decision is between peak and RMS detectors, but sometimes there are hybrid forms of these two types available, too. A peak detector operates a simple detection and reflects every signal voltage peak in the control voltage it produces. The RMS detector, on the other hand, generates an average value with a certain integration time which can be different for each circuit.

Peak detectors are mainly suited for true signal limitations with theoretically infinite compression ratios (limiter), and thus for applications where the level needs to be limited to a certain maximum level (often out of technical considerations). An RMS detector would not work well in this application because of its setting time as an inherent to their functional principle. Very simple peak detectors only analyse a half-wave of the input signal while the other half is dropped during the detection process.

Signals with asymmetric wave forms like voices or brass instruments can cause unpredictable reactions of the detector which might become very different when switching the polarity. Highly precise modern RMS detector circuits in combination with modern gain cells are predestined for compressors, because in my opinion they match the nature of the human ear nicely. By taking logarithm they produce a linear relation between the input level and the control voltage. The result is that a certain difference of level of the input signal in dB relates to a concrete difference in voltage and not a multiplying factor.

When the control voltage for a gain cell is generated this way, the result can be predicted precisely without depending on the absolute value of the input level – I feel this is a more unobtrusive kind of control behavior. If you ask me, the saying that RMS compressors are usually more of the 'slow kind', lacking the possibility to deliver fast attack times, is nothing more than wide-spread but unfounded legend. A modern RMS compressor with high grade components in its detector path can be really fast. When more noticeable compression effects are wanted, a peak detector can generate some interesting results, though.

The only thing one has to keep in mind is that because of these aforementioned reasons it can happen for example that single beats of a snare that sound exactly the same are processed with differing processing results because the wave form of the snare is asymmetrical. Some compressors offer the user the choice between peak and RMS detection; sometimes there is even the possibility to mix the characteristics in steps or continuously.

Signal Path
Besides the gain cell and the detector, the audio path, the design of the power supply and the selection of components have an influence on the signal quality of a compressor which should not to be underestimated. The developer has quite some options to choose from – the range goes from easy designs with ready-made integrated circuits to complex discrete circuits with especially high grade components. And here the situation is pretty similar to every other technical domain – it's quite easy to achieve a good average quality with today's possibilities, but the last few percent points of signal quality needed to arouse true enthusiasm from demanding users require a disproportionate amount of effort regarding components and design.

And this certainly does not mean collecting the most expensive components in one unit and then shaking the whole thing once or twice in order to get there. The art of developing audio circuitry comes down to investing lots of time and effort on testing and comparing several circuit and component options. You can give the same recipe to ten different cooks, and you will receive more or less different meals from all of them. Sometimes it is sufficient to just set some makeup gain without generating any gain reduction in order to test the quality of a signal path. This can reveal interesting details concerning the clarity and transparency of a unit already.

Even different types of potentiometers can have a grave impact on the signal quality, and the same is true of resistors or electrolytic capacitors in the signal path, too. A lot of the phenomena we have experienced cannot be described with technical measurements, however. And good measurement values are not necessarily a natural basis for pleasing sound experiences in demanding applications. If you ask me, certain effects which are definitively audible cannot be described with the measurement criteria we are using today. This becomes most obvious when looking at sound converters like speakers and microphones, but it also shows with electronic circuits.

These observations become especially relevant for mastering; the signal chain is kept as short as possible in this application and the effects of cable-length are much more critical here than in a live recording of 120 channels where lots of other factors are relevant for the overall quality of the recording. When you multiply the compressor topologies described above with the detector variants and the different amplifier circuits based on tubes, transistors or ICs, you can count a relatively large number of different ways to design an analog compressor.

The sonic results of these many variants are different from each other as well – every variant has its specific sonic attributes and control characteristics which can be positive or negative for the specific application they are to be used for. It seems it's the reason why lots of studios have quite a collection of different compressors, so that the individual character of each machine can be used for the application where it works best. Certain EQ settings could be used to approximately 'emulate' the character of a specific machine, but the dynamic behavior of a specific control element is really hard to copy with the 'wrong' unit.

Not so long ago, elysia has started to offer digital emulations of certain hardware units as software plugins, too. What we have learned is that the complete control characteristics of a hardware prototype can be reproduced by a good emulation very well. Generating a virtual control voltage, attack and release times, the characteristic curve of a VCA and similar parameters – all this can be emulated with high quality algorithms in detail; including certain non-linearities regarding saturation or frequency response.

It's not getting critical until the point where an ambitious analog developer would start playing with various components in order to shift the sound into a certain direction by nuances. Maybe the resolution of todays digital formats is not high enough to reproduce these details in the digital realm. Or maybe it has got something to do with the fact that many users still achieve different and faster results with an analog machine than with a digital emulation. To my mind and in face of all precision of the recreation, the feedback you get from an analog device differs from what you get from a plugin, and from all we have experienced this does have an influence on the result.

But one thing is for sure: both domains have a right to exist these days – you just have to know when and where to use which tool to receive the greatest benefit. If a complete production is made inside the box, inserting a piece of analog hardware can be quite laborious and a good emulation can be the better solution in terms of workflow, even it it only reaches only 90 or 95 percent of the performance of the hardware counterpart. And we have not even looked at the cost factor yet.

Of course, the quality of the converters is one of the crucial factors in this regard. If the conversion is not really good, the best analog tools will not be able to deliver their full potential and a plugin might even sound better then. Of course we are talking about something different when the signal path is partly analog anyway like in tracking or mastering...

Because of the mentioned variety in the realm of analog compressors it is not an easy task for a user to identify which units are the best ones to achieve certain results in certain situations. Learning by trying seems to be the best way to get there – although most users won't have the possibility to try every unit they are interested in. What the manufacturers could do to improve the situation is to communicate more clearly what the specific benefits of their products are. This could be of much more interest for the user than an excessive description of the specific technical design philosophy.

Especially in the segment of analog compressors there is a highly visible tendency of reviving tried and tested topologies from older decades with replicas. While it would certainly be nice to own a Fairchild 670 – from my point of view it would be much more interesting to accelerate the existing creative and innovative tendencies we see in analog and digital audio technology instead of cloning existing classics time and again. Of course it is harder to create ideas of ones own and then put them into practice.

I feel the trend is a bummer because today the designers have more possibilities than ever before to break new grounds. Probably the market will regulate this without caring about such appeals completely by itself sooner or later. A large part of today's users – if they are interested and willing to complement their DAW with analog outboard – probably need universal and modern units of a high quality which are much more than a one trick pony.

Ruben Tilgner (37) is a trained radio and TV technician. After completing his technical diploma he worked for the manufacturer SPL for 10 years. During this time he developed the 'Transient Designer' among many other products. In 2005 he and Dominik Klaßen founded elysia to develop and produce discerning analog studio technology and digital emulations.