-DeeT's "What Compressor" Page

original page: Jan 28, 2000
last update: August 19, 2008
(added photos, provided sharper schematic images)

"What Compressor"?

On this web page you will find the complete story of the "What Compressor": initial experiments, later refinements, design trade-offs, and the all-important schematic, parts list and construction tips.

The Wife of Invention

For those who haven't seen the 3630, it has a whole lotta knobs. You can have any compression ratio, any attack and release times, soft knee or hard knee, rms or peak... and I think I've left a few out! Anyway we hooked up the compressor and started experimenting with it. Unfortunately, we just couldn't get it to sound good with her voice. I had previously had a similar experience trying to get a good sound playing my bass through it. The best sound was always in the "bypass" position :(

I assumed, as I had done with my earlier failed bass experiment, that I simply didn't know how to use compressors. After all, the Alesis does everything a compressor could possibly do. Sure, it's a low cost unit, but I assumed the difference between this and a fancier unit would be balanced ins and outs and lower noise. Feature-wise, the Alesis should stand up against anything. I mean look at all those knobs!!

The Inspiration

The difference was striking and immediately apparent. The Joe Meek has four knobs, and no matter what you do, you get a sound that is useful for something. Even with large amounts of compression, the right setting of the knobs provides a transparency I wouldn't have believed possible. The Joe Meek can squish the heck out of a wave and you'd hardly know it.

Lucky for anyone who is enjoying this story, I'm not rich. I would have just bought a Joe Meek compressor and that would be that. [Uh, I don't mean to say that Joe Meek compressors are exceptionally expensive. They're really not.] Not wanting to shell out for a compressor and being utterly fascinated with the sound, I determined then and there to learn what was so special about the Joe Meek and attempt to imitate it. Little did I know I was embarking on a Journey that would take months and cost a good deal more than a Joe Meek compressor. Was it worth it? Absolutely!

The Joe Meek Difference

Danger: Curves Ahead

Here is a graph showing a typical compression curve, realizable with a VCA. The horizontal axis represents input signal level and the vertical axis represents output level. As you can see, the output matches the input exactly until a certain threshold is reached. Above the threshold, increases in input level still cause increases in output level, but a large change in the input causes only a small change in the output. This is the very definition of compression and a machine that can realize a curve this straight is an engineer's wet dream. Nowadays we have no shortage of happy engineers, and the Alesis compressor is more than capable of such performance.

By contrast, here is a graph showing the typical compression curve realizable using a photoresistor. The output level tracks with the input at first, but somewhere near some ill-defined "threshold", the output gain tapers off. For a while there, the curve actually looks like the perfect compression curve, but then the compression softens up and eventually poops out. For very strong input signals, the output again changes the same amount as the input, just with the volume turned down a bit. When one is shooting for the mathematical perfection of the earlier graph, this one doesn't make engineers terribly proud. But it worked in a pinch in the old vacuum tube days.

But let's put numbers aside for a moment. Do our ears do mathematics?? Look at both curves again and ask yourself which one would probably sound better to your ears. It's only natural that smoother transitions will be less intrusive than sharp ones. If your motivations for compressing are mathematical in nature (say, if you're implementing a compander-based noise reduction system or providing precise limiting to protect against overloading an amplifier), then you're far better off with what's behind curtain #1. But if you just want to listen to a compressed sound and enjoy it, don't make math, make music.

The Envelope, Please

Here is an example of an envelope generator's behavior. The top line shows a brief pulse in the input signal. This could be caused by someone yelling "boo!" into a microphone, then remaining quiet for a few seconds. The bottom line shows the resulting envelope. It rises with the input signal, but not immediately. While the input stays loud, the envelope signal remains high. Then comes the release time. The input signal is gone, but it takes quite a while for the envelope to recover. This has the beneficial effect that when the average input signal level remains the same the volume isn't constantly fluctuating up and down. The compressor thus "rides the peaks" in the input signal, without "riding the wave" and bouncing around a lot. (Electronics enthusiasts probably already recognize that an integrator is at work here. Good catch!)

The Joe Meek compressor is special in that it has two (countem, two!) different release times. Brief impulses in the input cause a quick release, but the loss of a sustained input level causes a slower release. To understand the need for dual release times, let's think for a moment about what our ears do every day.

Let's say we are sitting quietly, just listening to the gentle whirring of our hard disk and computer fan. Ahhh, isn't that nice? Now let's say you slam the space bar unusually hard: bang! Did you notice what just happened? Your ears had to deal with a louder sound, so they sort of "turned down the volume" inside your head. After the brief noise was gone, you could hear your hard disk whirring again, but it took your ears a small amount of time to readjust to the quieter environment. Try it!

Now let's say you're listening to that hard disk again, when your modem suddenly answers. "Beeeep...phwchchchchchchchchk!" Doggone it! I thought you'd shut that thing off. Anyway, your ears had no trouble getting comfortable with the louder sound coming from the modem, and after the modem quits, they adjust back. But did you notice? Because the modem sound was around longer, it took your ears just a little longer to recover from that than from the brief sound earlier.

So what does all this mean? It means that inside your brain is a Joe Meek compressor! Well, not really... but your ears really do "turn the volume up and down" in response to the sounds around you, and they do take different lengths of time to recover from loud sounds, depending on the duration of the louder sound. Now here's the key: if the compressor does to the sound what your brain would have done anyway, you tend not to notice it. That's the magic behind Joe Meek's marvelous transparency: it's distorting the heck out of the wave, in just the same way your brain does all the time. So, it doesn't sound mangled, even though it most certainly is. Cool trick, eh?

This is what the dual release times look like. On the top is the input signal level. A brief burst is followed by a brief silence, then comes a longer loud sound, and silence again. On the bottom is the envelope, employing Joe Meek's (and your ears') dual release times. The brief impulse causes the attack, followed by a decay that is slow, but not terribly slow. The longer sound induces the exact same rate of attack, but the decay is much more gradual.

The "Joe Cheep" Design

Hack alert! Hack alert! The files listed in the next paragraph are not the What Compressor but an earlier effort. If you're looking for the schematics for the What Compressor, please proceed to the bottom of this page. To build your very own "Joe Cheep" compressor, here are the schematic, circuit description, and parts list and construction hints.

Beyond Joe Cheep

She Sells Sea Shells

The traditional solution is something called a de-esser. The most common type of de-esser is little more than a special compressor that is particularly sensitive to sibilance frequencies. When the "S" sound comes along, the de-esser turns down the volume, and the sibilance problem is solved.

I didn't find this solution very satisfying and found another way to look at this issue.

The Highs Have It

The answer is that our ears are not equally sensitive to all frequencies. It takes a powerful bass signal in order for us to consider it "loud", whereas a comparatively small amount of power at a high frequency can send us running for the exits. So, a mix that sounds even to our ears must have lots of bass frequencies and fewer high frequencies.

If a compressor could be made to agree with our perception of volume, to regard levels as equal that we hear as equal, that might solve a lot of problems. For one thing, there would be no need for a "de-esser", because the "S" sounds just as loud to us as the rest of the word. The compressor would simply "do it right the first time" and not turn up the volume on the "S" sounds. Another benefit is that the tendency of compressors to "pump with the bass" would be mitigated, since the compressor wouldn't consider the bass to be any louder than we think it is.

Okay, so I knew what I had to do, but what is the curve? What is happening that causes our ears to be more sensitive to high frequencies? I'm no ear expert, but here's the line of reasoning that led me to the curve I'm using. I assume, first, that the apparent loudness of a sound is proportional to the amount of energy transfered to the ear drum. Physics says that an ear drum wiggling 1mm at 100Hz has half the energy as it does wiggling the same distance at 200 Hz. If my initial assumption is correct and I can ignore other factors, that means the ear's sensitivity to low frequencies drops off at 3dB per octave.

With this goal in mind, I developed a filter to tailor the compressor's response to my idea of the ear's characteristics. The addition of this filter rocketed the sound from "Joe Cheep" to "Whoa, Deep!", a delicious, sensual experience. We started recording with this new sound, and we were very happy with it. But there was one improvement still to come.

The "April Resistor"

I decided that a "mix" knob on a compressor could be a hard sell and set out to find a "magic ratio" of compressed to dry that sounds transparent, yet still permits a decent amount of compression. I settled on a maximum compression amount of 14 dB, which isn't super high, but it's higher than is usually practical for most vocal and instrument tracks, and the improvement in the sound was undeniable.

Can I Hear It?

Can I See It?

• Front View
• Rear View
• Inside View
• Front view of 8 channel "pro" version

How About Specs?

Compression modes: Stereo or Dual mono, true bypass
Dynamic Range: >120dB
THD @ 1KHz +4dBu: <0.05% for all knob settings
Maximum Gain Reduction: 14dB
Stereo Tracking Error: <1dB
Compression Ratio: Variable, Max. Approx. 1.5:1

How Can I Get Me One Of Those?

• Circuit Description, Parts List and Construction Hints
• Schematic, Sheet 1
• Schematic, Sheet 2
• Schematic, Sheet 3

I've also built ten prototypes so others could help me evaluate the design, but I probably don't need to be buried with all ten when I die. They're somewhat ugly (diecast box) but super clean electrically and mechanically, with all surface mount construction. Some of these prototypes are spoken for, but others are not. If you think you might like to buy a "What Compressor" prototype (subject to availability of course), send me an email.

Go to -DeeT's Hacks Page.