A different approach to the hiss-in-compressors issue

[Mark Hammer]

After finally getting the manual to a rackmount compressor/limiter I bought a couple of years ago (http://classes.berklee.edu/mpe/pdf_files/manuals_pdf/ashly_sc50.pdf), I was reading it on the bus to work this morning, and something they pointed out made me think.  The manual (well-written, I might add.  Kudos to Ashley.) discusses the perennial issue of noise/hiss in compressors.  One of the suggestions they make is to reduce the amount of compression used.

I thought about it, and I realized that one of the reasons why compressors can be hissy is that they increase gain under low-signal conditions too much.  So if you don't feed them anything, they crank up the input, and if that input source is hissy or the compressor itself is not well-engineered, you get all that hiss multiplied by whatever the gain is.

In the case of limiters, however, the goal is not to achieve some average overall level by boosting the low-level signals and attenuating the high-level signals.  Rather, the lower-level stuff is left pretty much where it is, and it is only the higher-level stuff that is affected.  In that sense, limiters should be quieter than compressors, all other things being equal, because limiters don't have the automatic boosting.

Of course, limiters and compressors are used for different sorts of things, and one of the reasons why limiters are less common stompboxes than compressors are, is because guitar players WANT that level boost so as to produce the illusion of sustain.  That's fine, but the noise problem seems to occur when the boost is situated "up front".  In other words,it is not the fact of compression that creates the noise problem, but rather the concentration of automatic gain boost in one spot.  For truly quiet compression, oriented towards what a guitar player wants, my sense is that one needs a kind of blend of limiting and compression functions in "the front" with gain added afterwards, at the point where noise can be more successfully managed.

So, the amount of automatic gain increase should ideally be lower than one normally finds.  I'm not sure what the default gain is in the Dynacomp/Ross, but if you mute your strings, that hiss comes up pretty noticeably if you don't have a pristine input signal.  By keeping the automatic gain increase to a modest level, and constraining the dynamics at the top end (i.e., a little closer to what is called "hard limiting"), where gain is reduced instead of boosted, one should be able to get a compressed dynamic range without the exaggerated hiss.

Alternatively, reduction of dynamic range doesn't have to all take place in one locus.  Why couldn't it be adjusted in two cascaded stages, so that each gives a little bit of boost to the low-level signals (but not enough to get the hiss), and provides some attenuation to the higher-level signals, with the result being that you get the same overall compression of dynamics and sustained level, without inviting all the hiss you get when you do it all in one spot.

Of course, the complement to this is that more gain has to be applied to the envelope follower (sidechain) in order to provide a suitable control signal.  Indeed, that seems to be one of the downfalls of many stompboxes.  In an effort to achieve a suitable sidechaiin signal within the context of a cheap and simple product, the gain cell has its gain cranked up and the output from THAT is then rectified.  If you transport this to, say, the Orange Squeezer context, instead of the op-amp in the signal path having a preset gain of 23, it would have a gain of, say 2 or 3, and a second op-amp would tap that and provide additional gain for the rectifier stage, instead of letting the audio stage do all the work.

What do you think?

[MartyMart]

Makes perfect sense to me , I'm a "limiter" fan , for sure and use them regularly as recording helpers
and "plugins" at work.
I like the "squashing the peaks" idea much more than "lifting" the troughs
Recently tried a Boss limiter and liked it a lot .
Seems like we need "multiband" compression, so as to leave any "high hiss" alone, just compressing
low and mid frequencies, say up to 4khz  ..?

MM.

[puretube]

think: voiced/unvoiced detectors in vocoders...

[ashcat_lt]

I could have sworn that the only difference between a compressor and a limiter was the ratio and the attack/release characteristics.  Neither "boosts the quiet stuff and lowers the loud".  Both simply attenuate the signal when it goes beyond a certain level (threshold).  You generally then increase the output gain to make up for the loss of overall volume. 

"multi-band" compression, or letting the high freqs bypass the compressor will help a little, but keep in mind that many of the transients from the attack of a guitar string (the real peaks in the level) are in that high freq area.

I would suggest that the best way to reduce the problem of "compressor" hiss is to reduce the overall amount of noise in your guitar signal before it reaches the compressor.

[Mark Hammer]

Your suggestion at the end is, of course, the best one to make, regardless of what one is using after the guitar. 

Compressors and limiters DO work more or less the same way...more or less.  I like what they say in the Ashley manual: the difference is  in how you use them.  "Compressors" are used to maintain a constant overall level.  In doing so, gain is reduced when the input signal exceeds some level, and as the input level drops, the gain comes up again.  The result is a relatively constant amplitude.  "Limiters" are used to maintain a certain maximum level.  So, gain is not really tampered with until a certain threshold is reached, at which point the gain is reduced.  But as the signal drops, that only means that gain reduction is "relaxed", rather than gain itself being increased.  That sounds like splitting hairs, but the upshot is that even though both adjust gain in response to high-amplitude transients, soft sounds stay soft with a limiter, where soft sounds get bigger with a compressor.  Of course, if you set the limiting threshold low enough so that gain is adjusted for almost everything, you can't hear the difference between a limiter and compressor.  In the Anderton EPFM compressor, gain change is achieved via an LDR in the feedback loop of an inverting op-amp.  Big input signal means the LED shines and the LDR resistance drops.  Because it is in the feedback loop of an inverting stage, that means the gain of that stage drops.  As the note dies out and the LED starts to dim, the resistance gets higher again and the gain increases.  The gain never goes past the maximum set by the parallel resistance of the LDR and a second resistor, though.

As was noted on earlier threads, often one of the differences between what gets called a limiter and what gets called a compressor is where the sidechain is tapped from.  If you tap the sidechain from before the gain/level adjusting element, then it ends up only responding to peaks, and tends to ignore drops in level.  If you tap the sidechain from after the control element, you're generally into compression territory.  I made a feeble attempt at a modded OS a while back that would try and do both: http://hammer.ampage.org/files/Tangerine_Peeler.gif  I've never built it so I couldn't say if the difference is audible or commendable.

In the case of something like the Orange Squeezer, the issue would seem to be separating the gain needed for producing attenuation via the FET, and the gain needed for keeping a reasonable output level.  The OS packs them both into one stage.  What it needs, more or less, is an input stage before the FET so that you get a little bit of signal boost to improve the S/N, followed by a wee bit of boost after the FET to compensate for the level change after attenuation, I used only unity-gain buffers in the Tangerine Peeler, and followed the whole thing up with a gain stage to restore level.  Whether this was a good way to do things is anybody's guess at this point.

One of the things that piqued my interest in this topic though was that perennial favourite, the Dynacomp.  What I can't tell from the schematic (http://www.tonepad.com/getFile.asp?id=9), because I'm not that clever when it comes to discrete or OTA circuits, is how the gain is distributed throughout the circuit.  Clearly there is some gain somewhere, because it can be louder than bypass, but where does that "loud" happen?  The input transistor seems to be a very low-gain or even unity-gain buffer.  I can't tell how much signal gain occurs at Q2.  Q3/Q4/Q5 are purely for the rectifier/follower, so how much is packed into the 3080, and is that a wise choice for where to stick the gain?

[ovnifx]

Traditionally, the definition ashcat gave is the "correct" one.  A device that raises the low levels while reducing the highs is called a leveller/levelar/levelor etc.  Compressors only give the illusion of "leveling" the signal because of a "makeup gain" stage at the very end of  the circuit.  The hiss comes from cranking up the gain on the pre- and post-compression signal overall with that makeup gain stage, and it "goes away" when the compressor circuit reduces the gain of the incoming signal.

Devices that don't fit this pattern are certainly around, but they are (in my understanding) the exception.

[R.G.]

The fact is that the noise of any set of gain stages is dominated by the noise of the first stage. And there isn't much you can do about the total noise for a given gain except put a lot of effort into making the first stage quiet. Any partition of a gain cascade into several sections still has the input noise mutliplied by all of the gains following it. And that leads to - yes, make the input as noiseless as possible, and make the input stage as noiseless as possible.

But there is one thing that hasn't been mentioned here. That is, noise is more noticeable in high frequency hiss than at low frequencies. That's what Dolby made his money on. I think we could take a page from Dolby's book and design a compressor that turned the treble content down as the gain went up. There might be a balance there that would preserve the  audible effect of compression without losing too much treble. And with the treble decreasing, so would the induced hiss.

So loud sounds would start sparkly, and get a bit less crisp as they decayed and gain came up. But I suspect that the listener would mentally add the "crisp" that they heard first to the later compressed sounds. At least, it would be hard to pick out the effect happening in the face of the more-obvious compression.

There may have been compressors designed like this already, and it seems that I remember something like this, but I can't bring it out of general storage into clear focus.

[StephenGiles]

Aphex deal with the hiss problem with the combination of its Dynamic Verification Gate and  Dynamic Recovery Computer in the Compellor by freezing the control voltage when the input has been detected to fall below a preset level. Unfortunately the circuitry for this does not appear in the service manual, although the patents provide a clue. But that of course is at line level. At instrument level I suppose that given a schematic is available, it could be adapted.

[tcobretti]

It seems to me like Mark hit the nail on the head when he brought up the word "threshold".  If we (by we, I mean you smart guys) could create a way to graft a threshold control (either fixed resistor or pot) onto a stompbox compressor it would make an enormous difference in the control over the noise level.  It seems like a simple hard knee compressor with an adjustable threshold that will fit in a stompbox would be perfect.  The only two controls could be threshold and output volume.

[RaceDriver205]

Aye, or one could just use an additional noise gate.
I guess what we want is a compressor that does this:

[Ardric]

Seems appropriate to link to AMZ's implementation of the SSM2166P compressor chip: http://www.muzique.com/ssm2166.htm.

This set of graphs from the datasheet is particularly useful:  http://www.muzique.com/ssm2166d.gif

Limiter, compressor, noisegate... they're all like a simple analog computer in the sidechain.  Given these DC inputs (one of which is our rectified signal), perform a little math and produce this DC output, which then controls VCA gain.  Not shown in the above graphs is the time domain, which is probably at least as important, and not as easy to muck with on this chip unfortunately.

There was a thread some time ago about complex envelope detectors for wah filters and etc.  Perhaps that's the same problem from a different direction?

[tcobretti]

I believe that the problem with the SSM2166P is that it isn't in production anymore.  Now it is the 2166S which is SOIC.  I'm still a newb - how do you work with an SOIC chip?  They are freaking tiny.

http://www.analog.com/en/prod/0,,765_1075_SSM2166,00.html

[puretube]

http://www.national.com/pf/LM/LM1894.html

[Mark Hammer]

So you're suggestion that the 1894 (single-ended noise-reduction/filter chip) be used in a manner like RG proposed? I.E., to sync noise/treble and gain adjustment

[stomper]

How about this?  Place a limiter at the beginning of the chain and a booster at the end.  The top drawing shows the concept, but it is impractical to work two pedals while playing.  The second drawing shows the limiter and booster in a single enclosure with an "effects loop" for all your other pedals.  The third drawing shows it implemented on a pedalboard.  "X" marks the compressor pedal.  The other boxes could be whatever you use, fuzz, delay etc.  

[Transmogrifox]

The mention of the SSM2166 chip made me think.  This chip has a "downward expansion" region, which I find to be somewhat hard to deal with, because the guitar signal just cuts out below this region.  If you were to make a compressor with a minimum gain of unity, then the downward expamsion thing would be ok.    We have an expansion region, then, between low-level signals and the threshold so that the signal is near unity when it's quite small (like noise), but just a little guitar signal will make the gain increase quite quickly.  Then the detector on the output senses the signal level is too high, and begins to overtake the expander such that you blend into a compression range.

Really, the expander's control voltage would be setting the compressor's maximum gain, and it would be clipped at a final level for any appreciable signal input, at which point it would operate like a basic compressor.  For small signal levels, the expansion control voltage would drop down, but be limited on the low side such that the amplifier's gain is at unity, so it doesn't drop out altogether.

I think what Mark is getting at is the concept of "milking" the SNR.  Another way to think of this is as an device that automatically optimizes Signal/Noise Ratio (SNR) for certain signal levels.

In any case I don't think anybody will argue with the importance of designing a low-noise front-end amplifier.  If you can boost the signal level with minimal increase in noise, then you have successfully improved the SNR.  It's that simple (tongue in cheek)