Multi-band compressors: Let's split!

[Mark Hammer]

The issue of treble loss, and retaining treble, in compressors regularly comes up.  Thinking about it today, I got interested in the idea of multi-band compression again.

Certainly one of the perennial issues in stompbox compressors is that they treat the signal in monolithic fashion: one envelope/rectifier signal and one across-the-board degree of compression.  The trouble is that most of the signal lives in the low end, so when a wide swing of the string over the polepiece tells the rectifier "Holy transient, Batman! We better reduce the gain!", the amount of gain reduction imposed on the harmonics is really an amount more appropriate for the fundamental.  (I am reminded of the Talmudic edict about not yoking an ox and an ass together, because the stronger one will force the weaker one to work harder than it can.)  The Marshall ED-1 compressor attempts to address this by inserting a sort of BMP-style tone control before the rectifier so that the envelope (and compression "feel") can be influenced by the lows and highs differentially.  But the unit still only uses one single sidechain for the whole signal, and applies whatever gain reduction is demanded by that envelope to the entire signal.

More appropriate, it would seem, would be separate gain reduction for different bands, such that the low end gets rectified separately from the high end, and different amounts of gain reduction are imposed on each portion.  Incidentally, it need not be a 2-way split.  It can be 3-way or more, if you have the parts and patience.  The signal portions with the widest amplitude changes get reigned in harder, and faster, while the portions that provide the definition to the note are allowed to retain a little more of their original character.

The concommitant questions, though, are these:
1) Where are appropriate points for dividing the frequency spectrum (we'll assume this is to be optimized for guitar, and not a "generic" design)?
2) Should there be differential gain reduction in each of the 2 or more bands that can be explicitly controlled?  Or is separate rectification going to take care of all of that for us and all we need to do is set the master gain of whatever feeds the multiple rectifiers?
3) Is optimal tone achieved by have different time constants for the various bands?  For instance, should gain reduction come on slower for the high end so that we don't lose the pick attack?

I'm having a hard time wrapping my head around this.  There are probably some theoretical resources about this (and no doubt Ton will point me to a patent document from 1953 or something! ), but I'm just starting out on this thought problem.  Anyone have any thoughts or experiences, or references, in this area.  Let me emphasize, I am not looking for high fidelity, or quietness, but rather somethiing that cuts nicely, and lets a SC pickup sound like one all the time.

[R.G.]

Hah!  A philosophical discussion!

1) Where are appropriate points for dividing the frequency spectrum (we'll assume this is to be optimized for guitar, and not a "generic" design)?

Hmmm... one could take different tacks. Such as (a) guitar spectrum, based on fundamentals and harmonics (b) limitations of the compressor, i.e. maintaining a smooth envelope (c) sound power in "typical" guitar signals per frequency (d) psychoacoustics. Without a better grounding in acoustics than I have, I'd just shoot from the hip and make separate bands out of the lowest four octaves in one or two octave bands, possibly making the highest one be "everything above X". I suspect that a little listening would make the inevitable mistakes apparent.

First shot is for ranging and second shot is for effect.

2) Should there be differential gain reduction in each of the 2 or more bands that can be explicitly controlled?  Or is separate rectification going to take care of all of that for us and all we need to do is set the master gain of whatever feeds the multiple rectifiers?

Here's an interesting question - if you compress the lowest two octaves, do you need to compress the stuff above that?

If so, I'd just compress equally on all bands.

3) Is optimal tone achieved by have different time constants for the various bands?  For instance, should gain reduction come on slower for the high end so that we don't lose the pick attack?

I think yes. However, this needs done subtly to avoid sounding unnatural. As if that's possible with today's hyperprocessed guitar sounds.

By the way, The SA571 is still in production and gives you two full compression channels per chip. A Linkwitz-Riley crossover can be done with about three opamps per band, so if you used duals, you could make a two band fragment with seven chips, six of them being dual opamps.

[R.G.]

Here's another thought.

Constant Q graphic EQs have something like one-octave bandpasses, ten or twelve of them. Using just the filter array and slaving that to a number of SA571s should turn up some interesting things to play with.

[Khas Evets]

Instead of multiband compression (which would be a complex build), what about simply eqing the signal sent to the detection block. This isn't exactly what you're looking for, but it would add flexibility. For a simple example, if you switch in a bass cutting cap, you would be emphasizing the compression in the upper register. I suppose I could experiment with a rack compressor with a side chain, sending an equalized signal to the side chain.

I don't mean to side track the topic.

[sfr]

Seems like the easy/cheap attempt at a start towards this would be to grab a couple compressors (I'm sure many of us have built both a Squeezer and a Dyna/Ross Comp, and isn't there a Dano compressor from the really cheap mini line?) and throw together something like the front end of Anderton's quadrafuzz, send the frequency split outs from that into the compressors and then mix 'em back together and see what you get.   

If I thought A) I could find the boards for the OS and Ross Comp I have populated and hanging around somewhere under the bench and B) had the time right now, I'd give it a go.

Actually - seems like something you could do decently with a few plugins and ProTools or something - I know I've done a similar concept with different effects.   Although compressing the signal before the amp and compressing the signal after the microphone are two different things, so this solution may not apply.

[RaceDriver205]

I do believe that my Boss Micro BR does 3-band compression. I might be mistaken though, I haven't got my head round all its features yet.

[Ben N]

I am reminded of the Talmudic edict about not yoking an ox and an ass together, because the stronger one will force the weaker one to work harder than it can.

OT: Not Talmudic, Biblical: Deuteronomy 22:10.

Great post, otherwise.
Ben

[dxm1]

Hmmm...

Talmud <==> Old Testament

This could get ugly...

[markusw]

The Trace Elliot Dual band comp splits the signal into two freq bands (no variable crossover though). Here is the schem http://britishaudioservice.com/schem/Pedals.pdf and the manual http://www.britishaudioservice.com/inst/PEDALS.PDF for compression details...

Might be interesting ... 

Regards,

Markus

[Processaurus]

Good ideas, I feel compelled to try this as a mod with my Ross/dynacomp that has a dry blend, in making a shallow HP filter on the dry and muddy up the compressed with low pass filtering.  Someone with a stock ross/dyna could try it easy by using the emitter of Q1 as a dry buffered signal and send that and the comp output to feed the respective two sides of a Big muff tonestack with equal valued caps and resistors on each leg.  Dry to the HP, comp to the LP.  You'd need some makeup gain afterward though. 



This one went over 99% of guitar players' heads, I don't think they make it or the distortion with pre and post grahic EQ anymore.

[Ben N]

Hmmm...

Talmud <==> Old Testament

This could get ugly...

Nah.. if ever I could say, it's all good...

[petemoore]

  To even out the response of the detection, low freq to high...
 Pick a bandpass freq band as 'sample' for the low freqs to be attenuated and sent into the detector.
 Pick a band pass for the high freq content, mix the low and high 'accordingly so that the detection of signal tends to be more equal from high to low freq inputs..to compensate for the guitar putting out higher voltage related to low notes. That's one ['easy I guess] way to even out bass/treble attenuation amounts, not that adjusting the pickup 'low on low/high on high' doesn't do relatively the same thing...or that having highs not attenuated isn't usable...like 'standard' compressors tend toward.
 The other thing..
 Dual freq comp...
 Take a bandpass signal to 'sample' low end, probably about 1/3 of the guitar output frequency spectrum [because bass is 'big' looking, and comps just 'seem to respond differently to about the bottom 1/3 of guitar freq's] and run that to the bass side env. detector.
 Take a bandpass 'sampler' to be balanced/mixed with the bass sample to envelope detector.
 I would guess the divide would be 'first targetted' around 5th fret/treble side, where the filters begin/end' rolling off frequencies would overlap, so putting in a bandpass 'space' between them...which would be filled as they are summed...adjustable for certain, especially on the 'first draft'.
 Anyway the basic point being that the env. decection filtering for bass side/treble side should probably be accentuating/attenuating different frequencies than the signal path bass/treble sides, figureing those out...what freqs makes the bass env 'dip' [and by how much] and what makes the treble 'dip' might be best figured out 'not by ear', using a meter to see 'what moves' the Env. detector and how much, get target values..then set by ear.
 I'm not so certain 'we' don't want treble swell effect w/bass compression.
 "Natural  w/Single coil pickups"...the other side of the coin would be does whacky things to the treble/bass side for lots of neat-o effect possibilities, sustained chords with movement of treble to bass, or bass to treble, slight to heavy 'breathing'..might be real cool, possibly approaching light env filter sound.
 For something like this, I think any self-crotiqueing-tweeker will want a very flexible first draft, and/or a Ver II. The problem certainly isn't lack of possibilities.
 LP filter on the bass env filter seems unneccessary, except maybe to tame or force a certain response from the env detector, highs won't twitch it, even if they do, bass probably isn't coming through the SP when highs are played, I guess a decent bass side comp could be allpass env detection w/SP being LP filtered?
 Here's where I'd try switched values to help prevent 'too many options to see straight' [easier to getting back to 'x' setting than using a pot] for caps or other:
 what freq the HP filter passes to the env. det. [probably two selectable cap values]
 maybe use a notch filter on the bass side env. to get more dynamic effect from the low E string.
 Then use pots for amount to env filters and comp SP's, should be enough for some confusion right there with all these sw's and pots, subtle effects.
 I think I'd try 'seeing' on [wish I had a scope] a DMM what the env detection circuits respond to freq/amplitude wise, hopefully drawing on some compression experience, perhaps looking at what your fave comp env detector actually 'does' when it's in different 'positions' of a note attack/sustain envelope...but remembering there is a sum/mixing of frequencies/amplitudes that happens at the end of each side of a 'split freq comp' [or try to while doing everything else].
 Something you could fiddle with for 5 long, yea, before you figure out what it can do or is going to be doing.

[petemoore]

This one went over 99% of guitar players' heads
  Too bad they didn't put 'eq' on the top half, 'mixer' is a confusing title for those sliders, unless it has a patch in/ for each slider, some assigned freq' #'s or something..unless it looks like what it has to be...looking at it, I'm confused.
  It looks like an eq with freq/comp sliders...slick, what..1 compressor with eq, but a similar multiband eq for the compression env. ?
  something like that...without split freq env detection.
  3 fairly narrow freq bandpass filters, low, med, high,
  tuned to respond well across the typical guitar freq spectrum.
  the top two having 1 knob like balance, the bass filter separately adjustable, interactive though, and used to be mix [what should be 3 evenly balance voltage amounts] and fed into the env. detector., add low noise eq.
 

[petemoore]

  But I like the:
  Env detector and comp for highs, each separately tunable/voiced
  ED and comp for Bass each tunable separately/voiced
       
>mixer for the hi/low signal paths.
     
 

[sfr]

Hmmm...

Talmud <==> Old Testament

This could get ugly...

I thought it was Torah = Old Testament, (at least the first five books of it - I'm not sure how the other books fit into things in Judaism) and that the Talmud was something of a collection of Rabbinic teachings.

Sorry, getting OT.

[markm]

Although alot of this discussion goes well over my head engineering-wise, I am very intrigued by the idea and if anything ever comes out of it, I'd be one of the first to build it and give it a whirl.
I'm a Comp-Junkie ya know! 

[Mark Hammer]

Hmmm...
Talmud <==> Old Testament
This could get ugly...

Nah.. if ever I could say, it's all good...

There is the textual basis (Deuteronomy) and the exegesis (Talmud) or rabbinic commentary-on/extrapolation-to the normative application.  I glossed over the distinction for brevity's sake.  "For brevity's sake!  For freakin' BREVITY'S sake!!  Lemme get this straight.  Hammer cut out relevant details to be ...brief?"   Well, stranger things have happened.

The Trace Elliot Dual band comp splits the signal into two freq bands (no variable crossover though). Here is the schem http://britishaudioservice.com/schem/Pedals.pdf and the manual http://www.britishaudioservice.com/inst/PEDALS.PDF for compression details...
Might be interesting ... 
Regards,
Markus

Many thanks for those links.  The Trace Elliot pedal, aimed at amplification of acoustic guitar, from what I gather, uses a 2-way split, but indicates that they employ faster time constants for the highs than for the lows.  One needs to qualify what "slower" and "faster" mean, though.  I gather that, for the highs, since they come one fast (rise time for 8khz is faster than for 200hz) and don't last long, a fast attack and fast decay are entirely reasonable.  For lows, the rise time is slower, and so is the decay time. 

Note that this also does wonders for minimizing envelope ripple and the "distortion" arising from that.  Since the ripple is most noticeable when there is less smoothing and bigger interpeak intervals (within a note), imposing more smoothing by having a more sluggish response helps to smooth out the ripple and make it less audible.  In the case of the higher frequencies, and ripple resulting from too fast a rectifier response will be largely inaudible because it is so high.  In general, I was thinking about preserving punch and tonal attack, but if the frequency-division helps to provide a smoother, more artifact-free, sound...well, hell, I'm in.

The Trace-Elliot pedal provides much excellent fodder for discussion.  Note the following aspects of it:

• It divides the highs and lows via fairly simple 2-pole lowpass and highpass filters.  That only two resistances in each filter section need to be altered to change the corner frequency makes it ripe for having "breakpoint experiments".  Also suggests that using a multi-band approach maynot have to be all THAT complicated to work reasonably well.  I haven't heard the TE unit, but they have a reputation for quality, and are in the price range that more complex and pricier circuits can be tolerated by customers, so I take that as an indication that simple CAN work.
• You will note that the highpass section is followed by a unity-gain inverting stage (the op-amp with a 47k input and 47k feedback resistor).  This would seem to be used to cope with group phase delay in the low-end or something like that.
• The rectifier circuit for the high end is different than that for the low end.  Note that the drive into the rectifier is variable for the low end, but fixed in the high end.  In contrast, the high compression sidechain has a variable output level (V4) as opposed to the variable input level (V2) on the low side.  Note the 100k resistor to ground after D10 but 47k to ground after D6.  Note, as well, that diode/1k bypass around 22k resistor R49.
• TE allows the user to bypass the gain reduction on the high end, separately from the low end.
• Finally, much like the original E-H Black Finger, and the PAiA Infinity Plus, among many others, the TE pedal varies/reduces gain by sticking an OTA (IC3) in the feedback loop of an op-amp.  I'm surmising that little blighter could conceivably be an LM13600 if we wanted it to.

Assuming the unit delivers desirable sound, there are a few things that can be drawn from this example as potentially good practices.
1) Two sidechains may be enough.  Not necessarily "best", but "enough".  (That Akai pedal is.....nuts)
2) LESS compression of high end may be one of the true underlying goals of multi-band compression.  Presumably one of the holy grails of guitar compression is to allow the subtle tonal variations and harmonic content that accompany variations in pick attack to be heard, without having to pay a major price in managing volume for that to happen.  Keeping the lows in check, while letting the highs roam free will result in a little more dynamic range than full-spectrum compression, but not THAT much more, given the comparatively puny amplitude of the HF components.

Note that I am not advocating cloning of the TE pedal, but rather attempting to draw some rules of thumb for perhaps collaborating on our own potential "1590-friendly" local design.  The TE pedal is but one design, explicitly aimed at processing acoustic guitar.  I'd be interested in learning about any other units out there, that have a similar OR different objective.  AS crazy as I am for manipulable parameters, the ideal is to have something that permits "ideal" compression tone with maybe a trio of knobs, and 2 toggles.  Perhaps, between the bunch of them, we can generate some sort of composite sketch of what our ideal compressor might be - a little "Meek 'n Mighty" stompbox!

[StephenGiles]

In the What Compressor, there is some interesting filtering in the path to the peak followers.

[Ge_Whiz]

I've always thought that the Old Testament could benefit from significant compression and noise reduction.

[R.G.]

For those (and there are several, judging from the responses here) who are interested in the history of the Bible, Talmud, Koran, etc, there is a fascinating book titled "Who Wrote the Bible?".

I was amazed to find that one can separate, for instance, the story of the Flood into two different but interally self consistent and complete stories of the Flood. It appears that there were two stories by two different writers interleaved. Likewise, it took this book to whack me in the face with the fact that Israel and Judah were two different countries at the same time. Stuff I'd heard but never placed in an adult context.

There's a lot more. The writings and the literary construction of the Bible are related to the sociopolitical goings-on at the time in the place where the Israelites lived.

It's a great read. For those of you who can sit still for this kind of thing.