EQ + (Engineer's Thumb x2+) = multiband comp.?

[MrStab]

hi,

a random idea popped into my head. i love my Engineer's Thumb, it's simple yet effective and it's become part of my live rig when i didn't really even need one, so there would be absolutely no benefit to attempting this idea i'm about to suggest... oh wait, yes there is: it would be cool! unlike me for thinking it'd be cool. lol

i built the parametric EQ here the other day, with eventual success: http://www.geofex.com/article_folders/eqs/paramet.htm (4th pic) and i started thinking: what if you took each split band from the EQ, and put it through multiple Engi.'s Thumbs, so both sides of the LM13700/600 (mirror the board and/or use multiple boards) are simultaneously in use over separate EQ bands before they return to the mix again. if that makes sense.

• is this possible/practical/really stupid? would noise be an issue?
• what would i need to remove in order to pull this off? i mean would the compressors benefit from their inputs being before or after the last opamp in the EQ schematic? like this (U1b):
  
• any other stuff?

if somehow the compressors need to be after the EQ's last opamp, then i thought a more sure-fire route would be just to duplicate the entire EQ circuit with one band on each.

i think it'd be cool to have the ability to just switch to a plain EQ, ie. turn the compressor off as well. fairly straightforward i'm guessing, just switch the input to the output stage or something. but i'm a long way off that bridge. doesn't even need to be that specific EQ circuit, but i'm guessing a lot of them work similarly anyway.

there are a handful of threads on multiband comps, and the Engineer's Thumb's DNA still has strands of more common (inferior!) commercial OTA compressors on the market, so i'm having a look at them and trying to soak up as much as can to see if any of it can be applied to these particular circuits.
but as it is these are fairly particular circuits, i thought i'd ask you guys for your input
!
cheers!

[mistahead]

I'd put the compressors late in the chain as you've done and LFO the inputs to sweep...

Some reason I like your idea... not sure why... but why not add more bands and compress/limiter each, maybe EQ them with a simple high/low-cut-and-blend... 

[R.G.]

Compressors all have issues with the circuits that determine what the output level is, so they can modify gain based on it. This has to have some kind of averaging, because it inherently involves converting the AC signal to a time-varying DC level, through whatever means. There is an averaging time constant; at least one, sometimes more, if you're doing different attack and decay times.

The averaging you need for treble sounds, 4k and up, is way different from the averaging you need for bass sounds.

Multiband compressors "fix" this a bit by splitting the signal into frequency bands where the time constants can be different based on what you have to average.

Multiband compressors have the problems of compressors, mitigated somewhat by being able to tailor the time constants. In trade, they get the problems of crossover networks, needing to do smooth, seamless, constant voltage and constant power splitting of frequencies, without introducing phase-shift problems at the crossover point(s), and some of the problems of mixers because the compressed stuff needs to be mixed back together, again seamlessly.

Parametric EQs are not the best possible choice for a splitter. I would look into (well, I have looked into) active crossovers for multiband work. Rod Elliot has some decent ones over at Elliot Sound Products. See the Linkwitz-Riley crossovers; http://sound.westhost.com/project09.htm

[mistahead]

Active crossovers you say...

Reading you say...

Thanks RG - reading list is longer than build list now, I think you win the Internet!   

[MrStab]

thanks for the explanation, RG. i've pretty much just tried to learn what you mean by crossover filters, sure enough it seems self-explanatory, if i have it right: basically bandpass filters with a rolloff as they approach the starting frequency of the next band? just being open bout my naivete here!

i'm not quite sure exactly what you mean by averaging time constant, i'm guessing it's to do with the attack/decay etc. of the note or the compressor's response to that?

definitely food for thought from what i grasp so far. if i managed to wrap my head round this, would it be feasible to just use two ranges (purely to strike a balance between the immediacy of gigs etc. and functionality, unless limiting it to this number would be counter-productive), and is the frequency boost/cut within this range is adjustable, either through varying the crossover circuit itself or further (non-parametric) EQ controls within each? or would i just need more crossovers?

yknow, that's just got me thinking of Cubase plugins i've used in the past, and how it won't let you overlap the boundaries of each section so it just shunts the next one along when you drag instead. if these circuits are adjustable, a dual-gang pot might be all you'd need if you only had 2 bands - one wired backwards from the other. i can't see how you'd apply it to more though, unless maybe you added a rotary switch for different "modes" and wired different bands to a few dual pots. so one mode would widen the extent of the high end whilst squeezing the low and mids. but that would be limited in functionality unless you could somehow have a say over two bands at once. triple-gang? lol

that's just an irrelevant idea, in any case i'm keen on the method you're outlining

cheers!

[R.G.]

thanks for the explanation, RG. i've pretty much just tried to learn what you mean by crossover filters, sure enough it seems self-explanatory, if i have it right: basically bandpass filters with a rolloff as they approach the starting frequency of the next band? just being open bout my naivete here!

Nothing wrong with naivete. We all start there. You're doing the right thing and filling in the voids.

Crossover filters are set up so that there is at least one lowpass filter and one highpass filter. The overall idea is that the output of the lowpass filter is everything below the crossover frequency, and the output of the highpass filter is everything above the crossover frequency. The term "crossover" comes from the use of crossover networks to route bass frequencies to bigger, bass speakers, and high frequencies to smaller, better-treble-response speakers. The sound literally "crosses over" from one speaker to the other.

But you have the general idea: as the output of the low pass filter tapers off, the output of the high pass filter picks up.

Humans are quite picky about their frequency responses, so they get irascible about there being funny peaks and valleys in the frequency response right at the place where the sound crosses over. It's not in general possible to have analog "brick wall" filters which have a constant output right up to one frequency, then stops entirely beyond that. There is always some tapering off of sound beyond the crossover frequency, on both the low and high pass sides, so there is a range of frequencies output by both high and low pass filters. For the crossover to work well, these have to add up to a constant signal level from one side of the crossover frequency to the other. This is quite tricky to design well.

A simple high/low crossover has one high pass and one low pass filter. If you want three bands out, you have two crossover regions, and two high passes, two low passes. It is tempting to try to make the middle output be a bandpass filter, all neatly integrated, but there are technical difficulties that make this quite difficult to do well and keep good performance in the two crossovers on each side of the bandpass. So these "bandpass" filters in crossovers are always done with the combination of a high pass and a low pass filter, not the bandpass circuits you see.

i'm not quite sure exactly what you mean by averaging time constant, i'm guessing it's to do with the attack/decay etc. of the note or the compressor's response to that?

AC signals alternate (Doh!) above and below some ground reference voltage. Figuring out how big the signal is can be a problem, because the average level is always zero; it alternates above and below the reference. Generally, the practice is to use some form of rectifier diode to pick out the peaks of the waveform.

This is a precision version of what happens in power supplies making DC out of AC for power. Here's a good way to look at that: http://www.geofex.com/Article_Folders/Power-supplies/powersup.htm
The first illustration on the top right shows how a diode lets a capacitor charge up to nearly the peak of the AC waveform on each half-cycle of the AC waveform, and then run down between peaks until the next cycle makes it re-charge again.

In power supplies, the AC coming in is nearly the same peak voltage all the time. An audio signal would have a peak level that varies up and down. We need to know how that peak level varies to do a good job of compressing. What we really want to have is for the capacitor voltage simply connect the dots between peaks of every signal peak, even if the audio signal suddenly jumps to many times the previous level, or suddenly plummets to nearly zero. The problem there is that those valleys between signal peaks look like the signal is dropping out too. We need to have a way to know that a fast-dropping signal is actually the signal getting smaller, not just the valley between two peaks. This is in fact what the capacitor is there for - it holds up the signal between peaks, averaging the peak signal, but running down to follow the AC signal if it suddenly gets smaller.

If you look at the upper right picture, you see that when the diode quits conducting after the peak of one wave, the capacitor voltage runs down as its load pulls current out of the cap. The speed with which this happens is determined by the capacitor value and the load resistance on it.  It happens that the product of the capacitance in farads and the resistance in ohms produces a time in seconds. This is the time constant, and that's the decay time constant in a compressor: how fast it can sense signal level dropping. This HAS to be longer than the time from signal peak to peak or you'll get ugly buzzing ripple on the output of the compressor from the capacitor telling the compressor to change gain too quickly.

And now we run into the central issue with frequency bands and compressors. If one averaging capacitor has to cover all frequencies from 20 Hz to 20kHz, there are 1000 cycles of 20kHz in every single cycle of 20Hz.  The right averaging capacitor time for bass (say, 20Hz to 300Hz) is very much too slow for treble (5kHz to 20kHz); what follows bass well is too slow for treble; what follows treble well is too fast for bass.

definitely food for thought from what i grasp so far. if i managed to wrap my head round this, would it be feasible to just use two ranges (purely to strike a balance between the immediacy of gigs etc. and functionality, unless limiting it to this number would be counter-productive), and is the frequency boost/cut within this range is adjustable, either through varying the crossover circuit itself or further (non-parametric) EQ controls within each? or would i just need more crossovers?

Two bands, each optimized for one range of frequencies is a huge advantage over a single compressor for the whole audio band. Three lets you tailor compressor response times even better - but you pay for it in twice the complexity of the crossovers. More than three gets complex, and you rapidly run into the law of diminishing returns.

yknow, that's just got me thinking of Cubase plugins i've used in the past, and how it won't let you overlap the boundaries of each section so it just shunts the next one along when you drag instead.

You're right - they're doing crossovers for you.

if these circuits are adjustable, a dual-gang pot might be all you'd need if you only had 2 bands - one wired backwards from the other. i can't see how you'd apply it to more though, unless maybe you added a rotary switch for different "modes" and wired different bands to a few dual pots. so one mode would widen the extent of the high end whilst squeezing the low and mids. but that would be limited in functionality unless you could somehow have a say over two bands at once. triple-gang? lol

In general, for smooth, "polite" compression, you don't need the frequencies to be adjustable. The frequency splitting is there for making the compression work better, not for doing audio stunts.

That's not to say that you might not want to do audio stunts.  This is, after all, an effects forum.

[MrStab]

here's my current ultra-layman interpretation of your description of the diode/cap system: the diode acts as a gate of sorts, but the cap smooths off the gaps between peaks to prevent it appearing as just a gap? a crude understanding but beats what i had 5 minutes ago! i'm not sure how i would go about applying that to this situation, though

it was actually a page describing the basics for subwoofer/tweeter distribution in speakers where i got the rough understanding of crossovers from. thanks, google. do the crossover points have to be balanced in order to not be perceived as a significant difference, then? like a weird kinda upset in the spectrum at one point, regardless of that point maybe being mildly additive in level?

so would you advise against the 3-Way Linkwitz Riley crossover in favour of just the 2-way circuit (mono, ofc)? i was thinking of a way to control the lines between 3 bands, as it were, using 2 dual-gang pots with one set of lugs controlling the low or high sides and the other sets of lugs always proportionately resizing the middle to a constant ratio, but your explanation of the precision required here has made me think that's probably not practical.

my strategy so far, taking this new info into account:
i could probably more or less mirror the Engineer's Thumb circuit, as the opposite side of the OTA is just blocked off with nowhere to go (bias inputs seem independent) and could take a feed on either side from the crossover circuit. would i then just mix the signals into the non-inverted input of an opamp, possibly with a pointless mix knob between the two extremes?

thanks again

[MrStab]

Unsure how to proceed, i decided to have a go at a stripboard/vero layout for this crossover circuit (well, a mono version):



there are probably some mistakes (e.g slight trace cut error behind one opamp) but here's my layout attempt:



one half of the TL072 takes care of the input buffer, the other half will combine the signals post-compression. not 100% sure on a mix pot (low/high balance) yet, but it could be a nice feature.

i'm unsure whether or not i have to vbias all the opamps - i know that's probably a given, but i'm unsure as it doesn't indicate in the schematic therefore i've only vbiased the TL072 in my layout for now.

i used the calculator app provided by that site to work out a frequency that seemed like a practical middle point whilst still being able to use common parts - 1k resistors and 100nF caps in this case. could i get away with using 2x100nF caps in parallel for the 200nFs required? i put output/input caps where the signals would be sent and received from the compressors, wasn't sure if that's necessary.

also, re. the whole concept of crossovers: in crude terms, the frequency i calculated is what the low channel will stay below and the high channel will stay above, right...?

[R.G.]

here's my current ultra-layman interpretation of your description of the diode/cap system: the diode acts as a gate of sorts, but the cap smooths off the gaps between peaks to prevent it appearing as just a gap? a crude understanding but beats what i had 5 minutes ago! i'm not sure how i would go about applying that to this situation, though

Your layman's interpretation is correct. The cap has to be there to smooth over the valleys between peaks. In a compressor, there will be a variable gain amplifier that is controlled by the voltage on that cap. Any ripple on the cap will be seen as a ripple on the signal the actual compressing amplifier controls. The little run-down ripple between peaks is even a problem, as bass frequencies make this run-down worse, because there is a longer time between peaks.

Going to multi-band lets you have more than one peak detector. The bass peak detector can have a big averaging cap, so there's little ripple on bass frequencies. The treble one can have a smaller cap, and follower faster notes better. A good multiband has two peak detectors, with different averaging times. The averaging times are those "time constants" I was mentioning.

so would you advise against the 3-Way Linkwitz Riley crossover in favour of just the 2-way circuit (mono, ofc)? i was thinking of a way to control the lines between 3 bands, as it were, using 2 dual-gang pots with one set of lugs controlling the low or high sides and the other sets of lugs always proportionately resizing the middle to a constant ratio, but your explanation of the precision required here has made me think that's probably not practical.

I would advise starting with a two-band compressor, and only going to more bands if you have a clearly-stated need for more. This is primarily on the basis of benefit achieved per effort expended.

i could probably more or less mirror the Engineer's Thumb circuit, as the opposite side of the OTA is just blocked off with nowhere to go (bias inputs seem independent) and could take a feed on either side from the crossover circuit. would i then just mix the signals into the non-inverted input of an opamp, possibly with a pointless mix knob between the two extremes?

I would make two channels of the ET (if that's the compressor you like), and feed each one from the output of one of your crossovers, high and low. I would find the averaging cap in the ET, and make it significantly larger for the bass side, smaller for the treble side. I would then mix them 1:1 into an inverting mixer made from an opamp, and not bother varying the mix. Inverting mixers are much better at not having interaction between the mixed signals. If that then turned out to be overall inverting from input to output, I'd use another inverting opamp right at the output to re-invert the phase to following.

[MrStab]

thanks for the further input, RG - it's much appreciated.

the only part of the Engineer's Thumb circuit i can find that seems slightly similar to the configuration in the Geofex diagram is what seems more like polarity protection. any clues as to what i'm looking for, a likely size or anything? all i'm going with at the moment is any caps near diodes. the schem is here http://www.freewebs.com/valvewizard2/engineersthumbschem.jpg

by inverting mixer, do you mean just one channel's output into an opamp's non-inverting and the other into the inverting input, or something more like this:



? phase is an issue that concerns me, as i might find it hard to tell. never ask me to fine-tune a guitar with harmonics or ask if a picture is hanging straight...

[tubegeek]

That is indeed an inverting mixer - also called a "virtual earth mixer" - look on Rod Elliott's site "ESP" here:

http://www.sound.westhost.com/project30b.htm

[MrStab]

thanks for the link, tubegeek. most of those circuits seem to be intended for more full-fledged mixers/desks etc. - could i get away with using the simple 1-opamp method i posted? i guess the non-inverted input being grounded is what makes it virtual earth? i recall a headphone amp i looked at a while ago which had a virtual ground and the V- wasn't connected to the rest of ground - is this the same? if so, how could i apply this to a circuit in which all grounds are connected to V-?

i asked an audio engineer friend for his opinion in the frequency split i should use, and he thought 600Hz seemed more appropriate. so i'm thinking something 600-800ish just so i can keep it precise with non-obscure parts. he might be coming at it from a more general audio perspective, but i asked him to bear in mind the frequency limits of guitar/amps etc.. What do you guys think?

i think once i identify the smoothing caps in the Engi's Thumb, i can move forward. there are two candidates for where it might be on the schematic but i'm just not sure

[samhay]

I was thinking about this a couple of months ago. The work-in-progress schematic is below. It uses 2 stock Engineer's thumbs (as a starting point) and the active crossover from my Anharmonic tremolo.
The schem is not verified and is probably full of errors, but if treated as a block diagram, might be helpful.

Sorry about the large image, but it's a large schem.

[MrStab]

thanks for sharing, Sam. at the moment i'm just over halfway building 2 Engineer's Thumbs on the same board, so i still have time to work on my approach to frequency splitting.

i see yours is adjustable, which would be really cool, but from what i've inferred from RG's comments on crossovers, variability didn't seem to be advised. have you built this circuit more or less, even if that particular schem isn't verified? if so, how'd it work out?

[R.G.]

Be sure to adjust the filter cap on the signal sense section so you can take advantage of all the work you've gone to to split the frequency ranges.

[MrStab]

ive made one of (what i think are) the smoothing caps the standard 47uF and the other 100uF, not vastly different but easily changed. not sure entirely which cap you mean for sensing, though - ruling out other caps, there's a 10nF connecting to Threshold, 100pF to Ratio and a 4.7nF between the OTA and one of the opamps. just started on the crossover part of the circuit, but i'm gonna call it quits for now and pick it up tomorrow (well, today as it's 5am in these parts)

[PRR]

I think you are asking GOOD questions but for a DIFFERENT field.

Multi-band limiters are widely used in broadcast. They work with pre-mixed whole orchestra/band sources. They make real money. The way to make money in radio is to get the MOST of your licensed power, to reach more listeners, and charge more for ads.

To make the most of their power, broadcasters compress all sounds UP to as LOUD as possible.

When you do that, on a pre-mixed whole-band recording, the staccato lead "punches holes in" the legato bass. (Even if plucked, pop bass is fairly steady and long-decay.) It sounds funny, also bass-less.

Splitting the spectrum mid-band is a simple (tho imperfect) way to split typical lead sounds from typical bass sounds. With the rise of Disco and severe cymbals, another split in the upper midrange was wise (especially on FM which has limited szzzzzzz power).

As broadcast limiters are a VERY competitive and somewhat lucrative market, over-kill is part of the racket. No doubt there are 4 5 6 and 27-way limiters-- more gotta be better than less, right? Anyway, more knobs, the longer the Program Manager has to mess with it trying to find "the sound" before he throws the crap in the dumpster.

Single instrument played live by a sensitive player is different. Your volume peaks are not stomping on someone else's musical line. If you are stepping on your own notes, you adjust your fingering and strumming.

As R.G. says, there IS a compromise between best timing for bass and best timing for highs. But guitar is not a particularly wide-range instrument. The fundamentals cover only 3 octaves. There's 2 or 3 octaves of harmonics.

You generally do not want to mash your harmonics different than your fundamentals-- most variations of that just sound weird.

There is probably a use for a multi-band guitar limiter. Is there a need? Hard to say. IMHO, it doesn't urgently need doing. While everything has already been done before (the audio ancients stole all our ideas), there must be stuff that needs re-doing more urgently. Your opinion may vary.

> this crossover circuit

Fourth-Order?? Sell that to some ignorant hi-fi fan, or to the PA guys who might really benefit (except they have better stuff in digital now).

Don't hard-slice your guitar sound. You don't want one sound up to 291Hz and a different sound at 309Hz. You play runs/melodies right through this point (ANY point you pick). While 300Hz is perhaps a good *area* to split guitar, it should gradually transition like 200Hz to 400Hz. FIRST-order crossover. This is as simple as an R-C to pass lows and a C-R to pass highs.

Re-mixing crossover-ed sound has pitfalls. The first-order is minor, you trim one filter a part-octave for a smooth transition. Simple mixing of Second-order outputs causes a NULL (no output) at the crossover frequency. You have to invert one side to sum nicely. (I see Sam posted the 12dB/oct lo-pass plus subtractor--- this does sum OK.) Higher-order can be trickier, though over a narrower band, so may go un-noticed by the listener though may be annoying to the player who knows what he strummed. In loudspeaker work there are much bigger problems from different drivers in different places so the theory of crossover summing is amusing, often moot in the room. But you can't hide behind that in all-electronic Xover summing.

Despite my rant, do it. You have a good building-block in the ET. Hear a little of my rant, the crossovers can be trivial. So $5 or $10 and a whole lot of self-education? Play with it. The first conception is probably NOT what you will end up following, so stay loose.

[samhay]

I never got round to breadboarding this - it hasn't progressed passed the thought experiment stage yet. I know each block works, but the above schem will need a little fine tuning.

You don't have to have a variable crossover frequency - you can replace the dual-gang pot with 2 equal fixed resistors - but it gives you more flexibility, e.g. if you want to use it with a bass, which was the initial rationale for this design.

I'm not sure how you were thinking of doing the final mixing. The active crossover I have suggested will give an almost-flat mix if the 2 gain stages are equal. However, they usually won't be. At this stage, I have independent volume control for each band of the compressor. This is a necessary evil (although you could equivalently have a mix and master volume) as the gain of each stage will vary with the compression/ratio setting.

I have updated the schem to highlight a few points and to fix a couple of glaring errors. I envisaged that each compressor would have variable attack and release times (equiv to R.G.s timing cap comments), so the controls would be crossover frequency and then 2 each of ratio, attack, release and volume. With 9 knobs, 11 op-amps and a dual OTA, this would be a pretty large build and I am have not yet found a need pressing enough to give it a try.

[MrStab]

much food for thought, Paul - a wealth of information slowly dripping through the narrow cracks of my understanding. my own (again, semi-layman) experience with multiband compressors is mostly limited to pseudo-mastering attempts, where as you say a whole band mix is different from limiting different areas of the guitar spectrum.

an audible, sudden divide between the two ranges was a concern - i'd just hoped i could get by on not having too much difference between the intensity of the compression applied. but that in itself borders on defeating the point. to keep up the multiband approach, it seems what i really need is something more variable like Sam's method. the original 3 band scheme would maybe have alleviated (or worsened?) the potential problem somewhat, but RG advised that it might be too much hassle for little return, and it'd require a 3rd compressor.

all is not lost, outside the multiband world. prior to this, i'd felt a need for varying the compression i use between tunes/sections, so with that in mind it definitely couldn't hurt to have two compressors on one board. if i didn't change many critical frequency-altering values within either circuit, i could have them switchable. but then i could i also have them combinable, even if as a "secondary mode", with the flick of a switch.

this project has arisen purely out of the same urge which gave me a memorable 240VAC shock when i was 9: "why not?". lol

i figured having a level control for each compressor pre-mix was inevitable, Sam. i have a source of thinner-but-still-sturdy, non-Hammond enclosures which go up fairly large for cheap (or i could just repurpose an old DVD player or something), so i've pretty much expected this project to be a bit on the large side.
the only reason variability has me scratching my head is because the site RG linked to seemed really adamant about variability and said something insane like 8-gang pot would be required to make their method variable - think your method would offer the precision needed?

thanks guys

[MrStab]

i'm gonna try a vero layout for your crossover section, Sam. just a bit unsure what you mean by -VB and -VA on your schematic?
could probably pull off that whole section with one quad-opamp