THE ENGINEER'S THUMB... At last, a better compressor!

[merlinb]

btw, using two OTAs ... increases the relative noise floor by SQRT(2) at idle via the fudging bias resistor R13 (1M) ... ie., the AGC/NFB mechanism is always on, even with no signal ... as a result, a relatively noisy circuit that gets worse with higher ratio

Unfortunately you are incorrect. Using boths OTAs in parallel (there are two in the package, remember) reduces overall noise, albeit not by 3dB. At idle (ratio at max), using one OTA I measure an output noise of about 120uV in the audio band, but with two it is reduced to about 106uV, a 1dB improvement at essentially no extra cost. Certainly not worse, and a lot quieter than the Dynacomp with which it directly competes.

the headroom fix is actually rather simple if you think about it, the noise is as well ...

Why don't you share it with us, oh wise one?

[rutgerv]

I was trying to understand how the original ET actually implements its threshold.

The threshold is a by-product of the offset characteristics of the side chain opamp. In theory the circuit should start compressing with infinitely small signals, but in practice the precision rectifier just doesn't rectify signals smaller than about 10mVpk; it pretty much ignores them. That's why you can't see any obvious threshold-detection mechanism in the schematic; it's sort of a happy accident.

Thanks so much Merlin! I suspected something like this.

@PRR: good suggestion! I already read most THAT application notes on this topic, but they keep suggesting their own integrated components for the envelope detector, while I really liked the sidechain of the ET (partly because of simplicity). the SSI part fI know fairly well, as I have designed several Eurorack modules with them.

@Eb7+9 It seems you mix up two people posting here. I do not have a problem getting the ET side chain to work. I merely needed some more understanding of the ET to be able to tune the side chain to my linearized SSI2164 VCA.

Rutger

[Rob Strand]

Unfortunately you are incorrect. Using boths OTAs in parallel (there are two in the package, remember) reduces overall noise, albeit not by 3dB. At idle (ratio at max), using one OTA I measure an output noise of about 120uV in the audio band, but with two it is reduced to about 106uV, a 1dB improvement at essentially no extra cost. Certainly not worse, and a lot quieter than the Dynacomp with which it directly competes.

I played with parallel OTA/VCA about 25 years ago.  It always reduces noise.    How much it reduces comes down to finer points, like internal resistances which create thermal noise instead of shot noise.    There was a thread about 5 years ago.

[rankot]

Merlin, two more questions - I've noticed that you removed diode going in parallel with base and emitter of BC327. Why? What was it's purpose?

Also, is there a simple way to add some post compression gain? Working like this, Level pot seems to be of not much use - it's very near the unity gain at maximum setting. Shall I try to change R15 or R17? EDIT: Never mind 2nd one, I've found answer here: https://www.diystompboxes.com/smfforum/index.php?topic=97123.msg1092872#msg1092872

[Eb7+9]

@Eb7+9 It seems you mix up two people posting here. I do not have a problem getting the ET side chain to work. I merely needed some more understanding of the ET to be able to tune the side chain to my linearized SSI2164 VCA.

Rutger

sorry, I looked at the internal diagram and saw what looked like a class-AB ota on its own ... but then again, not sure why you'd want to use this IC here since the input wants to see a current, not a voltage, and the side-chain puts out a buffered voltage before going into the V/I converter ...

[merlinb]

Merlin, two more questions - I've noticed that you removed diode going in parallel with base and emitter of BC327. Why? What was it's purpose?

It was for reverse polarity protection, but I decided there was no ultimate risk of that.

Also, is there a simple way to add some post compression gain? Working like this, Level pot seems to be of not much use

From your other thread we now know you haven't built my ET circuit, you've built some other circuit of your own devising, so I can't answer that. My ET Iss.4 already has some gain built in with R14.

[rankot]

From your other thread we now know you haven't built my ET circuit, you've built some other circuit of your own devising, so I can't answer that. My ET Iss.4 already has some gain built in with R14.

I've built three of them - and the last one is actually issue 4, together with that 4k7 at R14, but I don't get much gain with it, too. So I actually did what you proposed earlier, to reduce R18.

[jonny.reckless]

btw, using two OTAs ... increases the relative noise floor by SQRT(2) at idle via the fudging bias resistor R13 (1M) ... ie., the AGC/NFB mechanism is always on, even with no signal ... as a result, a relatively noisy circuit that gets worse with higher ratio

Unfortunately you are incorrect. Using boths OTAs in parallel (there are two in the package, remember) reduces overall noise, albeit not by 3dB. At idle (ratio at max), using one OTA I measure an output noise of about 120uV in the audio band, but with two it is reduced to about 106uV, a 1dB improvement at essentially no extra cost. Certainly not worse, and a lot quieter than the Dynacomp with which it directly competes.

the headroom fix is actually rather simple if you think about it, the noise is as well ...

Why don't you share it with us, oh wise one?

You might do a little bit better on noise replacing the TL07x with a NE5534 or NE5532 after doubling up the OTAs which should give you close to 3dB improvement if they're the dominant noise source. It made a nice audible improvement on my build. NE5534 is externally compensated so you can set the open loop bandwidth for the application which I found helped a little with stability and transients. I needed to strap everything down to ground pretty tightly to get a veroboard version working without oscillation. It was much easier with a PCB I seem to recall I got about -83dBV A weighted (the stock design was around -74dBV) on the output with no input and the sustain at max, but I might be remembering that wrong. With the OTAs mostly off it's just the op amp self noise you're worried about at quiescent. You've measured -79.5dB with the pair of OTAs so that's not wildly inconsistent.

[merlinb]

You might do a little bit better on noise replacing the TL07x with a NE5534

He would also need to change the source impedance though, perhaps you can explain that to him.

[Eb7+9]

... and a lot quieter than the Dynacomp with which it directly competes.

I didn't realize you were competing with the Dynacomp, or just competing - that would explain part of the confusion ... other than trying to beat a very poor noise figure we're talking two very different types of responses here, one servo based and the other not ...

let's be clear, your 570/571 derivative circuit can't be referred to primarily as an OTA based compressor - not by any stretch ... in the Dyna the OTA is "on-at-rest" and made to simply turn off with signal envelope - nothing complicated, the signal is carried the whole time by the OTA ... in the THUMB's - the opposite on/off thing happens with the OTA, and the overall gain reduction process relies on the precise cancellation of signal currents that could always become distorted "somehow" // the signal is carried by an op-amp ... as well, in your case, there's this situation that requires the use of R13 (1M), the bleeder resistor, to prevent the OTA from screwing things up if it were allowed to shut off completely ... won't go there now

otherwise the differences are plentiful ...

aside from the implementation details, the THUMBS circuit employs a fed-forward side-chain approach whereas the Dynacomp, along with the more natural sounding servo-style AGC circuits, are all fed-back side-chained ... this is not an arbitrary choice, as it lines up with standard control/systems theory and behavior: see Figure 1. here ...

https://core.ac.uk/download/pdf/41828182.pdf

Even though you might not want to admit it (and I can understand why) the active feedback topology you're using is not that new or different; like I said it's perfectly isomorphic in structure to the 570/571 stuff - I can't imagine nobody's ever said this to you elsewhere ... 

again, it's worth consulting the literature // see Figure 7. here ...

https://www.onsemi.com/pub/Collateral/NE570-D.PDF

notice the same general function gain blocks are present, whereas the delta-G used in the 570/571 is simply a 2-quadrant multiplier configured as a wide-input OTA ... both envelope detectors (yours and 570) are "ideal" (low-threshold) style rectifiers - yours is half-wave, theirs is full-wave; not a big deal either ... theirs converts to current directly, yours needs an extra v/i stage - no biggie other than in the details, both are linear ... compare again, the Compressor example in the NE570 paper is a feed-back (servo) style design

https://apps.dtic.mil/dtic/tr/fulltext/u2/a045608.pdf

a fed-back structure not only aims to regulate the output but also the control process as well (two things here) so that the output envelope ends up behaving in a much more even and measured way ... the idea here is that by taking the side-chain feed from the output the corrective process gets proportionally reduced as well (ie., doesn't end up exaggerating or over-shooting the process) ... some compressor designs take care of the second part better than others (LA-2A), why compressors don't all behave the same against dynamics despite addressing the same simple formulation of the problem ...

->> conversely, if one wanted to build a more dramatic or aggressive sounding unit one would do the opposite and run the loop fed-forward, which then is no longer a loop ... that's what you're doing here, producing a more dramatic squash - which can mangle signal real good // not what I'm arguing about either, other than to point out that fed-fwd and fed-bck AGC circuits don't work or feel the same, so let's be careful not to lump one with another simply because they both contain an OTA somewhere inside ...

don't get me wrong, I'm not against exaggerated effects in this sense ... the THUMBS tries to do what its implementation style normally does, no surprise there // rather, it's the limitation(s) of the inner part of the circuit - and especially more so the two part mistake - that will cause problems for some builders ... that's what I'm addressing here

this has nothing to do with fed-fwd or fed-back structuring but more so with how the gain elements are designed-in ... in fact, if you drive hot pickups in the THUMB'S (the way it is) the circuit will not auto-regulate well, or at all ... there's a guy on YT with EMG-81's and the circuit won't give him clean no matter what he tries, to me it's easy to see why ... to be fair, the Dynacomp suffers from the same problem - but it's also a very early design with not much to go on for comparison at the time // ... this is also why optical comps tend to be much better in this one regard ...

again, part of the problem here is the serious bottle-neck that takes place inside the corrective loop - something the 570/571 delta-G circuit doesn't suffer from ... if you're not careful you can easily end up with serious distortion there, especially during the pre-recovery part of the wave-front // that is, even when the circuit can still (later) self-regulate to produce a clean'ish output following a signal burst ...

this is where the ~220r/10k divider ratio going into the OTA comes into play (analyze this from an overall unity gain point of view and you will see why) ... in the Dynacomp that ratio is substantially larger as 2k/17k ... both circuits exhibit headroom issues when it comes to their OTA's, even on relatively low-output sources (below line level) // hence the delta-G circuit block in the 570/571 ...

imposing extra constraints in the area of internal headroom (by adopting an OTA inside this well-known structure) is not what I would call a general design - let alone some kind of improvement in the world of AGC circuits - not by any stretch ... I'll admit it, it's a little ballsy to try pulling this off using a conventional OTA as it opens up a small can of worms // fun situation for doing analysis ... but, from a more practical point of view, consider that all other well known and proven designs (1176, LA-2A, Envelope Lock, ...) can handle a wide range of input signals, from very low to very high - and still be adjusted to operate the same "relative" way ... not so with the THUMBS

mind you, the "finally at long last" corrected version of the circuit is right around the corner ...

---

as for noise, my only comment there is that some of the key resistance levels could be brought down by a decade at least, so it's another oversight there imo ... noise would be made that much better if these were changed, that's all ... the reason why those resistors are high is because the side-chain isn't very efficient the way it converts voltage to current, which I feel is likely why the second OTA was brought it ... again, very easy fix while maintaining a single-OTA structure // and while keeping noise low in the process ...

---

about the treble-boost mod ...

you're now at rev4 of a circuit that has, all along, overlooked a very basic "design" issue // which is the basis of many of my comments here ... to demonstrate what I'm talking about, and without spilling any of my beans today, let's look at what happens when things aren't quite centered in there if you know what I mean ...

in a nutshell, there's a condition you can end up in where the circuit is shaving a ton of harmonics off in the process, which translates into "getting dark when compressing" ... notice the OTA does nothing until we're asking for compression // that should be enough of a hint right there ...

I understand that there "was" a need for this Treble mod - it's obvious to me why it arose, and obvious that it's a band-aid applied there because you have not (yet) figured out what causes other problems // as these are related  ...

again, the treble-loss and distortion issues some people are having come from the same poor design choice ... correct this and the treble boost "fix" is no longer needed

in this sense I would study both the basic NE570/571 AGC circuits as well as the musical Dynacomp, as they both contain the two crucial elements your basic application is missing ... it's a no-brainer, trust me // once you figure this out the circuit can be simplified and cleaned up a bit as well ... and it will work optimally across all builders

---

so ...

I put together a version of your circuit using a discrete implementation of the NE570 side-chain (as per Walt Jung's "simplified" diagram) so that now it REALLY looks quite like the NE570 hardware emulator of sorts that it is ... except for the wide-input delta-G block replaced by a standard OTA (w/o input diodes) and still maintaining the non-inverting opamp gain stage and getting very clean output, it's a 570/571 type circuit for sure ... just to be clear, inverting or non-inverting makes no difference to the dynamics here, one could do either or - it's the NFB current mechanism that matters, and is the same in both cases ... and it works just the same, maybe a hair better, with the 570/571 FW rectifier block ...

I then wired up the side-chain as a fed-back structure and, sure enough, the circuit starts sounding more like the usual servo loop designs ... but it also exhibits the same input headroom issues, and in this sense has very limited application // again, why the NE570/571 came out like it did not using an OTA ... the THUMB's circuit can be made to work fine on only one OTA as there is absolutely no need for two ... in fact, once you guys figure this snaffu out you'll see that using two makes things even worse from a build success point of view ...

I recommend studying the MXR Dynacomp a little more closely and isolating the two "design" components that are missing in the THUMBS ... these two things will help the circuit work better overall ... another way of confirming this is by asking some basic questions about what the THUMBS circuit "needs" in order to produce a clean output

no need to compare the 570/THUMBS to a Dynacomp ... different beast for many a different reasons // ...

the Dynacomp can receive its own face-lift and be made into a much better sounding and feeling unit as well, all while maintaining the primary role of the OTA in the structure ... but that's for some other day

[PRR]

Why are you so intent on tearing-down someone else's work?

[idiot savant]

Why are you so intent on tearing-down someone else's work?

I don't post much, but I've been around here a long time.

There's a difference between constructive and destructive criticism, that seems lost on some folks.


Anyways, the little reference to Walt Jung made me think of the old-school "OTA-trick"

https://electronotes.netfirms.com/AN399.pdf

Wonder if there's any benefit to using that over the normal V-I converter.

If I remember right the old PAIA Gator did something similar..

[garcho]

There's no such thing as an Eb7+9 chord. It's Eb7#9. Unless you're trying to describe an augmented chord, in which case you put the plus in the wrong place and don't need the 7.

[11-90-an]

There's no such thing as an Eb7+9 chord. It's Eb7#9. Unless you're trying to describe an augmented chord, in which case you put the plus in the wrong place and don't need the 7.

Now, now, don't fan the flame...

(Thanks for sharing your design, Merlin... it helped me in my general understanding of OTAs and compressors... )

[Rob Strand]

this has nothing to do with fed-fwd or fed-back structuring but more so with how the gain elements are designed-in ... in fact, if you drive hot pickups in the THUMB'S (the way it is) the circuit will not auto-regulate well, or at all ... there's a guy on YT with EMG-81's and the circuit won't give him clean no matter what he tries, to me it's easy to see why ... to be fair, the Dynacomp suffers from the same problem - but it's also a very early design with not much to go on for comparison at the time // ... this is also why optical comps tend to be much better in this one regard ...

The non-inverting amplfier design means it leaks the raw input through.   For extremely hot signals there's a chance of overload which can't be limited away.   IMHO, blending the raw signal can help hide the perceived treble loss on a compressor.

[EDIT May 2023: As it turns out the Engineers thumb doesn't clean blend.  It does approach a limiter.
https://www.diystompboxes.com/smfforum/index.php?topic=130630.0
]

[Fancy Lime]

https://youtu.be/D6etS-nrW0U

Seriously, though, should we be thankful that after THIS year, we still manage to bicker about compressor minutia or is the fact that this discussion is going the way it is just a sign that we've all had it with the nonsense and just need someone to yell at? I recently punched a wall so hard, I had to plaster it over and paint over the blood spatter. CSI taught me that there is no blood spatter on the first punch. CSI either lied or has a sissy punch.

Have a peaceful end of year y'all,
Andy

[merlinb]

Eb7+9 is just trolling. You can tell because he keeps talking about how he knows how to 'fix the problem', but he never quite gets to the point of explaining what he thinks the 'problem' actually is, or what the fix is! It's just word salad.  
(From the way he types I think he might be mictester from the other forum, who has had beef with me before.)

[Ben N]

Hmmm. I confess I was going to ask for a translation.

[DIY Dood]

Is there some way to eliminate some of that duplication?

Yes, use a switching jack socket:

What is the practical effect (usage) of a side-chain input when we're dealing with a single guitarist and a pedal?

[DIY Dood]

Newb here... I've read through all 39 pages and studied a bit. I will likely have some detailed questions after I study my notes.


For now, is there anyone left producing circuit boards, preferably V3 or V4?

Is there a Vero layout for either of these versions?

Thanks,