Some adjustments to Jon Patton's Bear Hug compressor circuit

[Groovenut]

So many thanks to Jon for this cool little compressor! It's a great comp and after many ours of playing and finding only a few short comings, I decided to see if I could lesson the effect of the one that bugged me most, the over-shoot. With this type of compressor, over-shoot on the initial pick attack seems to be pretty common when using a pick. I did a bit of experimenting and this was the fix I cam up with for the "splatty" attack when the comp over-shoots.

First here's what the waveform looks like during the over-shoot when the mosfet is center biased



Here are the circuit changes I made to minimize the effect



and what the waveform looks like now during the overshoot



Thoughts?

[PRR]

> Thoughts?

It clips because the initial turn-down is slow. Also the waveform goes asymmetric.

You are (stock) getting 4.5V peak-to-peak. That's essentially all you can get with a 9V supply and much of that dropped in the Source resistor.

Yes, half-supply is not a for-sure-good bias with these parts, unless the MOSFET biases with just about 2V.

Your LEDs clamp and round the initial clipping, and do seem to smooth things so the compressor action happens a bit faster. FWIW, we find similar hacks in many old tube limiters (grid-clamping).

The limiter seems to take 10mS to start working. That's not a wrong choice, short distortion is not obvious. However 10mS is the long end of "short", and repeating that blip of distortion eight to the bar may make it obvious.

A different hack would be to shorten the attack. If I have found the original, there is a 4.7uFd cap driven by a doubler through 0.1uFd from an amplifier with 2.2K-4.7K source resistance. 2.2K+4.7uFd is indeed 10mS.

Hem and haw. Frankly I think the plan does very well for how simple it is. If the LEDs do the job for you, that's great. If you want "more precise" limiting, you probably want a more complicated gizmo. Serious limiters are serious engineering and often complex to build.

[Groovenut]

Thanks Paul! I greatly appreciate the input 

I think the easiest way to get a faster response is to make the 4.7u cap driven by the doubler smaller so that it takes less time to "fill the bucket" and activate the fet. However, for now the LEDs work pretty well for a simple fix.

I will experiment more and report my findings back.

[Transmogrifox]

Here's my work-in-progress take on this kind of compressor:


The envelope detector method used here has a faster attack on high amplitude signals but not as fast for low amplitude signals.   It helps tame the distortion on quick pick attacks.

Here's a simulation shot of a 1.0Vpp attack. 



Then a 200 mVpp attack:


This is with "attack" set at 2.2k.  Everything shortens up considerably by rolling that back.
It sounds pretty good as is.  It's not quite the same as a minor "adjustment" to the Bear Hug, but it's not entirely different either.  It does help mitigate the problem you're having by allowing generally longer attack times for lower level signals while more rapidly attacking high level signals.

[Groovenut]

Transmogrifox, thanks for that!

Would you mind giving more info on the J1 arrangement? Looks like you have the source tied to 9V which should give more than enough voltage drop to turn off even high Vp fets, and you have it working against a 10k resistance to set the minimum gain of J2. Can I get a more detailed description of the linearizing components R12, R13 and C4? Without putting it into the sim, I would guess they help J1 function in a more linear on to off sweep.

Thanks for posting your circuit. It does indeed look very interesting!

[PRR]

> linearizing components R12, R13 and C4?

R12 appears to be the "plate resistor" for M2. Also the recovery rate resistor. An interesting variant. I confess the drawing layout had me baffled a while.

"Linearizing" would seem to be the R13 R9 network. These would typically aim the gate at half the _AC_ voltage source to drain. Here we have 33:100, not half. Not sure if that is an exact answer allowing for other factors I am missing; or a "good" ratio found by experiment. Also "linearization" (and distortion) works different here, where we change a "cathode resistor" instead of the lower arm of a resistor divider.

[Transmogrifox]

R12 appears to be the "plate resistor" for M2. Also the recovery rate resistor. An interesting variant. I confess the drawing layout had me baffled a while.
Here we have 33:100, not half.

It is closer to half than it initially appears. If you look at the AC signal path from the Drain at R13 you will see that R9 and R12 form a parallel combination, so it's actually 33:(100|51) = 33:33.7, so it still follows the convention you would expect.

The purpose of R12 is to prevent the gate from being pulled lower than its useful range.  It takes much longer to recover from an extreme overshoot like that, and it shows up in the form of a volume dip after the initial attack.

R5 combined with its AC parallel equivalent, R6, set the minimum gain.  Again this mitigates the initial attack overshoot preventing the volume dip immediately following the attack. R5 further serves the function of reducing the thump on attack since it holds the DC state around J1 more static, keeping less extreme low frequency feed-through as J1 turns off.

For an easier understanding of the circuit, ignore the ratio control circuit for a moment.

You can see J2 is just the amplifier transistor used to set the gain of the circuit.  "M2" is biased just a little bit below its turn-on threshold, so it only responds to the positive 1/2 cycle of the signal sensed from J2 drain.

The response at M2 drain is that of a current source.  This is why higher amplitude signals attack faster:  It's a current source, and higher currents correspond to faster discharge of C6.  The smaller you set R8, the more current you get per input voltage, and so the attack gets faster with lower value of R8 Pot.

Then when the signal lets off, C6 charges to 9V through R9 and R12.  R9 dominates as the release time.  If you wanted add a release control you could change R9 into a pot at the expense of changing the linearizing characteristic set by the current combination.

The DC blocking from C4 was inherited from the most current rev of my rEAgenerated tremolo in which I saw the current leaking through this node preventing getting 100% depth.  In this case it's less important as a turn-off aid as it is to reduce the amount of low-frequency thump feed-through from envelope follower at the gate.  Of course, removing it altogether will make the biggest impact on thump.  This combination was more of an empirically-formed combination found in simulation looking at what had the best effect on distortion while still keeping suitably small thump feed-through.

Now add in the ratio control.  This works by letting the "EF" follower lower the bias on M2, which dynamically increases the threshold.  By jacking up the threshold dynamically, this makes it so the output doesn't tightly converge to the static threshold, but it lets it increase by a small amount.  This effectively lowers the ratio.  It's no exact thing, but it does soften it up a little bit compared to the traditional feedback compressor.

The take-home message about the network around J1 is most of it is aimed at reducing thump.  I saw in simulation that the thump feed-through pushed the signal asymmetrical and increased clipping on attack.  It also formed a low-frequency feedback path around the envelope detector which was causing those second order effects, such as the initial attack overshoot.

The linearizing effect of R13 and C4 help, but they're not the salient characteristic if this circuit.  It would still probably sound fine if you took them out.

Let me know if it's still foggy what all that stuff is doing.  I think I covered it

[Cozybuilder]

I was happy with my Bear Hug, but now I've got to build another.    Nice work Law & Trans!

[PRR]

> I think I covered it

Indeed. Thanks!

[midwayfair]

Just saw this. Sorry I wasn't around to comment ... I've been learning about microphones, preamps, transformers, and other nonsense plus writing and recording an album in February.

So, a couple things:

First, thank you Transmogrifox for all the work you've put into the various little tweaks to the design and the explanations.

Groovenut: Groovenut, if you haven't already done so, you might look through Transmogrifox's other posts on the subject where he talks about the attack time in the original Bearhug (which apparently isn't quite so simple as I assumed and depends on the compression pot setting). You may also be interested in taking a quick look at the Lovesqueeze, which also uses an LED limiter in the feedback loop of an op amp. (If you go way back to my original posting of the circuit, this was sort of a mashup of the Lovesqueeze and the Improved EA Tremolo.)

My problem with the LED limiter would be that there is distortion connected to larger signal sizes regardless of the compression setting. If you only ever put the compressor right after a guitar with a predictable (<2V) signal, then the LEDs won't do anything except on the attack of an uncompressed note. But larger signals can clip an LED -- which clips sooner than the power rails on the MOSFET -- even if the FET is driven up high enough to put the gain at 1x. Just something to consider.

I do recognize that some people have problems with distortion in this device ... and in fact the reason it uses a MOSFET is because when I first designed it, it had the lowest distortion and sounded the best when it distorted. The feedback cap between the gate and source should also minimize transients, since there's no gain applied to signals of a particular frequency, but I don't know what you've been using for that cap, and it obviously needs to be tailored to personal taste. (Josh at 1776 wanted the 150pF there, which I found way too large for a fingerpicker like me ... but you might try it if you didn't already.)

PRR's right that getting a truly invisible attack is very challenging when the goal is to make things as simple as possible for sticking it in a tiny box. However, the 10mS isn't written in stone. You can cut it down to 1mS in the passives network by using a 470nF cap instead of the 4.7uF (i.e., 1/10), and then 10x the resistors in parallel with the cap. Whether or not you'll actually get 1mS in the real world I couldn't say because I'm not sure how small you can get with the hold cap; at some point you would need a lower-impedance driver, which means adding an emitter follower at a minimum, and probably a high-current one, which maybe isn't such a good idea in a tiny box.

[LightSoundGeometry]

So many thanks to Jon for this cool little compressor! It's a great comp and after many ours of playing and finding only a few short comings, I decided to see if I could lesson the effect of the one that bugged me most, the over-shoot. With this type of compressor, over-shoot on the initial pick attack seems to be pretty common when using a pick. I did a bit of experimenting and this was the fix I cam up with for the "splatty" attack when the comp over-shoots.

First here's what the waveform looks like during the over-shoot when the mosfet is center biased



Here are the circuit changes I made to minimize the effect



and what the waveform looks like now during the overshoot



Thoughts?

Hey Groovenut ..lol

I just talked to you through email on reverb..I really liked your pedals..not many thing catch my eye but yours did lol

I love your work man, its very nice looking and I am sure it sounds just as fantastic

I been trying to put everything inside 1590a too ..I need to learn how to board etch

[ljudsystem]

Sorry for resurrecting this old thred but I'm interested in trying out this mod:

You can cut it down to 1mS in the passives network by using a 470nF cap instead of the 4.7uF (i.e., 1/10), and then 10x the resistors in parallel with the cap.

I'm not sure which resistors those are though, is it R5, R6 and R7 in the Bearhug pdf?

[PRR]

Doc:
https://cdn.shopify.com/s/files/1/0022/3952/9069/files/Bear-Hug-V2-1.pdf?16138734983046189115

Quote
>> You can cut it down to 1mS in the passives network by using a 470nF cap instead of the 4.7uF (i.e., 1/10), and then 10x the resistors in parallel with the cap.
> I'm not sure which resistors those are though, is it R5, R6 and R7 in the Bearhug pdf?

No.

[ljudsystem]

Ok, thanks. I thought they controlled the decay, I guess they do both attack and decay?

[PRR]

The smaller capacitor is faster attack. Also faster decay, so we increase those resistors.

[ljudsystem]

Ok, two more questions:

* C2 (47pf) controlls treble, right? If I raise the value It will cut more treble?

* C2 (10nF) and R5 (2k7) form a hpf? But, when I enter the values in a calculator I get a cut off at 6 khz. Uhh, what am I missing?

[PRR]

> C2 (10nF) and R5 (2k7) form a hpf? But, when I enter the values in a calculator I get a cut off at 6 khz.

C2 is separated from R5 by Q1, which for our purposes is a layer of pure-pure glass.

C2 and R2 set an input high-pass at about 2Hz. A bit generous but does not cost extra.

[iainpunk]

very interesting thread to read, but i thought the initial overshoot was a feature, not a bug...
i find it gives more attack and power to the longer sustaining note, especially for bass!

cheers

[Vivek]

What are the attack and decay time constants for this compressor ?

[iainpunk]

i thought it was
10ms attack
100ms decay
but i could be wrong here since it differs between versions

cheers