Lovesqueeze comp - howzit work?

[Mike Burgundy]

This came up in a different thread ( http://www.diystompboxes.com/smfforum/index.php?topic=100976.0 ), but it's rapidly disappearing off the grid and I'd like to understand the way this thing works.
Schem:


Main circuit path non-inverting opamp with LED limiting and a FET adjusting the gain, okay.
The way I understand the detection circuit is IC1b drives the Schottky diodes. These have a lower forward threshold than regular Si, say 0.4V. That is the threshold voltage - below this nothing changes, above this gain is adjusted.
Once the signal swing gets high enough (either + or -) one diode starts conducting and puts a charge on C10. The amount of charge applied to C10 is determined by how much gets through the diode(s). If you plot a graph of the voltage swing (say a regular sine) and add horizontal lines at V=+0.4V and -0.4V, the charge applied is the area between the sine and the 0.4V lines.
That charge leads to a voltage offset between the capacitors poles - how much depends on the charge and the size of the capacitor.
When the signal ducks below the threshold, compression needs to (gradually) disappear - this is achieved by bleeding the charge off of C10 through R11 - "release".
The voltage on C10 is used to control the varaible resistor Q1. JFets are normally ON, so 0V on the gate = low resistance = minimum feedback = no compression. When a negative voltage is applied to the gate, the Jfet turns progressively OFF, increasing negative feedback around IC1A and so lowering gain - we have a compressor.
Now here's the bit where I'm confused:
If you were to use regular Si diodes (Vforward say 0.7V), my understanding is that the threshold will go up (needs more voltage before the diodes conduct, charging C10). The threshold can be compensated for by increasing IC1B's gain. Ashcat pointed out that this will lead to a change in ratio - which I think I can see, thinking of the graph I mentioned earlier. More gain means the differences between "just over the threshold" and "way over the threshold" are exaggerated, so higher ratio.
Summing up:
D2/3 - threshold. Higher Vforward = higher threshold.
C10 stores compression control voltage, slows response down. Bigger value = less response (attack, ratio).
R11 bleeds off charge on C10 - release. Lower value = quicker release. Also affects attack and ratio.
IC1B = buffers/amps compressor control, more gain = more compression through higher ratio and lower threshold


As far as I can wrap my head around it, higher diode Vforward means higher threshold, and all other things remaining the same, less compression.
Midwayfair pointed to practical experience that Si diodes compress a lot *more* - so what gives?

(edit:typo)

[samhay]

Cool circuit - have not come across this before.
A quick spice simulation shows that the voltage on the FET (I used a 2N5486) gate pulls lower with BAT54 Schottkys than 1N4148s at D2 and D3 and the signal clamps more quickly - this may work/feel like less compression. Not sure how best to determine the compression ratio with a sim as it clips pretty badly with a signal pulse - those LEDs in the feeback path of the 1st op-amp are there for a reason.
As an aside, an obvious mod would be to vary the value of C10 - lower values decrease both the attack and release times and vice versa.

[midwayfair]

This circuit takes 15 minutes to breadboard, Mike. It would be trivial to build it and measure the voltage at C10/R11/D3. You can also measure the resistance drop across the FET.

Did you take into account that the diode are backwards from what we're used to seeing in a rectifier that creates a positive voltage?

[Govmnt_Lacky]

Do we just have to guess what the film cap values are? C1, C2, C3, and C12.

How were you able to sim the circuit without this info?

[Mark Hammer]

I'll guess 220pf for C2, and .01uf for C1/C3.   And, given the other component values, you can probably get away with 2200-4700pf for C12.

Note that you can play with the values of R8 and R10 to achieve different gains in the rectifier stage.  That may prove helpful for variations in diode forward voltage.

[samhay]

I took the cap values from Harald's vero layout - http://www.sabrotone.com/wp-content/uploads/2011/09/LoveSqueeze.gif
C1 = 33n, C2 = 100p, C3 = 1.5n, C12 = 15n. Slightly funky values, but the frequency response looks about right.
So how does this sounds anyway?

[midwayfair]

So how does this sounds anyway?

It's good, but it's more like a boost with some compression than a compressor. You can't squeeze any more gain out of the first half without creating distortion, and it's possible to overdrive it on the very front of the attack of notes with high output pickups. (Kinda like that pulse you were using in your sim.) So you have two options if you want to make it more of a compressor: hardwire the gain about midway and move the comp knob to the threshold section, or add a third control.

Or you can build something similar with discrete transistors, and a MOSFET booster for stage 1, which is MUCH harder to overdrive and has better noise performance without all the extra caps for filtering.

[samhay]

Jon, thanks for the description - sounds like the typical downfall of the dual op-amp compressor. I posted a slightly different take on this idea yesterday (http://www.diystompboxes.com/smfforum/index.php?topic=101118.0; sorry for the hijack) to make the most of the single-knob compressor. Your bearhug compressor looks pretty cool by the way and is on my list of things to try.

[Mike Burgundy]

This circuit takes 15 minutes to breadboard, Mike. It would be trivial to build it and measure the voltage at C10/R11/D3. You can also measure the resistance drop across the FET.

True enough, although I'm not really able to at this time.  The only room I have to tinker is about 5 degrees C at this time, for one. Everything electronic is in storage, for two. Was just wondering about the theory, and did not expect it to be too difficult a question ;P

Did you take into account that the diode are backwards from what we're used to seeing in a rectifier that creates a positive voltage?

What exactly do you mean? The diodes are back-to-back, the voltage taken is negative, more negative = more feedback = more gain reduction.

[ashcat_lt]

Seems to me like switching to Si diodes without changing anything else increases the threshold. 

Increasing the gain of the detector stage without changing anything else will decrease the threshold but increase the ratio. 

Swapping to Si and then increasing gain to compensate would leave the threshold the same and increase the ratio.

But then I get a bit confused because of the feedback nature of the device.  I mean, the detector turns down the gain of the comp stage, which also turns down the signal at the detector input.  So increasing gain in the detector decreases gain in the comp which then...  Oh heck I don't know anymore!

[slacker]

What happens depends what sort of resistance range the fet is operating over. With a max of about 3.3k in the feedback loop of the first stage the gain drops close to 1 pretty rapidly as the fet's resistance increases. Above a certain point increasing the resistance will essentially do nothing, for example 100k gives a gain of 1.03 and it goes down above that to a gnats above 1 for the max of 1 Meg set by R16.
Therefore, driving the fet's gate more negative beyond a certain point won't give any more gain reduction. It will just mean it takes longer for the gain to go back up.

What's the purpose of D2? Seems like it just clips the positive signal, but D3 blocks that anyway, so I can't see what it's for.

[Mike Burgundy]

That's one of the weirder things, but it does work well - and many compressors use it. Studio D&R compressor, Samhay's new design, heck even a Squeezer attenuates *before* detection, dunnit?. I guess it's less accurate, but it works and may well be very musical in its balancing act.
On D2/3: good point. Bit of a guess here, without D2, isn't it possible for C9 to charge itself up on all those positive signal peaks for D3 never to go into conduction?
(Heck, breadboard may be in storage, I like the mental exercise too!  )

[midwayfair]

True enough, although I'm not really able to at this time.  The only room I have to tinker is about 5 degrees C at this time, for one. Everything electronic is in storage, for two.

Are you in the US, and do you have access to your multimeter and a test rig? I could mail you my Bearhug socketted-to-hell test PCB (it uses the same rectifier) and some diodes, and you could mess around with it. I would need it back after you were done, though, because I have to pull it out when people ask me whether a certain substitution will work.

I also have a working Lovesqueeze clone that I could pull the diodes on and replace with sockets. Might be a little easier to deal with than the PCB by itself.

[jonasx26]

What's the purpose of D2?

It adds a DC offset to the AC signal. Google "DC restorer".

[slacker]

Cheers, just looked it up, never come across that before.

[Mike Burgundy]

@Midway: thanks a lot! Only I'm in the Netherlands, and the problem of workspace at very uncomfortable temperatures still holds - I'll build it anyway someday. One can never have too many compressors ;P Temperatures will rise, and the reasons for me to be unable to work at home will disippate wirth time too. When I do build it, I'll let you know, and what I find. For now, I've been simming it last night, and at least the sim does seem to confirm my suspicions: @certain settings/input: BAT54 yields about 1:4 compression, -1.9V on gate, roughly 50mV output. 4148 @ same settings/input yields 1:2 compression, -1.8V on gate, and roughly 100mV.
so, less compression, and more output for Si diodes.
Could it be that you've been running into the LED limiter? That'll compress the heck out of it, once you've reached that point.

@Jonas: thanks, that's it. Confirmed by the sim: it pulls the signal between C8 and D2/3 negative, so that peaks are one diode drop above 0. Sim also showed that without the restorer the compressor will initially work, but after a very short while it will charge up C8 beyond conduction possibilities for D3. So it will only compress the first transient, and leaves the rest alone ;P

[samhay]

Mike - out of curiosity, what have you used as an input signal for your simulation?

[Mike Burgundy]

all kinds, but this one was a  2kHz 100 cycle burst @ 200mV. Regular voltage source, no PU sim. Gave the circuit 1000ms to settle DC, then added the burst and looked for 100ms. You can see the attack and release nicely.
This circuit is quite a balancing act - especially the FET.

[samhay]

Thanks Mike. I Just had a play with my spice sim using your input pulse, which seemed to work quite nicely. I was measuring V(out) at the end on the 100th cycle. I haven't really looked at compressors in this detail before - is that a reasonable thing to do, or should I be looking at the initial spike (1st cycle) and/or their ratio?

With my crude analysis, it looks to me like it is acting as a limiter with a soft knee with the compression knob at 1k+. Any input signal above about 0.3V will cause the LEDs to start clipping. I used BAS54's, but can't imagine it will be a whole lot different with signal diodes. 
The 0.3V ish threshold seems to be a bit low to me if it is acting like a limiter. It might place nice with quite low compression settings - is a log pot recomended? - but either way, a bit of signal reduction (pot as voltage divider) before the 1st op-amp might help quite a lot.

[Mike Burgundy]

Maybe, I'll have to play a bit more.
If you plot output and gate voltage together, you get a good view of what's happening. After the start of the pulse, the gate is pulled down rather quickly in about 5ms - that's your attack time. Following that, gain on the opamp, and output are reduced - thats compression. The initial attack is still at 1:1 level, after that gain is reduced, which gives you a nice estimate of the ratio. When the pulse ends, there is no output (duh), but you can see the gate recover from it's negative voltage in about .6s. That's your release. It's a bit long to my liking, but I'll have to check a real-life build first. I think it's actually a lot quicker - the JFet is ON enough to effectively stop compression before the gate reaches 0V.