Official Forssell Limiter Build thread

[rafafredd]

A shame this project seems to gain no interest.

I do have some old pair of red boards here that I have lost track of. I mean, I can't find the suitable schematics to have those boards populated. Do you know if thses schems could match those boards? I would be interested in making these into functional units, after all.

 

[sonicwarrior]

Do you know if thses schems could match those boards?

I guess so. Of course I have only included the DOA schematics. The NE5534 output buffer schematic is attached.

 

[jensenmann]

Rafa, I have a bunch of schematics. Drop me you email via pm

 

[insideout]

Do you know if thses schems could match those boards?

they are on Fred's site eg http://www.forsselltech.com/media/attachments/SideChain_B.PDF
http://www.forsselltech.com/#!/articles/9

 

[kato]

Also, does anyone have an overlay for the PCB they could share.
I've seen the one Frank posted but I could use the one with parts #'s for REV 1.C.

I uploaded the overlay for rev. 1.c here:
https://docs.google.com/open?id=0BxbGM-GHrmdJOGZhc2VOWFg5aGc

 

[nhaudio]

Anybody has a parts list for rev 1.c ? All the links in the meta are dead.

 

[mitsos]

Not sure about the version, this is marked Rev 3, maybe Rev3=1.C ???

 

[nhaudio]

thanks will check it out

 

[kato]

I've also skipped the input DOA on the board I was having issues with a while back (I fixed the issue but wanted to experiment after reading kevs posts), and tried a cinemag CMLI-15/15B feeding the 4.99k.
Compared to the other board(both using same DOA's and components) the cinemag loaded one sounds so much better.

Thanks for this LowFreq. After reading about Kevin's call to Fred about ditching the input buffer entirely, I was thinking input transformer so as not to suffer the full unbalanced db drop. I missed your post the first time around - probably cuz I wasn't considering adding an extra transformer to the project. I'll give it a go and report back. Thanks!

 

[kato]

Perhaps this is a dumb question, but I'm really not certain even after giving it quite a bit of thought.  So here goes:

Does the audio actually go through the sidechain?
I understand it literally 'goes through' for adjusting the compression settings, but does that same audio 'make it' to the output?

The block diagram tells me no (? I think?)
But the $5 Burr-Brown ICs in the sidechain says 'probably yes.'

 

[shabtek]

i think the BB allows 24v operation if one were opting for 990's in the signal path, otherwise a cheaper pin-compatible will do

 

[Harpo]

Does the audio actually go through the sidechain?
I understand it literally 'goes through' for adjusting the compression settings, but does that same audio 'make it' to the output?

No. Sidechain only makes the LEDs in the optos shine up. If they shine up, the value of the resistive part in the opto cell goes down.

The block diagram tells me no (? I think?)
But the $5 Burr-Brown ICs in the sidechain says 'probably yes.'

Correct thinking, but the block diagram (with input/output reversed output stage) doesn't say, why a more beefy opamp is needed to charge the 10uF cap in parallel to the release pot and to deliver enough current for the opto cell LEDs to show up. With lower supply rail voltages a NE5532 seems sufficient and a TL072 will not do it.

 

[kato]

Thanks friends, I appreciate the advice.

I see earlier Clintrubber suggests upping the value of R24 and R25 in the sidechain to allow the use of TL072. Apparently this makes it easier for any IC, including the OPA2604, although I don't yet understand why.

You just saved me $12, between U9, U5, and U6. I can't see any advantages to running this at 24v. I'll try the NE5534, possibly changing R24 and R25 in the process.  Thanks again.

 

[etheory]

Apparently this makes it easier for any IC, including the OPA2604, although I don't yet understand why.

Upping the values of those resistors reduces the output drive requirements for the op-amp.
The lower the value, the lower the output impedance the op-amp will have to drive, the more duress it is placed under, the more it is likely to complain.
Increasing the feedback resistor also increases noise, but, to be honest, very marginally in this context, and in a side-chain application, which is already insanely filtered (a side-chain is simply a low-pass filter anyway), it won't be the slightest of issues.
The original spec from Forssell is most likely just overkill.
He is a purist, which is awesome, but in this case, there may be no point in such a high specification for what is really just a dual op-amp full-wave signal rectifier.
Try replacing those resistor values with something like 10K, which will work with almost any op-amp, and you'll probably be just as happy, and have spare change for an extra coffee you can use for debugging anything else that goes wrong.

 

[etheory]

What are people's opinions on the sound of this compressor?

I recently attempted to model this as a digital VST plug-in (directly from the circuit diagram), modelling the exact side-chain and optical element details (I used a NSL-32SR3 with a model I found here: http://www.lynx.net/~jc/NSL32-SR3modeling.html) but the result was quite "jumpy" sounding unless attack and release were basically set to zero.

The thing I've noticed is that for any release or attack that is slightly "long" (longer than the duration of a single beat), the compressor seems to violently "jump" to compress. The result doesn't seem particularly smooth or warm, more "crazy" and "extreme", even at a fairly high threshold. This seems to be a side-effect of the LED diode turn-on for the optical element being quite "instant" vs gradual, producing a pretty hard-knee.

I really need to get a hold of a VTL5C1 and measure it's properties and model that for comparison.

Did anyone build this with a NSL-32SR3 and have any comments on what they thought?

 

[ruairioflaherty]

I don't have any direct experience with the Forssell opto compressor, even though I do have a pair of boards sitting for years.  I did however own a Millennia Media Origin channel strip, designed by Fred and featuring an opto compressor.  I never looked at the circuit (I did with the NSEQ) but I'd guess  that it is based on the Opto design he has shared.

That particular compressor behaved exactly as you described, it was jumpy and grabby.  Longer attack times only seemed to cause a delay before the compressor grabbed the signal hard.  It was not very useable,  I found that mixing it in parallel was the only real way to get a good result.

Please start another thread when you have time about modeling the circuit in a VST, that would be interesting for many people here I imagine.

Cheers,
Ruairi

 

[Henke]

In my own vactrol-based compressor designs i make sure to offset the control voltage to make the LED stay slightly "on" (idling at 0.5-1dB gain reduction) to get a predictable behaviour

 

[etheory]

Good points.

Thanks ruairioflaherty for at least theoretically validating what I was hearing.

The two things I'll note are:

1.) It would seem that to add a ratio control, all that would be needed is variable gain to the opamp driving the 500R resistor going to the LED. This would also be the point to add bias (if required) as suggested by Henke.

2.) I think that both the attack and release are WAAAAAAAAAAYYYYYYYYYY too long. Like chasms and light-years too long. On paper the release is a max of 5.5 seconds ((500K + 50K)(550e3) * 10uF(10e-6)), and Attack of 0.165 seconds (500K * 0.33uF), yet in a practical circuit, that time is significantly extended by the dynamic properties of the circuit being charged continuously by a varying input. If I set the attack and decay both to zero (i.e. about 0.5 second release and 0.000033 seconds attack (assuming you current-limit attack using a 100R resistor to avoid overloading the attack-buffer op-amp, or in the real circuit the capacitor ESR is higher than this)), the circuit seems to produce a fairly pleasant but flat-line crushed sound (probably good for vocals, but not much else).

So I think I need to do some work to tune my VST simulation, but I feel this will also help inform some changes to the real thing, which I have the parts to build also, and probably should for reference, and to measure against.

As for a tutorial about modelling a VST based on a real circuit, my suggestion is to read the QUCS user documentation, and some of the technical details here: http://qucs.sourceforge.net/, specifically: http://qucs.sourceforge.net/tech/node16.html. I'm currently using a customized variation of Newton-Iteration/Fixed-point iteration and some other secret-sauce, beginning with an MNA nodal-analysis and s-domain frequency analysis for a 3-variable network (which is enough to describe the full circuit including the feedback-loop) in real-time without approximation. What I mean by "without approximation" is that I'm doing the full solve of all components, with as much detail as a spice simulator would. I'm not substituting models for simpler ones. This involves a real-time iterative solve of a 3 diode network (the two diodes in the side-chain + the LED diode), requiring 4 exponential evaluations per iteration (2 diodes + 1 led + exponential response of the optical element resistance). The details of how to do this fast enough for real-time use are trade-secrets for the meantime. Once I finish my commercial plug-ins, I'll be releasing some technical papers about how I do it. There are some things I've discovered that are new, and would constitute new technology for real-time simulation of circuits with regard to high-precision and low CPU usage.

 

[DrFrankencopter]

Not sure a Spice model is needed for a plugin. I'd think a few transfer functions, and a little non-linear modelling of the Vactrol should work quite nicely.

Use the curves available on the data sheet: http://pdf.datasheetcatalog.com/datasheet/perkinelmer/VT500.pdf

You'll see how resistance is a function of drive current and time. If I were modelling it I'd just do some interpolation between the 10ma and 40 ma curves (maybe saturate to those values for currents out of range on the time graph).

Cheers

Kris

PS: Been a long time since I've been here. Hoping to blow the dust off my Forssell limiter and finish it before the end of September. A mere 10 years after I started it 

 

[etheory]

Not sure a Spice model is needed for a plugin. I'd think a few transfer functions, and a little non-linear modelling of the Vactrol should work quite nicely.

This is what most plug-in developers do, and it really annoys me.
It's simply not enough.

To get the proper operation, you need to model the rectifier, since that is a part of the threshold (since you lose one diode drop at that point), along with the proper release and attack circuits, then the LED for the opto needs to be properly modelled, since it's the second part of the threshold (another diode drop) and is highly non-linear (it follows a Lambert W curve). Then you need to model the optical element properly (which I've done by measuring the real thing) to get the right gain response.

My VST models the entire circuit, and it sounds a lot better for it. I've used far too many plug-ins with a "close-enough is good enough" approach, and I just don't see the point, they simply don't sound as good. This is also one of the main reason why I've started writing my own plug-ins. The commercially available ones in most cases just aren't very good, or they use "tricks" that don't sound very good to implement things like feedback, when, with a little bit of work and more complex math, you can solve the equations properly and get a truly plausible result.

YMMV.