thoughts on this limiter?

[Jim Zuehsow]

I threw one of these together back when the article came out. Nothing special.. it certainly didn't make me get rid of my 1176's. The question is, why would I want to build one of these today, when THAT has such great chips that really make building a great limiter so easy, unless of course, you just want to have a FET limiter. Just not worth the time and effort, plus, been there, done that.

 

[recnsci]

[quote author="Wavebourn"]Smart dances around a 50/50 voltage divider from drain to gate? [/quote]

Drive FET differentialy. Like, take two identical voltage dividers; put FET
between lower resistors of voltage dividers; drive dividers with transformer;
use instrumentation amp configuration to amplify voltage across FET. This
will also noticeably reduce CV bleedtrough.

cheerz
ypow

 

[jdiamantis]

[quote author="NewYorkDave"]The use of the current mirror to charge the cap is a little unusual. It seems like a sensible and straightforward design overall. The parts are cheap--try it![/quote]

Used as shown you don't get all of the benefits of using a current mirror or current charge pump to charge the cap. The benefit here is a linear attack time vs voltage offered by the circuit. If you use a multi-time constant circuit with a current mirror, the actual gain of the side chain is program-dependantly variable, giving you a program dependant attack, release, and ratio. If you use a current sink for release, recovery time is linear as well (looks great on paper; doesn't sound a natural as the anti-log release provided by a simple resistor)

The math is a bit complex, but this scheme is used alot in older broadcast processors; including the first Optimod FM processor (used FET's as gain reduction elements), and later iterations of the Optomod and other company's boxes.

 

[mr coffee]

If you use a current sink for release, recovery time is linear as well (looks great on paper; doesn't sound a natural as the anti-log release provided by a simple resistor)

Is it possible to combine the two approaches to get a faster initial recovery without the long tail of the simple resistor approach (using a current sink)?

 

[JohnRoberts]

[quote author="mr coffee"]

If you use a current sink for release, recovery time is linear as well (looks great on paper; doesn't sound a natural as the anti-log release provided by a simple resistor)

Is it possible to combine the two approaches to get a faster initial recovery without the long tail of the simple resistor approach (using a current sink)?[/quote]

I started to write one of my typical just wrap a this around a that and you get.... but realized this is not trivial stuff so I dug up a schematic of an old companding tape noise reduction (kit) I sold back in the '80s. This was IMO better than the more popular well known tape NRs but right around then DBX started making their NRs in japan and their assembled units were cheaper than my kits, guess what happened to my kit sales?

OK, about my circuit http://circularscience.com/522_NR.pdf

First a little explanation about the NE572. For those not familiar this was nice chip with a pretty accurate full wave rectifier built in, and a decent gain cell. Not VCA quality, but a step up from 3080 or similar transconductance amps. Further the 572 is designed to dump it's FW rectified audio into a cap for smoothing and using a mostly linear control voltage from that cap to control the gain cell. The 572 had two FW rectified outputs designed so one could be fed into a small cap and the other into a larger cap. The small cap output could charge the slow cap so you would essentially get the attack of the small cap and release of the larger cap. An improvement, but I took this a few steps further.

By connecting what was supposed to be the small cap port to the - input of an inverting opamp I effectively clamped it to ground disabling it while extracting it's reasonably accurate FW rectified signal. I convert this FW current to a voltage and then cap couple it back into the top of the slow capacitor. Since the FW signal being drawn out is opposite polarity to the identical current feeding into the cap from the second FW output there is a first order cancellation of the AC component charging the cap with longer time constant (sort of a cross between a cap multiplier and a ripple filter). But now by careful sizing of power supply swing and resistor values you can establish a very fast attack/release mode for large transient overloads. If a large transient comes through and clips the opamp against the - supply it can no longer match the current coming into the top of the cap and it will charge up rapidly. Also the base of Q2 will hold that input clamped close to ground so it doesn't trigger fast attack circuit inside 572 (Q2 also drives a peak indicator advising you to back off recording levels.

In addition to these two charging time constants, C15 and Q1 provides another faster attack time constant for transients. While it looks a little unconventional, by being cap coupled it mitigated against tracking errors that can occur in companding tape NRs when encode/decode amplitude and phase response are less than reliable (think cassette decks).

This circuit obviously has many attributes driven by the dubious bandwidth tape path (note reduced bandpass of side chain) but the general concept has utility in control voltage modulation management in other processors. I offer this as food for thought rather than a canned solution. I no longer have any schematics handy for compressors I did using variants on this technique.

JR

 

[mr coffee]

Hi JohnRoberts,
Whew! I am studying your NE572 compressor design but it will take me a few hours or a day to take in exactly how you have applied the rectifier part of the chip. Sounds very intriguing. Even with your explanation I'm still having a bit of trouble understanding that part of the schematic!

I've used the '572 in a more app note incarnation as a noise reduction path for analog BBD delay chips (like the good ol' SAD-1024A, which still has it's aficionados). and which, as you note, although it is not VCA quality, it IS head and shoulders above the 570\571 era compander chips and the CA3080 (I guess the LM3080 is the only one still made nowadays).

Thanks very much for sharing this schematic. I'm afraid the C15 to ground in particular and the C14 from the FWR op amp output just haven't computed yet. I got that it has to do with ripple canceling, but it just hasn't sunk in yet.

Thanks again for the stimulating discussion!



:thumb:

 

[JohnRoberts]

If you're familiar with the 572 series that helps a lot.

C14 is primary path for canceling the AC component at C13, the effective control voltage port.

C15 is there to provide short term charge for quicker attack if the level change exceeds the threshold for Q1 to start conducting. First C15 dumps into C14 which swings negative due to large signal increase. Then when the transient passes and bottom of C14 returns to near ground, the top of C14 dumps charge into the smaller C13.

An important thing to notice is that C14 and C15 being cap coupled only affect dynamic or short term modulations of the control voltage. Long term it all averages out and the only DC path to define control voltage is coming from the 572 FW rectifier, so tracking accuracy will be good despite lousy frequency/phase response of typical tape recordings.

There are more tricks in that design I didn't even mention as they are specific to tape NR. I have also done my share of companding NR around BBD and CCD analog delay and they have their own requirements.

One perhaps amusing side note, when I upgraded the companding NR in a studio delay line/flanger from 3080 based to the measurably cleaner 572, I actually got comments that some customers preferred the sound of the old 3080 version. It's hard to argue for lower distortion in a BBD flanger, but that technology got trumped anyhow by early digital with it's own new flavors of distortion.

JR

 

[mr coffee]

JohnRoberts,
Thanks for the further explanation. Very clever design. Still studying your design to try and understand the full implications of the ripple canceling scheme.

I actually got comments that some customers preferred the sound of the old 3080 version.

"They don't make 'em like they used to" "The original version sounded better" <LOL>

The song remains the same, eh?

Still haven't quite got the transistor discharge circuit through my head ... current sink circuit without a DC path just hasn't quite registered. I'll study it some more.

Thanks again!

 

[PRR]

> long channel vs. short channel JFETs ... I don't understand enough semi- physics to make sense of that

It's just end-effects.

Compute a duct for a bass-reflex speaker. The effective length is longer than the hunk of tube you cut. A 1" diameter 100" long tube does act like it is about 100" inches long, but a 4" diameter 4" long tube acts like it is 6 inches long. A hole in sheet-metal has finite acoustic length. Conceptually, a hemi-sphere of air at each end "thinks it is in the tube".

An FET has a no-bias channel as big as your mask lines, and a biased channel which is much-much narrower, conceptually one atom wide.

If you build FETs in the kitchen sink: a 10mm long channel 1mm wide will be fairly linear, even when bias pinches-off the channel to one atom wide. Slenderness ratio varies from 10:1 to 1,00000000...:1, which is all "large". But a 2mm long 2mm wide channel pinched to 1 atom wide varies 1:1 to 1,00000000...:1, from short to long. Production FET sizes are microns not MMs, but always large by atomic scale. A long-channel is always long; a short-channel must vary in slenderness.

The short-channel gives more gain at high current and modest voltage, or for fixed die area. This makes it preferred for RF and CPU application. Same as 6L6 and 6DJ8 are very non-linear tubes, with large change of gain at high current, which makes them loverly in RF and power application; 6J7 and WE-300B are low gain but nearly the same gain at all practical current and preferred for simple clean audio. (And the cathode-grid "channel" has the same slenderness issues as a sand-state FET, just more mechanical dimensions to fiddle with.)


Leach limiter: Aside from other frills: a specific feature is that it will settle on the EXACT gain setting. As a servo loop, it has near infinite gain and near zero error. It takes time to settle to very low error, but it gets to low error fairly fast. It is an asymmetric integrator, with ideally infinite control-path gain at DC. You servo-guys will know more than I do about the terminology and implications.

I don't see this feature as useful for music. We don't need an exact level, we don't have a steady-state signal. And many of our "favorite" limiters are seriously "flawed". Really good limiting is really boring. A bit of zing and clip and error ear-fakes the dynamic sparkle we are squashing out.

It is a perfectly good simple limiter. It is an excellent tool for teaching you to THINK about the complete limiting problem (not just P- versus N- niggles). But if you need a limiter, some of the $99-$199 boxes will do the job near-perfect. If you need more or less than near-perfection, you have a long road ahead.

 

[JohnRoberts]

[quote author="mr coffee"]JohnRoberts,
Thanks for the further explanation. Very clever design. Still studying your design to try and understand the full implications of the ripple canceling scheme.

I actually got comments that some customers preferred the sound of the old 3080 version.

"They don't make 'em like they used to" "The original version sounded better" <LOL>

The song remains the same, eh?

Still haven't quite got the transistor discharge circuit through my head ... current sink circuit without a DC path just hasn't quite registered. I'll study it some more.

Thanks again![/quote]

Perhaps my explanation wasn't as clear as it could have been. There is obviously a dc path between all three capacitors but there is no DC path from there to ground or elsewhere other than to and from the 572 rectifier/gain control port.

Any fast attack or release tricks that add or remove external charge from that circuit could cause a long term gain error. While clearly more important for the encode/decode application, this reduces mistracking due to level or phase shift errors between the two processes. For example in the stock 572 circuit if phase shift alters the peak to average ratio of the music coming back from tape that would cause a significant mistracking error. This was one of the benefits claimed for DBX NR's use of RMS detection.

The transistor specifically looks like a zener diode in one direction but mostly high impedance with a parallel resistor in the other. So independent of all the transient attack speedups the control voltage always ends up with the same area under the curve as for the simple one capacitor circuit. It's like the fast attack gain reduction is "borrowed" from the average gain and paid back later so it always averages out the same.

JR