Calculating Attack & Release Times..... help, I give up:(

[PRR]

> there is no direct discharge path for the cap other than the gate of the FET, I'm assuming the existing values probably result in a very long release time.
> no idea how to model release time

"model release time".... if you had a simple circuit, you would use a SPICE pulse (or switch) to apply a voltage, then wait (forever, or until math-error accumulates) for it to decay. And a perfect diode to a cap, NO other path, would take forever to decay (in real life, it may actualy drift "UP").

HOWever... in your plan, if you assume a voltage on C62, where can it go? Through R86 to R55 to a bunch of voltages "following the signal". As abbey road d enfer says more concisely.

I would have real doubt about any sidechain numbered up to C62. Rule-of-thumb: when I take off my socks to keep track of what-does-what, it is too complicated for me, and perhaps too complicated to be doing what I want to do.

I would have strong doubt about any scheme with "no ...discharge path for the cap". If true, the cap can never decay. In a real world, stray leakage will drift the cap voltage to ANY voltage, including unhappy zones. Now, if there is a RESET button, my doubt is weaker; still I'd think about something to drain leakage, how large that leakage could be, why I would want to hold a voltage for very long time, if this may be a chore for a PIC/STAMP (digital can hold until the lights go out).

 

[tv]

The "k.i.s.s." way in ltspice.

(or use the PULSE and setup the Tdelay, Ton and Tperiod parameter.)

 

[JohnRoberts]

Or you could build it up and see how it acts.. Abbey nailed it, the discharge path is back through the same resistor that defines the attack in series with the precision rectifier's feedback resistor. The far end of that 4.7K follows the input voltage  during discharge cycles for a roughly 5:1 ratio of release to attack.

Warning esoterica alert- For a circuit like that where there is no need for the side chain to be low impedance (10uF) I'd be temped to scale the time constant components to use a film cap (say .1 or .22 uF polyester).

This is a rather subtle point, but in circuits like this where the side chain capacitor is charged with one impedance and discharged with another, dielectric absorption in the cap can actually make a difference in the networks response to level changes. I repeat this is very subtle and won't sound bad (or good), just ever so slightly different than an ideal capacitor.

I once intentionally used a cap dielectric with known high DA (tantalum) in the side chain attack/release circuit for the previously mentioned CX decoder, because the encoder circuit used tantalum, and I wanted to track as accurately as I could. This was clearly over-engineering, but rearranging real deck chairs, not imaginary ones like many audio phools. 

I repeat, I don't expect this to make an audible difference significant to this circuit's behavior, but I mention it for instructive purposes, since DA is widely attacked as a source of audio evils. Here is a circuit where it actually does make a difference, however modest.   

JR

Note: DA is modelled as multiple smaller caps with Rs in series all in parallel, inside the composite capacitor. This suggests that a high DA cap will have a more complex att/rel curve in the limit, while it is still dominated by the gross R and C values.

 

[ruckus328]

Guys, thanks.

-Regarding the schematic TV referenced - yea that does look familiar doesn't it.  Little different but not by much.  I see they have paralleled a resistor to the cap as I want to do.  I'm going to move the distortion null circuit post buffer as recommended, seems like a no-brainer.

Abbey - I wasn't thinking "backwards" in the signal path when I was trying to analyze it, oops.  That being said, if I do parallel a resistor to the cap I would not want this additional discharge path back at the opamp that currently exists, so simply blocking it via a diode will do the trick, right?  Seems that's how it's done in the buss comp at the beginning of this thread as well as in the schematic TV linked.

One thing that I could never wrap my head around (I'm sure I'll get properly schooled on this one) - What's with the funky full wave rectifier in the SSL circuit?  Not typically how I'm used to seeing it done.  Why the second set of diodes?  Couldn't it just be simplified as it's done in the schematic TV linked?

John - Regarding the film verses tantalum, interesting.  I have contemplated changing the .47uF tant to a film in the buss comp before, never gave it a try before though, maybe I should.

 

[JohnRoberts]

Guys, thanks.


John - Regarding the film verses tantalum, interesting.  I have contemplated changing the .47uF tant to a film in the buss comp before, never gave it a try before though, maybe I should.

I expect this to be a very subtle difference and perhaps less than the cap tolerance.

JR

 

[tv]

One thing that I could never wrap my head around (I'm sure I'll get properly schooled on this one) - What's with the funky full wave rectifier in the SSL circuit?  Not typically how I'm used to seeing it done.  Why the second set of diodes?  Couldn't it just be simplified as it's done in the schematic TV linked?

My guts are pretty sure that each type of the rectifier will result in a slightly different "sound character" when actually used.

The opamp gain-stage will give you a different character compared to a discrete gain-stage. IMHO the real question you should ask yourself is whether you want a mojo comp/lim or a clinical fast FET jobbie. The clinical vibe will imho call for a precision-rectifier circuit, while a mojo approach will call for a rectifier that you will have to fine-tune for "flavor" (which is a very time-consuming thing).

Perhaps look into one-BJT phase splitter similar to shure level-loc (Q5-7).
http://www.freeinfosociety.com/electronics/schematics/audio/shurem62v.pdf

 

[JohnRoberts]

There are several variant precision rectifiers. While most should be very similar, some of the cheaper, reduced parts versions may differ from ideal in some way. I wouldn't expect huge differences between variants, of similar function (like between all peak detectors). Some precision rectifiers are average detecting, so that will vary in behavior. While att/rel of peak detectors can and will make a difference, likewise full wave vs, half wave rectification will differ..

JR

 

[MikeClev]

ESP project 67 (http://sound.westhost.com/project67.htm) is using the diode drop to set the threshold for the onset of limiting... do I have that right?

ESP has a brilliant article on precision rectifiers here: http://sound.westhost.com/appnotes/an001.htm

Just by coincidence, I recently started to modify that design for a more fully featured circuit with a threshold pot, attack and release pots and a makeup gain stage and pot, to build and test on vero board. Wish me luck!

 

[JohnRoberts]

ESP project 67 (http://sound.westhost.com/project67.htm) is using the diode drop to set the threshold for the onset of limiting... do I have that right?

No not exactly.. if you are talking about figure 2.. the threshold is the pinch off voltage of the JFET, plus the diode drop of the rectifier diodes. Vp can vary so a design like this in production will limit at a range of levels depending on the JFET used. . 

ESP has a brilliant article on precision rectifiers here: http://sound.westhost.com/appnotes/an001.htm

[TMI-warning]

Ok, buckle up for the rest of the story...  While that over view covers a lot of ground most of them are variants of each other and share similar weaknesses. Precision implies accuracy which for small signals can be trashed by cheap opamps with large dc offsets. 

Another consideration is not unlike zero crossing distortion, where all those commutation diodes have to swing from conducing in one direction, to conducing in the opposite direction, in zero time at the instant of zero crossing.  Reducing this to terms the opamp has to deal with, the output of the opamp has to swing +/- a few hundred milliVolts in response to some physical input voltage change. In fact this output voltage change divided by the opamp's open loop gain predicts how much of an error voltage has to be overcome at the input. In practice since opamp open loop gain falls off with increasing frequency, this input error looks like a low pass filter to very small signals.

There are strategies to mitigate this, involving using extra parts and circuit tricks.

My old TS-1 design used all the tricks I was willing and able to put into that design (30+) years ago, and my precision rectifier, in front of my dB convertor was still down -3 dB at 20 kHz , at -50 dBu.  Trust me most of your library of precision rectifiers will not get close to that (perhaps all). Some fair rectifiers were done inside NR chips like NE570 (I actually used a NE572 just for the rectifiers in an old compressor I did way back when), more recently the rectifier in the THAT log detector does not suck.

But YAWN so what.. for general purpose side chain use the frequency response of your rectifier at -40dBu and below is not a huge concern (for a measurement instrument more so). Therefore I only mildly object to calling them "precision", they are better than just a raw diode in series with a hold cap. I guess precision is code for compensating for the diode drop. So what do you call more precise rectifiers? 

For extra credit look at my old TS1 schematic that is out on the internet somewhere.. while some of the parts I used in my old rectifier are obsolete now (the transistor array).  [/TMI]

Just by coincidence, I recently started to modify that design for a more fully featured circuit with a threshold pot, attack and release pots and a makeup gain stage and pot, to build and test on vero board. Wish me luck!

yup Good luck...

JR

 

[MikeClev]

Thanks for the explanation John, it is much appreciated as always!

As for precision, that's ESP's description for it not mine. On reading that article I kind of figured that if there were that many different ways to make a 'Precision Rectifier' there must be something pretty imperfect about all of them!

 

[JohnRoberts]

I am guilty of not wanting to copy other engineer's work. In fact when I was managing an engineering group, part of that management work was to prevent my engineers from reinventing old wheels. There is no reason why a company like Peavey should have a half dozen different precision rectifiers in different products doing the same thing. (Precision is a general term for them,, not Mr Elliot's personal choice.) There are many ways to skin that cat, but not a sh__t's difference between most of them.

I even used another engineer's topology for a mic preamp in a console I did, after i determined that my version wasn't really better (and cost $0.25 more). See.. I can be a team playa... ,

JR

 

[PRR]

For audio speech/music limiters you don't want a Precision Rectifier, you want a Peak Catcher.

 

[PRR]

> just a raw diode in series with a hold cap.

If "threshold level" at the diode is say 3V, there's nothing wrong with a naked diode in studio environments. The base 0.6V offset is trimmed in design or prototype. +/-0.1V "error" with supplier or temperature is half-db variance.

 

[tv]

For audio speech/music limiters you don't want a Precision Rectifier, you want a Peak Catcher.

a precision rectifier is going to catch them peaks with greater vigor and zeal..

 

[JohnRoberts]

For audio speech/music limiters you don't want a Precision Rectifier, you want a Peak Catcher.

a precision rectifier is going to catch them peaks with greater vigor and zeal..

Not to quibble but precision rectifiers can be peak, or average, or whatever detecting...

So vigor and zeal are in the details of the design.

JR

 

[PRR]

> Not to quibble but precision rectifiers

Precision rectifiers generally fold one polarity over the other to give the full wave in unipolar form.

They also generally have low to medium output impedance, not directly suited to peak-catching on a cap.

By the time you add additional diode(s) for the peak-catching, vigor and zeal (budget and space) are fading.

And since, for _limiting_, we only need -10dB to +10dB of threshold, the "problem" of diode drop is easily solved with the other kind of vigor and zeal: simple Brute Force. Slam 1V-10V at the diode. The ~~0.6V offset "error" calibrates-out at 3V nominal, offsets and softens the knee at 1V very slightly, is quite moot if you want to follow +10V over-peaks.

This is of course very different from metering, or wiiide-range compression, where brute-force might lead to 30V or 100V peak signal. Then some vigor/zeal trickery avoids awkward brute-force. (And very-very different than tracking companders, where some error may be accepted IF the comp and exp can be precisely complementary.)

-----------------------

> ESP project 67 ...is using the diode drop to set the threshold for the onset of limiting... do I have that right?

Not really. What JR said except he missed R4 R6. These stand the FET on +5V. Then add the FET offset, -2V to -8V, and the diode drops 0.5V.

Hmmmmm... 2N5459 has high and wide-spread of Vgs(off). It appears that a highest-Vgs 2N5459 could be somewhat "on" even when sidechain output is zero. Elliot's postings are great examples for study, but some are not fully specified for blind assembly. In this case, R4 R6 could be tweaked to match Q1 offset, then a bit more to get a hard action. That means the OUT level depends on FET Vgs, which suggests separate paths for program and sidechain. Or, and I think this is Elliot's background: adjust the rest of the system (preamps, mixer, power amp) ad-hoc until it sounds good.

Note that any fog around diode-drop has become insignificant compared to JFET spread.

While you can find a JFET with Vgs spec 0.5V-1.5V, making this uncertainly "small", that's not a great answer. JFET attenuator distortion goes by ratio of signal level to Vgs. Signal level smallness is limited by resistor hiss. So you generally want a high Vgs FET, which is why the '59 was specified.

Also be aware of the R8:R7 R11:R13 balance to fiddle FET distortion.

 

[JohnRoberts]

I wasted too many years trying to perfect encode/decode NR systems but I won't waste your time too with how complex these side chains could be, a relatively simple technique to finesse the Vgs spreead and get a modest repeatable threshold voltage is use something simple like a base emitter junction of a transistor as the threshold. The transistor will have gain that can easily ignore the JFET variations.

Simply put, if the output is more than a diode drop from ground, it turns on the transistor that turns on the JFET... Of course I expect you to do the work, so you figure out how to connect all the dots, but i've seen plenty of dirt simple bass compressors executed this way.

I have also made simple full wave detecting circuits using two transistors, (one common emitter and the other common base, while only one polarity of signal swing has current gain.

You should be able to find a one transistor circuit without wearing out your google button. I expect the two transistor approach is less common. 

Of course I may have missed a resistor or two...  8)  I'm just the broad strokes guy...

JR

 

[PRR]

> full wave detecting circuits using two transistors

This is still my favorite. One diode, one transistor, full wave.

 

[JohnRoberts]

The two transistor approach I was talking about could be made by replacing the top left diode with the base junction of a PNP, that would not only charge the cap on down swings but provide current gain on the pull up..  Of course the circuit as drawn works fine and is better matched between polarities. Of course my way the PNP that replaces the diode also replaces the emitter follower. 

Different strokes... there are always more ways to skin the cat, than cats who need skinning.

JR

 

[ruckus328]

Hey guys, thanks for all the input, greatly appreciated.  Been a very busy couple weeks so haven't had any time to put into this but on vacation now so trying to get this design wrapped up this week.  I've had a working prototype for a couple months but had to park it with everything else going on, so hopefully not too much tweaking to do to get it up to par.  Have a couple things was wondering if anyone can give thier 2 cents:

1) Regarding changing the attack/release of original talkback (and referring to the schematic I posted on Page 1/reply #10), going to scale down the cap (C62) to about .47uF and parallel resistor vales same as SSL buss comp and tweak from there.  However, I just wanted a to verify that I should add an additional diode between D10 and R86, to prevent the cap discharge path back at the opamp, otherwise adding any resistors in parallel with the cap (C62) isn't going to have the desired results, right?

2) Regarding ratio's, (again referring to my schematic on reply #10), will simply increasing R85 (or decreasing R9) lower the ratio?  But then doing this (as I understand it) will also greatly adjust the threshold point.  ie: less feedback will mean much higher input signal required for the fet to start opening, so lower ratio would mean much higher threshold point and vice-a-versa.  Because of that don't know if adjustable ratios really is practical with the existing circuit........

3) I had spent may days trying to devise a circuit to accurately monitor gain reduction, as there's no G/R metering on the original.  It was really shots in the dark, but I did end up figuring it out.  As my understanding was the FET is not linear, simply measuring the CV to the gate of the FET won't get me accurate readings, so I ended up basically doing as an 1176 and duplicating the audio FET circuit, feeding it some DC, then meauring the voltage across the fet, sending to a diff amp, scaling it etc etc, and got metering accurate within a db. 

However, it is rather comlicated doing it this way and involves matched FET's, and noticed in the schematic TV linked that it is just intending to send the CV directly to a metering circuit.  Wondering if maybe I went overkill and just monitoring the CV afterall will be good enough?  Seems the guy that did the other schematic thought so.