Time constants in compressors

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

Viitalahde

Well-known member
Joined
Nov 7, 2005
Messages
727
Location
Kuhmoinen, Finland
This morning's thoughts:

I started wondering about the basic attack/release time constant circuit used in many compressors.

Code:
                   R1
IN >---------/\/\/\/-------o--------o-----------> OUT
                           |        |
                           |        /
                      C1  ---       \ R2
                          ---       / 
                           |        \
                           |        |
                           |        |
                           o----o---o
                                |
                                |
                               ---
                               |||

The basic circuit: R1 defines how C1 is charged, and R2 defines the dis-charge. Attack & release.

Now, without doing any maths (yet), I am wondering if there could be a sonic difference between these two circuits?

A) R1 & R2 are large, C1 = small

b) R1 & R2 are small, C1 = large

Both cases designed for same mathematical charge & dis-charge. Case A is obviously a high impedance circuit where B is a low-impedance circuit with high capacitance.

If both circuits act the same way on the paper, I suppose the driver before this could make a difference? Low-Z with high capacitance.. tough to drive.

From somewhere I remember the Fairchild 670 has a beefy (8 watts or so) sidechain driver?

I quess the ultimate question is: how do the designers end up with the R/C combinations in sidechains? They design the rectifier & driver and calculate around that, then listen & tweak?
 
As I've been messin´ around alot with the N*eve circuits I came to a point where the attack / release acts strange, if the attacktime is slow(high R1 value) and the release fast( small R2 value), C1 won't charge full as R1 / R2 acts like a voltagediver! And as the Ratio switch is before the timing circuit this kind of settings will get you lower ratio? :?
 
You use a slide rule, fool!

es_3.jpg


And then, you see those big ears? :razz:
 
[quote author="Viitalahde"]This morning's thoughts:

I started wondering about the basic attack/release time constant circuit used in many compressors.

Code:
                   R1
IN >---------/\/\/\/-------o--------o-----------> OUT
                           |        |
                           |        /
                      C1  ---       \ R2
                          ---       / 
                           |        \
                           |        |
                           |        |
                           o----o---o
                                |
                                |
                               ---
                               |||

The basic circuit: R1 defines how C1 is charged, and R2 defines the dis-charge. Attack & release.

Now, without doing any maths (yet), I am wondering if there could be a sonic difference between these two circuits?

A) R1 & R2 are large, C1 = small

b) R1 & R2 are small, C1 = large

Both cases designed for same mathematical charge & dis-charge. Case A is obviously a high impedance circuit where B is a low-impedance circuit with high capacitance.

If both circuits act the same way on the paper, I suppose the driver before this could make a difference? Low-Z with high capacitance.. tough to drive.

From somewhere I remember the Fairchild 670 has a beefy (8 watts or so) sidechain driver?

I quess the ultimate question is: how do the designers end up with the R/C combinations in sidechains? They design the rectifier & driver and calculate around that, then listen & tweak?[/quote]

You probably know this, but for the sake of those who may not:

Before one gets too busy with the slide rule realize that the "IN" port is not a typical voltage source, but has high Z for one current polarity and low for the other. Sometimes this is done with a diode or diode enclosed in an op amp feedback loop, sometimes with a transistor etc.

Suppose we are sensing magnitude peaks and converting these to positive voltage pulses. So the voltage at the left of R1 charges the cap with roughly an R1*C1 time constant (R1 << R2), the attack time constant.

When the voltage falls below the cap voltage the current from R1 ceases, for a drive circuit that acts like an ideal diode. At this point R1 is open-circuited and C1 discharges toward ground with an R2*C2 time constant---the release time constant.

Having lower or higher impedances for the network shown will, as you acknowledge, have consequences---how hard is it to drive?---how much does the loading by the circuitry downstream of it affect things?---but if those effects are taken into account, there should be no effect on the function. If you make the cap large you will have to use a 'lytic; too small the R's get big and leakage currents become important, and things could even get a bit noisy.
 
It starts with the music, and what you want to do to it, and what you don't want to do to it.

Artifacts are what you don't want.

Remember, the 660/670 were designed not for rock and roll.

Maybe one setting is for a cymbel crash.
Maybe another is for a vocal peak.
And another for a overall rise in volume from an orchestra.

Once you have that, then you pick some times you think might work.

Then you listen. Most of the tweaking done on the 660/670 was those time constants. The rest of the circuit is actually pretty simple.

Balanced transformer coupled audio makeup.
No caps in the signal path.
Then a balanced voltage amp.
A power supply, and those time constants.
 
[quote author="Viitalahde"]
A) R1 & R2 are large, C1 = small

b) R1 & R2 are small, C1 = large

Both cases designed for same mathematical charge & dis-charge. Case A is obviously a high impedance circuit where B is a low-impedance circuit with high capacitance.

I quess the ultimate question is: how do the designers end up with the R/C combinations in sidechains? They design the rectifier & driver and calculate around that, then listen & tweak?[/quote]
Not sure, but is you main question how R and C are chosen given certain attack & release-times (so the RC-products already known) ?
 
without doing any maths

Well, do the math and your questions will be answered. :wink:

We can spend a whole lot of words describing what can be summarized in a simple formula.

T=RC

Ask yourself: for a particular value of C, what must be the value of R in order to allow C to charge within a given time (T)?

Then it'll become obvious why compressors designed for a fast attack time use a sidechain driver amplifier with a low source impedance. And it'll also become obvious why an Altec 436C (as an example) cannot achieve very fast attack times.

Also: you have to look at the R-C network in its complete context, wrapped around an amplifier. In other words, be aware that the likelihood for unwanted feedback (motorboating) increases as you reduce the time constants. When I was doing the mod on Soundguy's 436 recently, it became apparent to me pretty quickly why the designers of the 436C chose those particular time constants: if you reduce them more than slightly, the amp motorboats.
 
> Case A is obviously a high impedance circuit where B is a low-impedance circuit with high capacitance.

This is a very real problem.

A "good" limiter will capture a sub-milliSecond peak and hold gain down for a second.

To hold gain down, you hold a negative voltage on the grids. The grids need a DC path to absorb their leakage. If this path is too high resistance, the I*R drop will cause a voltage that screws-up your control voltage.

At the other side, you have to grab enough current and charge in a milliSecond to feed the grid resistor for a whole second. The charging current can be at least 1,000 times higher than the discharge current. This can easily be a Problem.

Tubes usually specify 1Meg max grid resistor. 660/670 had eight grids to feed. 1Meg/8= 125K grid resistor. It takes -50V to get deep limiting. 50V/125K= 0.4 milliAmps grid current, which has to be maintained for a whole second. To capture this much power from a milliSecond transient, we need 0.4mA*1,000= 400mA from the rectifier and its driver. And we need it at 50V, so 400mA*50V= 20 Watts peak power, or a 10W sine-RMS amplifier.

0.1W is easy. 1W isn't too hard. 10W amps are big toys, especially in vacuum tubes.

The lesser tube limiters don't work this hard, it would be too expensive. Just two grids, and cheating the max resistance spec using 1Meg or even 2Meg for two tubes. (Not all tubes need to go to 1Meg max, just the worst-case ones. In pro limiter work, leakers would fail calibration and be replaced.) Also they don't have super-fast attack, no 1,000:1 ratio of attack to release, no 1,000:1 ratio of charge to discharge current. Also lower voltages on the tubes. -20V in 1Meg is 0.02mA, a 500:1 attack/release ratio suggests 10mA charge current, which can be tapped from a 30mA+30mA push-pull stage.

The Fairchild does better than 1,000:1, so Narma cheated the grid resistor up to about 270K(?) to reduce discharge current towards 0.2mA. With the attack times he needed for disk-cutting, this gave ~300mA @ 50V = 15 Watts peak attack power, which could be done with a pair of 6V6 and a 3:1 transformer.

OK, change the ~1Meg 1uFd discharge side to 10K 100uFd. In the old days, you could not trust a 100uFd electrolytic to leak less than a 10K resistor, but today that's fine. If you want milliSecond attack, you need a 10K/1000= 10 ohm attack resistor. If you use 20V max control voltage, you could need 20V/10Ω= 2 Amps attack current, at 20V is 40W peak 20W sin-RMS.

The attack amp never really delivers the 10W-20W long-term, only on the big peaks. In transistor work, we can rely on that to use a small heatsink. But a linear tube amp generally has to idle at fairly large current so the Fairchild's 20Wpk amp is as big and hot as any 10W sin-RMS amp.

You are almost always caught between the highest resistance your grids can tolerate, the peak power needed to meet your Attack/Release ratio goal, and your budget. The resistor values are usually FORCED on you, especially with an all-tube sidechain. Transistors change the economics at both ends. A $5 chip can replace the P-P 6V6. We have good negative-swing DC amps now, so we could use a 22Meg release resistor and buffer it down to a low impedance at the grids. We could even do the time constant at -5V and amplify it down to -50V for about $2 (plus a negative rail).
 
Bcarso, Dave, PRR and the rest - thank you!

I wasn't asking about calculating the time constants T=R*C, but was just curious about the relationship between R and C. High or low R or C?

I pretty much got now - it really is a matter of the rest of the circuit & economics. Sidechain driver, the "port" the CV is dumped to.. So when you have your limiter topology and you know its charasteristics (FET, remote-cutoff triodes etc) you are given a "window" to calculate your time constants to.

When I posted my question, I did not actually think of charging currents, but it makes perfect sense. A few consumed watts on the attack period starts to raise sweat on the designers forehead. :razz:

http://www.elecdesign.com/Articles/Index.cfm?AD=1&ArticleID=4478

This link Crusty2 posted is a good one and gives me some thoughts. The problem of a small release resistor<->big capacitor versus huge release resistor<->small capacitor is an obvious one. Huge charging currents vs. hard-to-buy potentionmeters.

Of course, in normal recording/mastering use, release times of over 1-2 seconds are rare, so it's not a real problem in this use. Looks like release is easier and attack is harder, design-wise? If the driver amplifier can't deliver all the current the attack period needs, it translates to slowed-down attack?

I wonder if anyone has used either a FET or an LDR as the release resistor? Those would offer interesting opportunieties for auto release experiments.
 
FETs and LDRs are pretty poor-tolerance resistors in general, as especially so at high values, although matched pairs with one side dedicated to a servo could be workable, depending. But I doubt that one would try to use them as a variable element in this application.

One circuit I've used for pulse-stretching that has the appeal of simplicity is a simple emitter follower with a diode in series with the base and the emitter going to the capacitor. You get the beta current gain so it allows the preceding stage to run at relatively low power. It has the disadvantage of two diode voltage drops, and the diode has to be a low-leakage type. If you stay below the emiiter-base breakdown voltage of roughly 6 volts you can skip the diode, and the typical below-breakdown leakage of a decent bipolar is very low.

Then buffer with a JFET, or a CMOS input op amp.

For the sand-averse, some of the supertriodes like the 6H30 have peak currents in the ampere region. Of course you wouldn't be running them class A :razz:

Another technique for really fast capture with long hold is to cascade a fast peak detector with a slower one.
 
> release is easier and attack is harder, design-wise?

No. Consider if Narma had penciled-in a nice 10Meg release resistor. He could use a smaller cap(s) and much less attack power, lovely. BUT the grid current of eight grids in 10Meg would cause several volts unpredictable error in the control voltage. He could have used a cathode follwer buffer, but that adds error and needs a negative power line. Part of his solution is to run the tube grid resistance a little higher than the per-tube spec, knowing that he won't get 8 worst-case grids and if he does, the budget includes calibration which will weed-out the worst leakers. Buying a few extra mini-tubes, which can be re-sold to radio makers, was cheaper than adding another pair of 6V6 or going up to 6L6.

> raise sweat on the designers forehead.

The designer can spec any part he wants; the production engineer sweats the cost. (Obviously a designer of things that can't be produced at profit has a short career....)

Plot the cost of both ends. A milliWatt of drive power is cheap, 100mW not much more, a Watt is money and 10W may be big money. 1K to 900K release resistor is all the same, over 1Meg you may have to sort-out leakers, over 10Meg you have to find super-clean tubes or add a buffer. So you have a curve that is high at both ends but pretty flat between. If the width of the low-cost flat-spot is wider than your Attack/Release ratio, you can pick anywhere in that area. In practice, you don't get a lot of leeway, especially for a high-performance limiter. Your Release resistor will be the highest value your gain-cell (or its buffer) can stand, and you put money onto the drive amp until you meet your Attack spec. In some cases this is negligible added cost, especially for moderate Attack speed: you can tap enough audio power from an existing stage to do it. And in other cases you just go overboard, like the Fairchild and the big GE.

Using JFET attenuators does raise the allowable grid resistance; often you can't readily buy a resistor big enough to upset the FET. 22Meg is often no problem. 100Meg might reduce cap size required Attack current even more, but 100Meg resistors cost more. And adjustable 100Meg pots don't exist (which is where Crusty's link goes; for a lousy 1Meg audio, I'd just look in guitar amp supply sites, or fudge to 500K, but for higher resistances that levered resistor idea may be useful).

Using transistor drivers or buffers sure reduces costs, but also spoils us: we expect low cost.
 
The original question does remind me of some original thinking about attack and release times and shapes from many years ago.... It's true that there can be a conflict; my method of overcoming a possible inflexibility was to introduce a 2 or even multi-step circuit, where a second R/C combination sits on top of the main one. The attack time of the upper small capacitor can be extremely short and doesn't need high current to fill it. This arrangement also provides a form of auto release... for transients the upper capacitor charges and discharges rapidly, for normal signals, the main capacitor charges, and discharges slowly.

The character of the sound of a compressor is very much a combination of the attack and release times and shapes, and the distortion characteristics of the audio during the compression process.
 
Ted, my next question was going to be about these R/C stacks (auto release).

I quess the basic theory is that the smaller capacitor on the top reacts to smaller transients while the lower one doesn't really react (charge) at all? Then the big hairy drummer does his tom fill and that makes the big caps work too. :twisted:
 
It's 'seriesing' them really!
Just stand one of them on top of the other; and then for extra interest, alter the standing voltage of the bottom 'ground' connection, or even make it dynamic, changing with the musical content.... but then it's starting to get complicated!
The Audio and Design F600 compressor (circa 1972) was the first one where I saw the 'auto release' circuit, but I believe that its origins are earlier than that.
 
Here are some small excerpts from literature that references dual time-constant sidechains in compressors. I cannot post more than these excerpts because the works are covered by copyright.

PDF
 
series.. ok for some reason i envisioned them as parallel.

If they are series, which one should come first, the transient stage or the low freq stage?

I guess we could look at the high freq stage as peak detection and the low freq as more of an RMS detection?

do you sum the output from each stage and how do you do it, through simple resistor summing?

EDIT: NYD that is great stuff!

time for a groupDIY discrete compressor I say!


In anyone's opinion, If I were to use a JFET as the level modifying device, would it be better to use the FET in series with the signal allowing it to flow through the FET or shorting the audio to ground via the FET?

I forsee the general issues with allowing the audio to flow through the FET but how much of a problem do you think it really poses? I've seen many large name consoles use FETs as mute elements, so would you say that the issues involved are worse as less than a full gate ON situation?
 
Hi Svart, I'm juming in on the FET discussion....
Operating the FET as a series element is difficult; there are problems caused by both the drain and the source operating at the audio voltage, and with wildly varying impedances as the compression acts.
Using a FET as a shunt device is relatively easy; as long as you can put up with the high levels of distortion caused by the difference between the ac potential of the drain (or is it source.... I can never remember) and the gate. This can be overcome to some extent by applying a degree of AC to the gate, but there is still a significant non-linearity left that shows itself as a quite nasty 3rd order distortion. The 1176 puts up with the distortion, as do the old Audio and Design F600 and F760 series.
But what's wrong with VCAs? :twisted: (joke)
 
Back
Top