Attack and Release Times

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

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

NewYorkDave

Well-known member
Joined
Jun 4, 2004
Messages
4,378
Location
New York (Hudson Valley)
I realized just today that there's at least one gaping hole in my knowledge of the fundamentals.

When a compressor's attack time is specified as so-many milliseconds, does that refer to the time required for the compressor to begin acting upon the signal, or the time to achieve the full amount of gain reduction? To put it another way: assuming the gain control element responds instantaneously, if you're considering an RC filter following a rectifier in a sidechain, is the nominal attack time T=RC (63% charge), or 5RC (100% charge)?
 
> does that refer to...

Whatever the spec-writer wants it to mean. There is no agreed standard. There are enough differences between limiters, both function and intent, to make any single standard misleading for some unit.

> the time required for the compressor to begin acting upon the signal, or the time to achieve the full amount of gain reduction?

Define "begin". Define "full". 1dB? 33dB?

Actually, for the simple limiters like 660 and 175, the time-constant is constant, near enough.

> if you're considering an RC filter following a rectifier in a sidechain, is the nominal attack time T=RC (63% charge), or 5RC (100% charge)?

Right, except 5RC is not 100%, and the 63% point is not magic to limiter users.

I design for 1*RC, and that seems to be the numbers on a 660. Of course the mS numbers are meaningless except to mark the scale: you have to diddle the knob to suit your program material.
 
[quote author="PRR"]
Right, except 5RC is not 100%[/quote]

Whoops... "Within 1% of final value." How's that?

(The best teachers are, oftentimes, ball-busters :razz: ).

Define "begin".

Good point. That's nebulous.

Define "full".

When, for an input of a given amplitude, the compressor has compressed the signal as much as it's going to compress without a change in that input amplitude.

The reason I asked is because I chose certain values of RC in a compressor I modified, empirically, based simply on what seemed to work best in that circuit. And I was pondering how I would answer the question (if it came up) of what was the range of attack and release times.

Hey, I just came across a Rane tech note in which the author grapples with the definitions of attack and release times. Here's the quote:

Attack Time

How long the compressor takes to control the signal after the actual overload occurs. In a good compressor with adjustable attack time, expect a range of adjustment from 500 microseconds (us) to 100 milliseconds (ms). Some manufacturers specify this time from the beginning of overload to some arbitrary gain reduction -- perhaps -3 dB. Others specify it from the start of overload to the time the gain stops changing. Perhaps the most meaningful is to specify from start of overload to when the system gain is within, say, 3 dB of the final control point for 10 dB of gain reduction. Unfortunately, in the war of specsmanship, this measurement suffers by comparison. In a system where this parameter is adjustable, ridiculously short times are often spec'd just to look good. Some manufacturers state the attack time as from onset of a very large, very fast overload to the beginning of gain control. This is an artificially short time since it really doesn't reflect how long it takes to get the overload under control. The important consideration is "how long does it take to control an overload?" Equally important as speed, is the shaping of the attack function. If badly done, even a slow attack will sound abrupt and "clicky." Unfortunately, the buyer is not usually informed as to this critical part of the design. This is where a reviewer's article could pay off, as well as a carefully done listening test by the potential buyer. If during testing, the compressor sounds intrusive when the attack is reduced below 1 ms, try a different compressor.



Release Time

How long the compressor takes to relinquish control once the overload passes. Same problems of specification as the attack time, but of less consequence. Normal adjustment range is from 100 ms to 3 sec or more. A short release time of 100 to 500 ms is a good starting range for spoken voice, while the longer times are better for instrumental music. This time inversely affects the distortion added by the compression process, as will be shown. Release time is usually adjustable. Here again, the buyer is at the mercy of the designer. Much has been done with release circuitry over the years to produce good compressors. Terms like "program dependent" and "dynamics dependent" abound. Some have genuine meaning, while some are hype to get the buyer's attention. Bottom line? How does it sound when compressing?
 
> for an input of a given amplitude, the compressor has compressed the signal as much as it's going to compress without a change in that input amplitude.

To bust your chops again: an exponential time constant will take infinitely long to reach "as much as it's going to". Which is not a useful answer.

To get some use out of it, we need an error-band. Say "within 1dB". So for an input that would cause 20dB GR after infinite time, we look for the 19dB GR point. However for a 10dB signal, at the same time, an exponential curve would give like 0.5dB, and the 9dB (1dB error) point would happen sooner.

Selection of an error-band is controversial. A disk-cutter working to maximize level and time per side, with look-ahead, may be working to tolerances better than 1dB. A system that clips harmlessly may be quite fine if error is often 3dB. Note that for ~10dB ultimate GR, 3dB or 4dB error is essentially the 63% point on the exponential curve.

An awful lot of limiters have 2-pole response. The two poles are not the same; that would be borderline unstable and in practice pretty ratty. But there may be a short time constant in series with the adjustable time constant. And cap-coupled rectifiers are prone to slow per-cycle time constants: a 0.1uFd coupling cap driving a rectifier and 1.0uFd time cap can't jack-up the time-cap more than 10% per half-cycle, and can be very slow to rise for a bass transient.

So: it's complicated. And ultimately meaningless. Number the knob from 0 to 11, or from "warpdrive" to "slug-slow".

> answer the question ...of what was the range of attack and release times.

"Proprietary trade secret based on extensive empirical experimentation", or "E^3". (We don't get superscripts here?)
 
It's really fun reading PRR winding up NewYork Dave!

FWIW, I've 'specified' one or two compressors in my time.....
I specify the attack time as being the time from the commencement of a transient, or the onset of a loud signal, to the moment when the gain is (sensibly) stable and compressed.
I specify the release time as the time from the cessation of a signal to the time when the gain has (again, sensibly) returned to the uncompressed state.
I think that's the only reasonable way to specify it, because many compressors, mine included, have non-linear attack and release curves; it's normal practice to incorporate programme dependent attack and release characteristics.... and I (once again) agree with PRR that it's cissy to put actual times on the controls..... compressors should be operated blind.... wind up the knobs 'till it sounds right!
Sadly, I still get lots of correspondence asking for my advice on numbers of milliseconds attack and how many dB of compression is 'best' for various signals...... Ignore the numbers!
 
One problem is that if you have a feedback topology and adjustable ratios, the attack and release times are functions of the ratio. That's why on things like the 1176 which has the feedback topology and ratios from 4:1 -20:1 there are no labels for the attack/release. Same with the Forssell opto, and the SSL buss comp. All this further supports the 'use your ears' camp!

Cheers,

Kris
 
I hate to advocate the gratuitous use of microcontrollers, but it does seem like this kind of stuff could use some parameter recall functions at least, or even some more sophisticated interactive adjustments.

I am ignorant of a lot of stuff out there---does anyone know if this sort of thing exists already?
 
The RNC1173 compressor has a digital sidechain. AD conversion, then a little computer then DA to control the VCA. But no parameter recall functions.

chrissugar
 
[quote author="PRR"] an exponential time constant will take infinitely long to reach "as much as it's going to". Which is not a useful answer. [/quote]

This is an oldie but a goodie:

A mathematician and an engineer are working in an office. Suddenly, a beautiful woman appears at the door. She says to the two men, "Now I'm going to take several steps toward you until I've halved the distance between us. Then I will halve that distance again and again until I reach you. And then you may have your way with me."

"Oh no", sighs the mathematician. "She'll keep getting closer and closer but the distance between her and us will never reach zero."

The engineer just raises one eyebrow and grins at his colleague. "That's true", he says, "but she'll get close enough."
 
[quote author="bcarso"]I hate to advocate the gratuitous use of microcontrollers, ...<snip>... does anyone know if this sort of thing exists already?[/quote]
Yup. It's called VD, and it's not as bad as it sounds.

Amek Virtual Dynamics uses asn A/D to sense signal level, then sends a CV to the VCA. It's used on a few downmarket consoles, but also on the 9098i. On the 9098i it can be pre-EQ, post-EQ, feed-forward, feed-back, pre-or-post anything more or less. all set by jumpers on the motherboard.

The standalone rack unit, the 'System 9098' dual compressor uses this, but 'hides' its digital sidechain heritage behind front panel Sifam meters and rotary 'hard' controls. it's still a digital sidechain and a 2-channel 'VD' card inside though.

People like it. When it's done right, and code is written well, it can be a thing of beauty.

Keith
 
I'm sure there are good digital solutions to this time constant thing.
Usally there is a software solution to most hardware problems if you want to go that route.

4 engineers were driving in a car. A mechanical engineer, a chemical engineer, an electrical engineer, and a software engineer.

Suddenly the car sputters and stops.

The Mechanical Engineer says " Maybe it's the timing belt" Whereupon, the Chemical Engineer says "must be water in the fuel". The Electrical Engineer proclaims "I think the spark is missing".


The Software Engineer pauses and goes "Maybe we should all just get out of the car and get back in.
 
> is (sensibly) stable and compressed.

> I will halve that distance again and again ... "...she'll get close enough."

Busting aside, these come to about the same thing.

While it will take infinitely long to reach zero distance/error, there is an interval when error is large and a point where error is negligibly small. The distance does not have to reach zero, only has to reduce to the length of your, uh, arms. In the limiter, error is large for a while, and takes a very long time to become very small, but the ear-resolution is around 10% or 1dB and for short periods perhaps 2dB or 3dB.

We should also distingush between limiting for engineering and limiting for effect. The audio engineer may want to eliminate clipping, or minimize audible clipping. To quash all clipping needs attack OTOO 50uS. If shutters clang or grooves cross, you need such speed. If clipping is just clipping, no horrible side effects, the ear is pretty deaf to clipping of a millisecond or so. If instead of clipping you have soft distortion, like tape, much slower attack may do "interesting" things to transients.

I don't limit for effect, only as protection against surprises. I want a pretty fast attack. Since I don't use ribbon-shutter optical film recorders or phonodisks any more, a little clipping is OK if nobody notices. I like to know where "about 1mS is". Most of my music is low-transient, so I have little trouble with ducking. Occasionally I have to pump-up speech for maximum clarity, fidelity not the prime directive, so I'll set a slower attack to let some speech sounds clip rather than have them duck the subsequent sound. 50mS is my starting place, but it depends a lot on the speech and on what the limiter is really doing. And on music, I prefer not to limit at all: 10dB limiting is a major boo-boo, seeing the -1dB light just wink on the loudest note is about right for tape, on CD I'd rather not see limiting at all.

> I hate to advocate the gratuitous use of microcontrollers

In broadcast sound processing, it is all about the mini-brain. AM broadcasters will eat hats to get 1dB higher average loudness, because that is 10% more range and (assuming uniform population) 20% more listeners and 20% more ad-rate for the same operating cost. That's where the heavy work is.

For an instantly available recall-limiter, DBX (I think) has a $799 loudspeaker controller that does X-over and EQ and delay, but also has limiting and you can save/recall settings. About 99 scenes if your finger does not wear out on the few-button panel interface, but you can link a PC and do an infinite number of settings.
 
Back
Top