-Only I'd like to start by taking a different tack:
Imagine that YOU are the limiter. You're a simple infinite ratio limiter.
To do the job, you need two basic tools: a means of determining the signal level, and a means of controlling the signal level.
So, I give you a level meter to look at, and a volume control dial to turn. Now you can turn the level down when it gets too loud, and limit the signal level.
That's it, -right?
Well, basically yes, but it overlooks some important details: Firstly, what is the meter reading? -the INPUT level (before the volume control) or the OUTPUT level... (after the volume control)? -This is not a subtle distinction, and in order to discover why, let's consider the latter example first.
So your meter (note, I haven't said 'VU meter', but we;ll get back to that later!) reads the OUTPUT level. This is good because it allows you to confirm that the job is being done correctly. -The task that you have is exceedingly simple: if the needle is to the right of the fixed limit, you turn it down some more until it is AT the limit, -If it's to the left, you can set about increasing the 'volume control', to restore the original gain structure once more. -that's the only info you need: above threshold or below threshold. No more info is necessary: you don't need to know how many dB over or under, just that the gain needs to go higher, go lower, or be left alone. -This is a Feed BACK topology, because the OUTPUT is being monitored, and any gain-corrective action is fed BACK to a point BEFORE the monitoring (signal level meter).
Now let's consider feed FORWARD. This uses the same two tools (a meter and a volume control) but in the reverse order in the signal path. Now the meter reads the INPUT, and the volume control then makes the adjustment. -In this instance there is NO way to 'check-up' on how good a job was being done... we just have to trust that the amount by which the signal threshold was exceeded has been accurately corrected... so now we need MORE information than the first example: we need our meter to tell us HOW MANY dB the signal is over the threshold (in the first example we just needed 'over/under' info)... AND we need a volume control which is calibrated so that we can tell how much gain reduction we're applying... in the first example, we didn't need ANY such information.
That's the topological difference, but now let's consider the meter: a VU meter indicates the RMS content of a signal. The 'average energy' if you like. -A peak meter doesn't care what the average is, it only tells you what the MAXIMUM height is. It's like a skyline analogy; some skylines are generally low, but have one stupendously high tower, others -like Manhatten- are all fairly tall. -Scale for the same peak readings , and the average is different... but -using a camera- zoom in to get the average height, and one skyline gets the high building tops chopped-off.
Similarly, a VU meter will read a square and a triangle with the same peak level as VERY different levels... but the clipping point of both is identical. so a peak meter will tell you they're the same level. -On the other hand, a taller triangle might have the same apparent volume (RMS content) as a smaller square wave, but a peak meter will tell you that they're different levels, while a VU meter says they're the same!
-Who is right? well, neither of them or both of them, or perhaps one or the other... depending on what you're looking for.
Relating this back to compressors, some units (like dbx) use RMS detection (the means of 'metering' the signal) and others (like the 1176) use peak detection. -B-I-G- difference in compression result.
But while the dbx designs use RMS sensing and a VCA, not all VCA compressors use RMS: the SSL buss compressor for example is peak detecting.
Some gain control methods do not lend themselves to feed-forward because they are not linear or predictable, or consistent from device to device: FETs and Cadmium-Sulfide Opto-resistors are examples, but vactrol type optos are much more consistent, so let's not lump all optos together! -VCAs however are VERY linear and predictable, and so is PWM. One gives log result, the other linear, but VERY consistent. As for detection, True RMS detection is very predictable, as peak detection can also be. but there aren't many vari-mu designs which are feed-forward... it's possible (and I believe that the Gyraf vari-mu design is FF) but it takes work and clever design to overcome certain trickinesses. -Most vari-mu's are FB.
Hmmmm... Have I strayed from topologies too much? -It's tricky to include examples which I'd consider relevant, without confusing the subject with distractions perhaps... none the less, I hope that clarifies a little. -Otherwise let me know what I've made a mess of and I'll try to straighten it out.
But as a summary, Feed BACK compression /limiting is when the detector (measuring) is affected BY the gain reduction due to being downstream. Feed FORWARD is when the measuring is upstream, and NOT affected by the gain reduction which happens afterwards.