[ now playing: Jethro Tull, "Wond'ring Aloud ]
OK - just what the hell is going on here, anyway?
This delay has now been demonstrated in so many titles, I've started wondering if there aren't just some natural phenomena at work here, as
DuncanS humbly submitted waaay back up the road in this thread.
As we know, the LFE channel, among other treatments, is (ostensibly) a low-pass filtered extract from the main program. I understand for the most part the way filters work, and the phase shift they produce. But something with which I still struggle is the concept of group delay. I've read all the technical explanations of what it is, and they seem to make sense enough, but I'm just not able to grok the overall effect for some reason. But Duncan's reminder has stuck with me throughout my reading of this developing saga, and has made me curiouser and curiouser. So I decided to make an LFE channel of my own, from scratch, to see the exact changes that occur to the timing of the waveform.
I wanted something with a nice big convenient bassy transient right at the start, so of course I went straight to "Morph the Cat."
Starting with the stereo CD mix of course, I loaded it into Cool Edit. First step was to do a mono mix, so I'd be starting with 2 identical, time-aligned waveforms to play around with. After a quick zoom-in to verify this, I then ran an 80 Hz, 4-pole Butterworth filter (a typical LFE treatment, I should think) on the left channel only, to do a direct comparison. Here's what came out:
The full-range version is on top, the filtered version of that same waveform below. I've used the selection tool to measure the delay; it starts at a waveform peak on the full-range, and ends with the corresponding peak on the low-pass filtered version. Way down in the lower-right corner I've circled the length of the selection (delay), which is 6 milliseconds.
To satisfy my own curiosity I did another round, this time moving the cutoff to 120 Hz just to observe any changes. Here's that output:
This delay came in at 4 mS.
The 2 conclusions I'm drawing from this experiment are:
1. Allowing for rounding errors caused by the 1 mS resolution of the Cool Edit timeline, these figures line up just fine with Duncan's formula above.
2. These delays are in the general area of those being discovered by so many here, and I believe they account in large part for what we're seeing. We all know that the processing of the LFE channel can vary widely from one title to the next, from the ridiculous full-range examples to any number of unspecified LP filter treatments. There's certainly no standard being adhered to with this stuff. In my estimation, this would account for a lot of the variation in delay times being measured.
So for me, the question now becomes: Why do some titles NOT show this delay?? Perhaps they were processed in the digital domain, using FIR filters (which can be configured for no phase shift or group delay) as opposed to the more conventional IIR types, which are basically digital versions of classic analog filters, with all the attendant phase behavior? That's my guess at this point.
Any other thoughts or ideas gratefully received.
Finally - Thanks DUNCAN!