Understanding Group Delay for Channel EQ

[fohman]

Hello all,

 

I want to understand my Group Delay measurement for the Channel EQ in Logic X. I am running pink noise through the EQ setting in the picture, and then out into Faber Acoustic Electroacoustics Toolbox, using the Group Delay (top) and Phase (bottom) transfer functions to measure things. More specifically, I've got a tone generator on an audio channel with no output assignment, but with sends to Bus 1 and Bus 2, of which Bus 2 has the EQ and goes to the "source" input for the Dual FFT transfer function, and Bus 1 goes to the "reference" side, and has a 15 sample delay to compensate for the inherent delay in the Channel EQ at my 48kHz project sample rate. Also, when I remove the delay and the EQ, both the Group Delay and Phase lines are perfectly flat at 0, and when I put a delay on the source channel and begin to increase it, the Group Delay line moves upward and the Phase line curves (and wraps) downward... and so I'm assuming that these measurements are accurate, but maybe not?

 

My question is this:

 

How is it possible, given that the Channel EQ only introduces about .31 milliseconds of overall delay, that there are frequencies at its output (in this case, around 30Hz) which register on the Group Delay at as much as -100ms, or "100 milliseconds ahead of" the reference signal? Even if the Group Delay showed a -1ms reading, how could it be possible for a device to output a sound that beats its simultaneous unfiltered copy to the finish line? Am I missing something?

[fohman]

Ok everyone, here's a response I got from someone trustworthy through email:

 

"You may be confusing group delay with phase delay. With phase delay (essentially phase at a given frequency divided by that frequency), you would expect a positive result, since there will always be some delay in the system under test. Group delay is the derivative (i.e. slope or rate of change) of the phase shift with respect to frequency. So, with a linear phase system, you would expect a constant group delay over frequency. When the time delay through the system varies with respect to frequency (nonlinear phase), you may encounter a negative group delay. It doesn't mean that time constraints have been violated."

 

And my response:

 

"Thanks for your explanation, I don't fully understand at the moment, but I'll keep thinking about it. So, is there a graph which will tell me, on a frequency basis, where things actually land in time? In other words, if I wanted to put a time stamp on every frequency passing through the EQ, to see where they all land on the other side relative to a set starting point for all frequencies, how do I do this?"

 

Does anyone have some ideas?