The Group Delay at a frequency is the slope of the phase at that frequency. It provides an indication of how the envelope of signals at that frequency is delayed in the course of passing through the audio chain - if everything is delayed by the same amount the shape of the signal is not changed, it simply arrives later, but delays that vary change the shape of the signal.

Anything that affects the phase will also affect the group delay, which includes crossovers, the response of drivers, the low and high frequency roll-offs of the various parts of the audio chain, any time delays in the system and the filtering effect of the room, its modal response and the frequency-dependent absorptions of its surfaces and furnishings.

There is a relationship between the magnitude response of a system and its phase response, wherever the magnitude is not flat there will be some corresponding change in the slope of the phase - the sharper the magnitude change, the steeper the slope of the corresponding phase change and hence the larger the group delay. Since every wiggle in the magnitude has a corresponding feature in the group delay, it is difficult to isolate parts of the group delay and attribute them to any particular cause.

In 'minimum phase' systems there is a particular mathematical relationship between the magnitude response and the corresponding phase. By calculating that 'minimum phase' and its corresponding group delay, the excess phase and excess group delay can be found. There are two useful pieces of information that can come from the excess group delay: (1) flat areas correspond to time delays, the amount of the time delay can be read directly from the excess group delay plot. This can reveal time alignment issues between drive units, or simply delays in the measurement chain. (2) areas that are not flat indicate areas where the response is not minimum phase, and that deviation is not simply due to a time delay. The particular shape and values of the non-flat parts cannot tell us what caused them, but knowing that the response is not minimum phase in that region is nonetheless useful because the filters we can apply are themselves minimum phase. In minimum phase regions of the response filters could be applied to correct both the magnitude and phase of the signal. In regions that are not minimum phase the magnitude could be corrected, but the phase would still have deviations, which means the resulting filtered signal would still be different than the original even if the magnitude is now flat.

 

Is it possible when you get a chance to present us an example of Interpreting Group Delay Graphs?

Thanks,
crom

 

This thread is a bit old, and I am a newcomer in this forum, but a long time and enthousiastic user of REW.

I discover by chance this method of the Excess Group Delay to define the alignement of my loudspeakers. It matches perfectly my listening impressions, but seems to deliver contradictory results with other methods I have used, based on the time alignment of the impulse.

Am I doing something wrong?

Dominique

 

I don't know if I can help. I haven't used GD or EGD for XO timing so I can't help with that. I have used 2 or 3 other methods so if you are looking to understand one those better, I may be able help with that.

Given the right measurements I could also compare your current timing to those I would suggest using the phase tracking method.

 

I was reading from another source that when comparing phase between drivers the one with the steeper phase plot is the one that is delayed compared to the one with the shallower plot. Therefore, we would want to delay the driver with the shallower of the two phase plots to match with the other one.

Is that correct, and a good rule to follow?

 

Yes, given any arbitrary fixed time reference, increasing the delay of a driver will increase the slope of its phase in the XO frequency range. This process will work fine in practice assuming any reasonable reference time. If this thought process and approach applies well to your situation then it is a good way to think of it.

I prefer to first set the reference time such that the HF driver bandpass range is reasonably close to 0° and then just adjust the LF driver delay ± as needed to get it's phase to best follow the HF phase.

Close direct sound phase tracking between drivers is my preferred target. It doesn't matter what method is used to get there.