I don’t think that’s really the case. Excess group delay will be flat if a signal is passed through a physically possible filter (like an LCR network, or a speaker), but will deviate when the signal mixes with delayed reflections, for example. Neither is a reliable indication of better or worse sound, but it can be used to exclude frequency response errors in certain areas from EQ correction (where the correction would be pointless or harmful).
As far as I know, Dirac uses both IIR and FIR filters. The latter are not physically possible and allow for separate correction of time and frequency. I’m not surprised that this would show up in the excess GD plot.
Speaking of group delay, after reading up on the filter characteristics of AKM DAC chips I started worrying about the different group delays of different filters. Now, if the same of DAC with the same filter is used for all channels then this won’t matter. However, in my system I use different DACs for LR and sub channels. Both have AKM chips but only one lets me select the filter.
So I went and measured the actual delay of the DACs, depending on the filter choice in one of them, and the sample rate. To my surprise, the choice of filter has virtually no impact. The sample rate unsurprisingly does:
| 48kHz | 96kHz | 192kHz |
F1 | 14.9ms | 7.4ms | 3.8ms |
F2 | 15.4ms | 7.6ms | 3.9ms |
F3 | 15.4ms | 7.6ms | 3.9ms |
F4 | 15.4ms | 7.7ms | 3.9ms |
F5 | 15.4ms | 7.7ms | 3.9ms |
F6 | 15.3ms | 7.6ms | 3.8ms |
What this means for Dirac setups is, if you run the DACs during Dirac Live measurement at a different sample rate than when listening to music, your bass timings will be off considerably. The only solution is either using the same DAC model for LR and subs, or sticking to a fixed replay sample rate and using that during Dirac Live as well.
I understand your points.
As for the DAC I use a multi-channel interface so the point is not applicable.
We can also leave out the point of excess of phase and speak only of GD. If the uncorrected system has few millisec of GD at low end, with the correction I can expect a variation proportional to that of magnitude, since Dirac does not use FIR at low frequencies.
But this does not happen. What happens is a global increase in GD over the entire low range, even reaching 120 ms sometimes.
I also tried to translate the target so that the low frequencies were little corrected in magnitude, shifting the large variation towards the high frequencies.
But nothing, same results.
For a correction system this is a strange behaviour, Although we consider that it only works with IIR, at low frequencies.
If I apply a minimum phase correction with Audiolense, this does not happen. Indeed, Audiolense always improves the GD, as expected.
However, there is a point to add. This is true for the main listening position.
On a larger area, Dirac is able to maintain greater consistency of magnitude at low frequencies, unlike Audiolense.
So, perhaps, the question of Dirac's extra GD is to be found precisely in this effect.
Then we can talk about whether the consistency over a large area or a low GD over a narrower area is better.
At the moment I am still trying to understand if the matter is like this or if Dirac is simply wrong (I doubt).
The fact that Dirac delivers highly variable results to each filter creation iteration, as well as measurements, however, does not give me much confidence in the robustness of the algorithm.
However, the fact is that the bass at 100ms of GD is audibly of poor quality.
I think
@mitchco can certainly make a valuable contribution to the discussion.