Neither in my case and probably we are not alone with this
-
Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
Excess group delay and mismatched phase response
- Thread starter pollock0424
- Start date
I'm just assuming it's the best psycho acoustically solution. Or the best compromise with other factors.Neither in my case and probably we are not alone with this
Because the alternative is an error in the filter creation algorithm I suppose...
Last edited:
I think I understand what you say, and in fact it demonstrates in a certain sense the hypothesis that a better consistency is achieved with the GD increasing.
But what I was saying is that if you measure a main + sub system with 80Hz Xover or the same sub with the same Xover alone, you shouldn't detect a coarse GD envelope change below 80Hz like shown in the graphs you posted.
Then coarse is a relative terms, the diff is probably not audible.
But in sub only graph there are about 10ms difference between 30hz and 55hz, while with sub+main there are about 30ms diff for the same freq points. The peak even changes from 55Hz to about 40Hz.
I can't explain it ..
Last edited:
Lion♡
Senior Member
Hmm...
Did GD need to increase over there in Dirac Pulse I upload as an example? ->
What I gave such a delay to Dirac Pulse in the previous article was trying to copy what appears on DLBC.
If my response was Dirac Pulse there is no delay or polarity required for this. Isn't this consistent?
And Just in case you misunderstand, I'm not criticizing Dirac. Rather, I like and support the DRC process, including Dirac.
Also I think what we're talking about is similar but slightly different. Was it insufficient in my example for my word?
This is the LR4 80Hz LP that corresponds to the subwoofer. No delay.
And this is Sub with 20ms delay and reverse polarity applied in advance. (80hz LR4 LP assumed Sub)
And now this is an alignment of that.
The reason why I'm posting the capture once again is that I'm talking about the delay (or polarity?) between Sub only and Sub+Main that you told me.
Ok, I better understand what you were saying. Could explain the difference in fact. I forgot that Dirac does not use steep crossover like other DRCs.
In any case, if GD increased is a compromise or not, only Dirac can say it I think...
The fact remains that, so much, in my case it is audible and not very pleasant.
PS. It must be said that Dirac works to optimize a volume instead of a point, so a single point measurement may not be representative...
I will try to make several measurements.
Last edited:
Lion♡
Senior Member
I can't explain it either, because I don't know myself. That's why I've been following these threads with interest, testing it alone in REW, and trying to figure out the consistent intent of DLBC. Also, I thought I had somewhat replicated the pure Dirac Pulse delay and polarity inversion without room error. But the important question is, "Why?" As I mentioned earlier, those two responses do not require delay or polarity inversion. However, if DLBC were involved, it would have resulted like this... but I still don’t understand why.
The hypotheses are:
- The measurement filtering algorithm is not universal enough
- The filter calculation algorithm is not universal enough
- Measurement errors/problems
- This is effectively what provides the minimum mean squared error in low freq response on the multiple points measured (so considering the specific phase response of the system+room)
Last edited:
Lion♡
Senior Member
If that's the case, maybe there was a problem with the pre-DLBC basic DL, but seeing that doesn't happen...
This is a possibility.
But I think it's too many cases to call it a measurement error by some users...
Do you only use DLBC?
I've often seen some people put minidsp on the base dirac and the subwoofer put minidsp separately and use it directly among the sub. (Of course they own DLBC too)
I have put DLBC up for sale to keep Audiolense, which gives me much more predictability and the possibility of creating digital crossovers also for woofer/tweeter. But I haven't found any buyers yet, so every now and then I play with it and use it for TV where less latency is needed.If that's the case, maybe there was a problem with the pre-DLBC basic DL, but seeing that doesn't happen...
This is a possibility.
But I think it's too many cases to call it a measurement error by some users...
Do you only use DLBC?
I've often seen some people put minidsp on the base dirac and the subwoofer put minidsp separately and use it directly among the sub. (Of course they own DLBC too)
In terms of preference there is not much difference, bass aside. In some cases, when I can make Dirac work well (thus playing with the microphone and the measurements) I get a little better sound, with the same frequency target. But it really depends on the process of measuring and creating the filter... according to criteria that I have not yet understood and which honestly shouldn't worry as users.
RenPa
Active Member
- Joined
- Dec 7, 2022
- Messages
- 123
- Likes
- 48
I tend to agree. DLBC seems to prioritise FR vs GD.
In my mind this DLBC filter:
should be better than this one bellow because front left and front right response follow up together and I have a less deep dip at 55Hz-issh:
Wrong answers REW. The first filter has this GD:
Which is worst than the second filter:
RenPa
Active Member
- Joined
- Dec 7, 2022
- Messages
- 123
- Likes
- 48
I can assure you. Since I came here in this thread I searched for the filter with the less worst wavelet as suggested by you and you can't imagine the difference it made. It Tends to confirm @OCA magic with Audyssey.... Our ears prefer less GD than straight line FR
RenPa
Active Member
- Joined
- Dec 7, 2022
- Messages
- 123
- Likes
- 48
All I know is since I found that 3.4.4 update was the only one reliable, Dirac finally put 3.11.00 update release that should fix measurements inconsistencies. So your 3rd hypothesis was good!
I would like to add that in my case, with DLBC, temporarily moving the crossover bar elsewhere and then putting it back in the exact same position (by typing the frequency with keyboard) randomly lead to a different estimated FR in the graph. It is evident from the dip(s) in the FR that move and change in level.
Try 10/15 times to move the XO bar and put it back in the same starting position and capture the screenshot of each estimated FR (which is recalculated with each movement of the bar). Then compare them all, and you will see that two specific FRs alternate.
At least that's what happens to me.
And it is not just a graphical issue because I have measured with digital loopback the relative filters of the two FRs that alternate, and they are different
Try 10/15 times to move the XO bar and put it back in the same starting position and capture the screenshot of each estimated FR (which is recalculated with each movement of the bar). Then compare them all, and you will see that two specific FRs alternate.
At least that's what happens to me.
And it is not just a graphical issue because I have measured with digital loopback the relative filters of the two FRs that alternate, and they are different
RenPa
Active Member
- Joined
- Dec 7, 2022
- Messages
- 123
- Likes
- 48
Exactly the same for me. My workaround is always the same with 3.4.4 . I calculate the same xo setings 3-5 times and I see 2-3 different predicted responses. After that it is always those 2-3 choices no matter how many more calculation I do.
What I just found is I can't trust the best «looking» predicted response to be the best sounding one. Now I know why. I must find the less wrost GD one...
That's what I did too before getting bored and falling back on Audiolense, that while not having such a sophisticated correction algorithm, is more predictable and makes the single centered listening point very very linear in time and freq domain.
The bass sounds dry in fact, but it should be technically correct ...
theREALdotnet
Major Contributor
- Joined
- Mar 11, 2022
- Messages
- 1,683
- Likes
- 2,962
Yes, same here. I believe the reason is that the Dirac algorithms don't find an analytic solution to the problem, but are heuristic by nature. Likely because of the computational complexity of the problem. They speak of a "genetic algorithm", I always pictured that as a variation on the simulated annealing heuristic strategy.
I found that even the FR looks different in reality than what Dirac Live predicts, hence checking the filters with measurements is crucial.
RenPa
Active Member
- Joined
- Dec 7, 2022
- Messages
- 123
- Likes
- 48
I agree with checking measurements is crucial but not for the FR response in my case at least. With 3.4.4 versioin of DLBC and MMM method + RTA in REW I always got exactly what DLBC filter predicted. But what I know now, it is not all the picture and I must check GD. Using wavelet taught me an easy way to see what happens that compensate my lack of knowledge even after reading many times the help files in REW. ...amateur...what a shame!
Did you report your GD excess to Dirac support?I agree with checking measurements is crucial but not for the FR response in my case at least. With 3.4.4 versioin of DLBC and MMM method + RTA in REW I always got exactly what DLBC filter predicted. But what I know now, it is not all the picture and I must check GD. Using wavelet taught me an easy way to see what happens that compensate my lack of knowledge even after reading many times the help files in REW. ...amateur...what a shame!
I suppose that if they receive different reports they could give more priority to the possible resolution of the problem (if deemed such).
RenPa
Active Member
- Joined
- Dec 7, 2022
- Messages
- 123
- Likes
- 48
Yes I did and hopefully I am not the only one
RenPa
Active Member
- Joined
- Dec 7, 2022
- Messages
- 123
- Likes
- 48
Following all comments here, I tried 10 different filters in DLBC to see if one would stand out from the crowd. It is hard with my knowledge to decide. GD is true only for one mic position meaurement? And for one channel at a time? DLBC filter is calculated on an average of 13 measurements. Why to prioritize one GD measure against 13?
How to choose between this filter results for the Front left channel
and this other filter?
How to choose between this filter results for the Front left channel
and this other filter?
