Disturbing time alignment results

[Dahlberg]

First of, I have prepared a filter where I have compensated for latency caused by the acquired Eq's I need for my dipole speakers. The process of linearizing drivers at a 250cm distance was thrown in the thrash. I do get pretty good results from the basic Eq process so I dont see the need for another solution regarding that and room correction will take of the remaning fine tuning afterwards.
So, I have recently done some measurements ment for time alignment and now I'm confused about the results. The subwoofers where rotated by -4000 samples and the midbass drivers by -2000 samples before saving a multichannel file for measurements. So for the left channel, the pulses turned up at 8400 and 9961samples, for the right channel at 8387 and 9963 samples. Both tweeters at 12000 samples.
After rotating the drivers back by 4000 and 2000 samples the drivers appeared to be located at ~40 samples in front and ~400 samples behind the ribbon tweeters. That would indicate a distance between the subwoofers and the panels of 286cm's and the midbass drivers would be located 30cm's closer to the microphone. This is all from using the markers in the graf to translate samples into distance.
In real life the tweeters are closest to the microphone followed by midbass at an additional distance of 7cm's and subwoofers at ~70cm's away from that. All drivers have a positive polarity and I'm using linear phase UB jpol11 first order filters.
Except from that the difference between the Right and Left subwofer doesnt make sense either. It doesnt look like much in the graf but the difference in distance to the microphone appears to be 9 cm's. In real life it's within +/- 1mm.

[UliBru]

Bengt,

welcome to the troubles of time alignment
It is quite comfortable to take the positions of the max. peaks (e.g. subwoofer 8400 and 8387). But a closer look reveals a different picture. Both curves are pretty similar at the beginning but at the peaks the curves are different. The left curve has a higher amplitude. This means that the output signals of the subwoofers are different. So even if the timing is perfect identical you get a seemingly different peak position.

And indeed we must keep in mind that the real speaker drivers add their own minimumphase behaviour (and thus delay) to the linearphase crossovers. Thus they may introduce their own delay in combination with their own "gain" function. And especially with low frequencies we have to face an additional problem. A reflected soundwave can change the measured curve and this can even happen before the speaker driver has reached the full peak introduced by the crossover).

That's the reason why Acourate does not promise a full automatic driver alignment. IMHO an automatic driver alignment is just a fake alignment.

So the rotation of the crossovers is a trick and it gives already good results. But it is necessary to be very careful as the conclusions can be wrong.
You will find that it is very difficult.
It makes sense to load the according crossover in another curve and to properly rotate it to the nominal position. Then you can compare the curve shapes to learn more about the differences (apply TD-Functions - Gain to adjust the XO curve height).

Beside the rotation "trick" you can also try to follow the ideas of sinewave convolution described by https://www.audiovero.de/freedownload/T ... lution.pdf

You will learn again that it can be VERY mind-boggling to solve the time-alignment problem. The reality can challenge us so much.

Beside the sinewave convolution you can also similarily try the convolution with windowed chirps.

[Dahlberg]

So I went "old school" using the delay function in AC. At first I inverted the midbass drivers and started from 0ms going upward to find the lowest amplitude at 70hz (crossover between sub and midbass). Delay was done to the midbass drivers to move them closer to the subs, measured with the spl meter in REW. Ended up at 11,25ms for 70 and 9,5ms for 350hz. After polarity switch amplitudes was back to where they were supposed to be, but the numbers are way to high and the tweeters are still at the wrong position. So I figured I'll keep the positive polarity and start over again.

Results:


The distance between sub and midbass is now down to 1/3 and the difference between ribbon drivers and midbass is 0,25ms. Ribbons still on the wrong side and even though there is just 4ms left between subs and midbass, 70 cm's is supposed to be 2ms.

Amplitude is identical at crossover frequencies between the different approaches after polarity switch.

Edit: So according to this it should be more or less correct to in Acourate invert the midbass drivers and rotate the ribbon drivers by 180 and the midbass drivers by 192 samples ?

[UliBru]

Bengt,

when you use linearphase XOs in Acourate the driver pulse responses should all be positive. So inversion of the midbass driver should not happen.

I know about the "old school" approach. You have to be aware that all you do is an alignment which produces the most cancellation with the inverted driver. This simply means that the steady state of the test tone has got the cancellation by summing opposite phases.
In reality the steady state is especially for low frequencies influenced by reflections. A reflection can shift the steady state to another time.
So after returning the inverted driver back from inversion the frequency response (steady state) shows a maximum level.
But this does not mean that the time is properly aligned. Only the phases are aligned after some time = when the drivers play the sine wave in steady state.

You can test this the following way:
E.g. 350 Hz = cycle time of 2.857 ms = 137 samples
So you can now rotate the midbass or tweeter by +-137 samples without a change in playback level (REW spl meter).

Now music is not steady state. Usually you do not listen to tones with a long time duration.

[Dahlberg]

This can also be seen in the 350hz table where the 1ms is almost as low in amplitude as the 3,75 ms one. That one was on the other hand so far away from the 70hz test, and since I couldn't figure that one out I disregarded it instead.
So all positive polarity and give it another try then.
How about just using a measuring tape and finetune until stepresponse and group delay looks decent enough ?

As I figure it, it shouldn't be that far of even though measurements say something else.

[UliBru]

Measuring with a tape can be quite misleading. As you cannot measure the acoustical centre of a driver.
Have you read the sinewave convolution paper? You should give it a try.

BTW I use the quickest driver = tweeter always as timing reference. The measurements do not lie.

[Dahlberg]

So if I'm not mistaken this is what has to be done, at least before evolving to sine wave convolution ? If I place the subs 400 samples ahead of the tweeters that will compensate for the difference. Same with the midbass drivers even though they will end up on the "wrong side" ?

[UliBru]

The analysis by rotated XOs has shown that the bass is 400 samples too late and the mid is 40 samples too early.
Compensating XOs thus will play the bass 400 samples earlier and the mid 40 samples later.
So in total the peaks bass, mid and tweeter peaks will match at the same position.

The only remaining question is if the peaks are really comparable to the ideal peaks of the linearphase XOs.

[Dahlberg]

Well I will create a project and do some measurements, again I will also read up on sine wave convolution.

[Dahlberg]

This is a comparison of the EQ'd xo pulse and the measured one. It seem that the pulse peak is distinkt enough. It was rotated to fit in the same position and some gain was added to the xo file for better visibility.


Edit: The pulse is for the sub