Hi
Because it was my case until recently , perhaps you don't have understood the notion of "farfield" of a loudspeaker .
When you apply filtering to your speakers , the sum of the responses of the different speakers wil be flat on axis of the tweeter only in the farfield . Within this distance there will be deeps at crossover region . These dips depend on distance of measurement , order of filter , frequency of transition and gap ( ctc spacing ) between the speakers on the baffle .
On a classic baffle of a loudspeaker the axis of tweeter , mid and woofer are parallel . If you want to time align these speakers you have to measure the drivers separetely in their axis ; than apply the delays . One solution is to place the loudspeaker on the side at the edge of a table ; Place the mic face to tweeter on axis as best as you can , and then don't move the mic : move the box along the edge of the table in order to place mid , and then woofer in axis of the mic . You can put a marker on the table and others on the edge of your box to improve the process . The mic should be at a distance of 4 time the diameter of the woofer . And the frequency range for the woofer should start at 80 or 100 Hz ( REW's performance will be best to detect IR start ) . That's how i succeeded to perform a four way LR24 .
FWIW, I've typically used step or impulse response of the raw drivers to check if estimated acoustic centers are actually correct (tweeter at, or slightly behind, woofer timing). Zoom in on the start and look where the response starts to come out of the noise floor, that's where the acoustic center is, for the highest frequencies where proper alignment is more critical than at lower frequencies.
All responses must have the same timing reference, obviously. I sometimes used a helper tweeter nearby to create a marker pulse present in all recordings. Don't move neither mic nor helper tweeter when changing setup for a different driver.
Ha, more like the joy (or frustration) working with optimisers to fit the real response to the target.
Usually a two-step process (for analog designs), first you curve-fit the required analytical filter blocks to the target function, and then you optimize component values of the actual circuit (set up from educated guesses/experience) to fit to the resulting correction target obtained in the first step.
Yes me too, ...... IIR impulse responses initial rises from zero need to be time aligned.
Below to help illustrate to others: electrical LR24's @ 600Hz. Red high. Green low.
Top panel shows the two sections' impulse responses in time domain.
Note how the initial rises start together. (100ms was added to both sections to move impulse starts away from zero, to let rises be seen easier)
Which gives flat frequency mag, a phase wrap at xover freq, and a group delay of about 0.75ms group between sections. (blue)
Ok, trying to directly answer your question.... by using example started above.
(Oh, first I should say I'm assuming interferometry is the same as the null method...probably should have made sure before i take off with this.)
Time alignment should be nothing more than distance between acoustic centers. Which for the electrical summation above, is of course zero.
A problem with the null method, is that if group delay becomes part of the fixed time alignment, a null will still appear but it will usually not be as deep a null as when group delay is not part of the time alignment.
Here's the null from inverting the low section of the above, with zero delay in place.
Note is goes below the charts -24dB range. (down to about -50dB on expanded scale)
Now, here's the high section delayed by the group delay 0.75ms. (an often suggest way to achieve fixed time alignments..)
Note the null is only about -18dB.
Also note with the group delay inserted as fixed delay, that the impulses are now inverted, so it requires a polarity inversions of one section.
Here's low section inverted. The mag scale is zoomed in to show the ripple.
Ripple is not terrible , but it's also not needed.
If you toy around with timing further, to the sample level, taking delay to 0.83ms (vs what's been used) will produce a null as deep as zero delay.
However, ripple get's small bit worse.
Ok, point of all this...IF you've targeted acoustic LR24 xovers, there should be no fixed delay between driver sections other than distance between acoustic centers.
With LR24's, tell tale sign that group delay is worming it's way into fixed delay, is a need for a polarity inversion in one on the driver sections.
So erik, my take is ..given you needed a polarity inversion on the mid...yes, group delay has snuck into your fixed delay settings
Which can throw intuitive physical distance into not making full sense vs measurements.
I've never delved this far into looking at how using group delay as fixed delay effects response, as it became clear a while back it's suboptimal at best.
One thing i noticed working up this post, is how closely group delay between the low and the high flat areas, matches the time difference between the low section's impulse rise and peak.
It's almost like using group delay used to time align IIR, is very close to aligning impulse peaks...which we know is a no-no for IIR.
Anyway, hope this made sense / helped.
Last edited:
@eriksquires: It would help if you could tell us your measurement program (ARTA, REW, Omnimic, etc.) , whether you are using single or dual channel measurements and your crossover simulator program. Also a description of your measurement process (mic position, etc.) including whether you use minimum phase measurements.would help.
Exactly.Now, here's the high section delayed by the group delay 0.75ms. (an often suggest way to achieve fixed time alignments..)
Note the null is only about -18dB.
I never understood why most software tries to align IR peaks which is fundamentally wrong. It should align the starting points.
Hi, interesting... How do you find linear scale more helpful?
I rarely look at it, other than it helps me to identify comb filtering.
(I'm far from being a mathematician...wish I was one...but those days are long gone.
I'm mainly measurement based... the sims I posted .....I end up verifying them electrically with DSP and dual channel before I trust them Lol)
It's a bit maddening isn't it?
Took me a while to learn to align impulse starts with IIR given how most software often works, (even with help of some kind folks here on DIY.)
Took me even longer to learn to align impulse peaks with linear-phase.
I had to stick with phase in the frequency domain for time alignment, until i learned that.
I often wonder how impulse alignments (and crossovers in general) ..are generally so obscure....
Anyway, a couple more examples for folks, to support impulse alignments. For any interested...
Here's same electrical 3-way, both IIR and FIR linear phase, with LR24 crossovers at 300Hz and 3000Hz
First is IIR showing need to align impulse starts.
Second is linear-phase FIR showing need to align impulse peaks.
For those interested in precise multi-way time alignment, one thing I can guarantee from working with time alignments with both IIR and linear-phase....
Impulse peaks are a hell of a lot easier to nail down precisely, than impulse starts are.
The lower the frequency of the passband, the greater the relative ease/accuracy of peaks vs starts.
D
Deleted member 375592
I think you get it as good as it gets. Congratulations!
... and thank you for posting the question. It helped me to understand the details of what I was missing.
... and thank you for posting the question. It helped me to understand the details of what I was missing.
Thank you for some enlightening posts above.
I had a reread of the original LR article https://www.linkwitzlab.com/JAES/jaes_papers76.htm. It is really well written and actually describes the accoustic center and choice of filter Q and N really well.
And topping it off by reading the classic https://www.ranecommercial.com/legacy/note160.html makes it even clearer
I had a reread of the original LR article https://www.linkwitzlab.com/JAES/jaes_papers76.htm. It is really well written and actually describes the accoustic center and choice of filter Q and N really well.
And topping it off by reading the classic https://www.ranecommercial.com/legacy/note160.html makes it even clearer
Sorry, it's just shorthand for a process many use when you don't have the ability to measure the start of the signal and arrival rate.
For a pair of drivers (mid + tweet or mid + woofer) You measure the individual response of each driver, and input them into a simulator like XSim or VituixCAD. You also measure the combined response. For all of these measurements you do so in cabinet without a crossover.
In the simulator, plot the simulated response plot and actual response plot. Because the real sum has acoustic delay but the simulated sum does not the two response plots won't look very similar, especially in the areas where the drivers overlap.
In the simulation you then the delay of the driver that is further away until the simulated response curves equals the measured combined response. When this matches exactly, like a key in a lock, you know you have calculated the correct acoustic offset.
This works because the combined output is a complicated signature of the constructive and destructive sum of the drivers which is so unique that it's easy to tell when your simulation has matched up with reality and when it has not.
Obviously there are some safety concerns with tweeters which need to be addressed. A large value cap may need to be used in measurements and simulations to protect them.
Last edited:
I was lucky that this is what I assumed. In my case, OmniMic was hard to read the exact values, so I started off by adding excess delay to one driver, which made it easy to tell two drivers apart on the impulse chart. Then using that delta as a starting point I inverted one of the drivers and nudged the delay for maximum null. I definitely felt more like a hobbyist and like I was doing it half-baked, but the end results were correct so I will just avoid bragging about technique and focuse on the results in this case. 😆
Thats a really neat method for finding offset!
Do you keep those measurements for simulating or do you make infinit baffle measurements and model the box and speaker element placement?
handiest is to use two channel measurement with correct procedure so the delays are included in the data and there is no need to calculate anything other than make the crossover, with great flexibility. Basically follow VituixCAD measurement manual. Thisbstuff can be done in lultiple ways, so what ever feels most logical might be just fine, at least until there is some error somewhere.
This method only provides relative offsets so it will only work in a specific baffle, and only when comparing the distance of A vs. B.