Of course you can simulate this, I just mentioned I proposed a method to Jeff to prepare the PCD to run these sims. Pretty straight forward.
Thanks for running these results, very helpful for the discussion.
The default of placing the t= 0 point at the peak is not satisfactory. Obviously, the transient starts when the impulse begins to rise which is always before the peak unless the system frequency response is flat to infinite frequency. If the system has a low pass response (which all speaker and electronic systems do), then the lower the cut of frequency the slower the initial rise and the longer the time till the peak is reached. I believe HOLM has an option called Causal. That is a little more realistic, but any of this auto placement is a little iffy because there is always a little noise in the impulse before the actual start. In theory, causal should remove all excess delay and if the system is minimum phase, the resulting FFt of the impulse should yield the minimum phase response. The locking of the offset is the right thing to do if you want the measurement relative to a fixed distance and you keep the mic distance constant. But again it depends on repeatability and how you set up for measurements. I tested the locked offset feature and found that it did not always work correctly. That is, on a loop test I took the initial measurement and then lock the offset. The phase of repeated measurements of the loop test should then be identical. Unfortunately they were not always the same. I presume this has to do with what PC the code is running on, what the sound card is and what the latency is..... I do know that when I do that in Sound Easy, if I loop test and hold the start of the fft window constant (lock the offset) the phase comes out exactly the same every time. I think this is because SE references T = 0 to the start of the reference pulse since it is a 2 channel acquisition system.
I have looked at HOLM briefly. It's a nice piece of software but I think there are some options that it would be better off not having. (Like setting T = 0 to the impulse peak).
There are lot of good measurement software packages around. Learn how to use one correctly and go with it, which ever one you like best. I like SOundEasy basically because I can do all my measurements and save them directly into driver files so I don't have to mess around with importing data. ARTA has some very nice display feratures. HOLM is reprotedly easy to use. (They all seem pretty easy to me.)
With HOLM, I only use the auto-find-time-0 option to get it close to the peak. After that I lock it down (there's a second check box you need to enable to keep the latency the same from one measurement to another), measure all the drivers with the mic in the same spot, set the 0 point to just before the first impulse (which would be from the nearest driver), and then generate frequency response for all drivers using the same 0 point.
IMO the auto find time-0 thing is more of a convenience feature. It's handy if you just want to look at FR curves (say from a full system measurement), but you need to lock it if you're measuring drivers for crossover work.
Agreed
IMO the auto find time-0 thing is more of a convenience feature. It's handy if you just want to look at FR curves (say from a full system measurement), but you need to lock it if you're measuring drivers for crossover work.
Agreed
With HOLM, I only use the auto-find-time-0 option to get it close to the peak. After that I lock it down (there's a second check box you need to enable to keep the latency the same from one measurement to another), measure all the drivers with the mic in the same spot, set the 0 point to just before the first impulse (which would be from the nearest driver), and then generate frequency response for all drivers using the same 0 point.
IMO the auto find time-0 thing is more of a convenience feature. It's handy if you just want to look at FR curves (say from a full system measurement), but you need to lock it if you're measuring drivers for crossover work.
Agreed
Did you select "casual response" or "First Peak"?
And here the Dutch forum that hosted the discussion (in Dutch)
Scheidingsfilter technieken - Hear.nl Forums
I am afraid Jeroen removed most pictures from his web space.
Yes, no, maybe. Since phase is arbitrary it's not particuarly important where the reference point is so long one has an adequate understanding of where it's placed and control over its placement. In HOLMImpulse it's generally preferable to derive the lock point from the tweeter as that impulse has the sharpest peak but, really, anything will work. The point is getting good data requires knowing the tools and their limitations.
If you're satisfied with SoundEasy, that's great. My personal experience is SoundEasy v16's MLS results are noticeably lower quality than HOLMImpulse's. I've never been able to get either a fast or slow sweep to complete successfully in SoundEasy whereas HOLMImpulse has always executed the sweeps properly. One and two channel measurement approaches are subject to different errors, so I wouldn't necessarily argue one is superior to the other. For example, two channel measurements are only robust against variations in audio interfaces if both channels experience the same latency shift. My general practice is to repeat all measurements in order to check for consistency. As my experience with HOLMImpuse is it's considerably more reliable and stable than SoundEasy, it's much quicker for me to collect data, duplicate or otherwise, in HOLMImpulse.
Rightly or wrongly, I came away from the from the initial description with the impression multiple drivers are measured on axis in this workflow. In which case measurements which take account of the path length differences to different drivers are useful as well. Particularly if you want to use the drivers as a phased array to control the speaker's radiation direction.
Forgot to mention I generally work with linear phase crossovers; controlling time alignment through slope isn't an option in that case.
Ah yes, makes sense. Thanks John. I wondered why HOLM kept placing the peak a tiny bit after the 0 point. Now I know why.
I agree that it may be soundcard and/or driver dependent. I checked and double checked my USB sound card - mic preamp on the time lock. It was always the same. And always the same on sweeps or MLS, which is handy.
Just today I managed to come up with a good crossover for the Altec A7-500.
Using the locked time really helped see where phase is. By spreading the frequencies some, phase came right into line. It sums well, too
FYI. Low pass = 600Hz Butterworth 3rd. HP= 840Hz Butterworth 2nd. Horn pushed back 5" from the front. Gives an acoustic 3rd order at ~725Hz.
Not really audible.
It doesn't really matter. One will place the t = 0 (let's call it the origin) point at the peak, one will place it right where the impulse starts. In almost every case, I then move it back by a few samples to be sure that I'm picking up the full impulse. It adds extra phase rotation to the frequency response curves, but it does that to every measurement, so it doesn't matter since all we care about is relative phase differences between the drivers.
The key things to ensure (IMO):
* Make sure the origin is before the impulses for all the drivers
* Make sure you measure them all with the mic in the same place, and with the origin locked
So I usually lock the origin to one of the drivers, measure all the drivers, then move the origin so it's just before the start of the earliest impulse, then measure all the drivers again.
At least, that's my understanding of how to use the tool, I could be completely wrong
Modeling Crossovers, Listen before you build speakers!
I came across this paper a while back. These guys did blind listening tests on various crossover architectures using some math models with Matlab and headphones. From their descriptions I was able to write a similar script for Octave. The crux was higher order crossovers produce a narrow notch if there is a difference in distance between the drivers as happens off axis.
AES Convention Paper 5908: A Virtual Loudspeaker Model to Enable Real-Time Listening Tests in Examining the Audibility of High-Order Crossover Networks. Cochenour & Rich
I assume the same could be done with a line level crossover by summing the outputs and feeding the results to a headphone amp.
I came across this paper a while back. These guys did blind listening tests on various crossover architectures using some math models with Matlab and headphones. From their descriptions I was able to write a similar script for Octave. The crux was higher order crossovers produce a narrow notch if there is a difference in distance between the drivers as happens off axis.
AES Convention Paper 5908: A Virtual Loudspeaker Model to Enable Real-Time Listening Tests in Examining the Audibility of High-Order Crossover Networks. Cochenour & Rich
I assume the same could be done with a line level crossover by summing the outputs and feeding the results to a headphone amp.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.