Does an in box measurement and then correction for in box measurement eliminate this near port/near woofer merge problem?
Well, for my active filtering endeavour i use VituixCAD to get a feel for the baffle-effect (step, diffraction) on different distances
Why?
Although, it also heavily depends on the system.
Stitching doesn't work all that great anymore for a cardioid with relatively big drivers.
That's because the directivity is already pretty high at lower frequencies.
So we can only use it for the on-axis response.
Even for closed systems (or BR etc, something with a volume) it can sometimes give a bit if false impression.
Luckily simulations are quite decent at those lower frequencies.
Roughly an octave below where the diffraction starts.
Stitching doesn't work all that great anymore for a cardioid with relatively big drivers.
That's because the directivity is already pretty high at lower frequencies.
So we can only use it for the on-axis response.
Even for closed systems (or BR etc, something with a volume) it can sometimes give a bit if false impression.
Luckily simulations are quite decent at those lower frequencies.
Roughly an octave below where the diffraction starts.
Could you suggest a more accurate method for scaling the near field response between woofer and bass reflex port ?
Member
Joined 2003
I think what @DaveFred means is to take a mic-in-box measurement for the low frequency portion, instead of trying to merge individual responses of port output and woofer nearfield.
I think the main downside of mic-in-box method is that the maximum usable frequency is usually quite a bit lower than the maximum usable frequency of a nearfield measurement, so may require additional steps to get a reflection free response from 200+Hz range for merge.
Member
Joined 2003
If aligning the response levels visually, I would try to have them converge towards 0Hz, main problem is the noise floor of the measurement gets in the way making this visual process a bit difficult at times.
Member
Joined 2003
The "error" at low freq in diffraction simulation between 1m and 30m distance will be applied to the low frequency portion being merged... It is significant. Any opportunities to reduce error in data used for design should be considered, this is one that requires little effort so just do it.
Member
Joined 2003
The difference between the two responses shown here: https://www.diyaudio.com/community/threads/vituixcad.307910/post-7784443
With the goal of creating a full frequence response that represents a reflection free response in the far field listening distance, not 1000mm, the process that Kimmo has laid out with the merger tool in VituixCAD provides the most error free results for what the average user is able to achieve indoors in your average home. I am open to hear what the alternatives are to provide a better result. Hoisting the speaker 5m in the air outdoors, or use of an anechoic chamber is not a suitable solution for most people, and most anechoic chambers also have low frequency limitations as well.
I have seen some other alternatives to merging low and high frequency portions such as using a guiding filter to replace the low frequency portion as a method of removing room interaction from the response. I've also seen some "cepstrum editing" techniques that attempt to "erase" reflection information from the impulse response. I can't speak to the accuracy of these alternatives, however.
With the goal of creating a full frequence response that represents a reflection free response in the far field listening distance, not 1000mm, the process that Kimmo has laid out with the merger tool in VituixCAD provides the most error free results for what the average user is able to achieve indoors in your average home. I am open to hear what the alternatives are to provide a better result. Hoisting the speaker 5m in the air outdoors, or use of an anechoic chamber is not a suitable solution for most people, and most anechoic chambers also have low frequency limitations as well.
I have seen some other alternatives to merging low and high frequency portions such as using a guiding filter to replace the low frequency portion as a method of removing room interaction from the response. I've also seen some "cepstrum editing" techniques that attempt to "erase" reflection information from the impulse response. I can't speak to the accuracy of these alternatives, however.
Last edited:
Ehm?
The freq resp below the stitching freq is being cut off.
The other data is from the measurements.
So as I said before, totally not relevant.
What is relevant is the actual expected directivity, nothing more.
Member
Joined 2003
Error of applying 1m diffraction simulation to low frequency portion can be significant, so VituixCAD / Kimmo suggest to use 10m-30m simulation here. Error of using 1m or 30m data for high frequency portion may be small, but it is really 1 minute of work so just do it.
Method is not bad, but you should take care that vent pressure stays a bit below cone signal at LF. Woofers with phase plug have quite equal radiating area in front and back so difference can be smaller than with bigger difference (dust cap) and damped enclosure. Here is one prototype with Purifi NAA, but I don't really know is ~1 dB difference at 5 Hz correct and adequate difference or not.
The same decision could easier with passive radiators which can be adjusted using cone areas. Passives drop pressure very fast at LF due to self resonance so adjustment with levels is inaccurate...almost impossible.
I have found that Microphone-in-box method as described by Joseph D’Appolito here:
https://audioxpress.com/article/measuring-loudspeaker-low-frequency-response
gives a very good low frequency response directly without need for scaling between woofer and port.
Above 150 Hz or so (where the port is usually inactive) you can merge to the near field response of the woofer.
Then the above summed response can be merged to the Far field response taken at 1 or 1.5 meters.
https://audioxpress.com/article/measuring-loudspeaker-low-frequency-response
gives a very good low frequency response directly without need for scaling between woofer and port.
Above 150 Hz or so (where the port is usually inactive) you can merge to the near field response of the woofer.
Then the above summed response can be merged to the Far field response taken at 1 or 1.5 meters.
The baffle influence is applied to the near-field low frequency measurement part. That is not cut off, as far as I understand.
The discussion is about the error.
We get an error because we are measuring to close by, not because the simulation is that much off.
Aka, we are not in "far-field" yet (technically still not far-field, but you get what I am saying here)
I am gonna generalize things a bit (otherwise we can nitpick special cases forever), but in general the error in bafflestep simulations is quite low.
Or a much better way of saying this, is that the results from bafflestep simulations have a high predictability.
We apply this simulation to the nearfield measurements.
This introduces another error, which is how accurate the calibration of the microphone is, as well as the error in distance.
Although practically speaking, almost everyone will eyeball the levels = stitching method.
The stitching frequency is determent by how well the speaker still work as a piston (so no obvious resonances and other acoustic related issues) and we assume it's still omnidirectional at that point.
In other words, if we take an imaginary scenario where that is actually very high, we are lucky and can basically also stitch at a high frequency (let's say 800Hz to just put a number to it).
This is also the reason why I mentioned cardioid systems especially with bigger woofers.
Since they are definitely not omnidirectional anymore
The "far-field" measurements are just cut-off at this frequency.
Any data below that frequency isn't relevant anymore, because we can use the near-field+bafflestep simulation for that part.
Instead of using just an abrupt point, we can smooth both out a bit (an octave or so), just to get a nicer transition.
So again, the error (far) below this frequency is totally irrelevant, because we simply dismiss it.
It's literally not being used
We just simulate the bafflestep as if we are at like 3 meters and apply this to the nearfield measurements accordingly.
And we just pretend that the 1 meter acoustic measurements are just at 3 meters.
As even VERY clearly shown in the example, there are differences but very small.
These differences are mostly in the diffraction, not the bafflestep (NOT the same thing!!!!).
(I am talking about the bafflestep transition frequency, not the low-end!)
Which is to be expected, since with diffraction we are not talking about simple acoustics math anymore, but very complex acoustic behavior that's even so-so with FEM/BEM simulations.
Again, this is somewhat generalized, because there are plenty of examples where this bafflestep transition is much to close to the minimum frequency of our time window.
Or the shapes or ratios are so unique that the entire bafflestep isn't accurate anymore.
But in that case we've reached the limits of the stitching method to begin with.
One of the reasons why you first want to do your homework and start with bafflestep simulations.
Also the reason why 3-way systems can be tricky, because in that case you also have to have a good idea where your crossover point is gonna be.
Although, as long as we know that the woofer as well as the midrange still behave like a omnidirectional source and we know the response well (from nearfield measurements as well as driver parameters from impedance), we could basically get a pretty decent idea from the simulations.
Last edited:
I was just about to write an addendum to my post above because I realized what you explained very clearly. Thanks!
Side note to make the above more complete.
In theory we could simulate the type of source to the bafflestep simulations.
In that case it would also work for systems that don't work like a monopole.
It's only a bit hard because of how the practical world behaves.
In general it's a little bit of a theoretical discussion.
I don't agree with people that say that any type of bafflestep compensation is useless.
The reality is that people often have speakers relatively close to the wall. (20-50cm or so)
So in a sense the absolute numbers we are getting don't say that much.
In theory we could simulate the type of source to the bafflestep simulations.
In that case it would also work for systems that don't work like a monopole.
It's only a bit hard because of how the practical world behaves.
In general it's a little bit of a theoretical discussion.
I don't agree with people that say that any type of bafflestep compensation is useless.
The reality is that people often have speakers relatively close to the wall. (20-50cm or so)
So in a sense the absolute numbers we are getting don't say that much.