My advice would be to make near field and gated far field measurements of a driver. Then used the standard techniques to merge the NF and FF responses into a composite frequency response which spans from low frequency to high. This composite response is what goes into the simulations and design process.
? Bsc is baffle step correction, it is not affected by room or location!
Baffle makes also small edge diffraction effects at higher freq. They depend on baffle shape and driver location, and change when off-axis so dsp-eq is not effective. But the designer must know them and use measurement at at least 3x baffle width.
Juahzi i have not been accurate enough: for bsc frequency at which it happens is not room related ( given by smallest dimension of box's frontplate, ballpark freq: 115/ baffle width in meter) but the level needed of correction is location/ room dependent.
On a pole outside at 6meter from ground you'll need a full 6db. In room close to boundary will probably needs way less or even none, in wall will require attenuation ( of lows), any other location will be in between.... in the range 3/4db.
On a pole outside at 6meter from ground you'll need a full 6db. In room close to boundary will probably needs way less or even none, in wall will require attenuation ( of lows), any other location will be in between.... in the range 3/4db.
Last edited:
That sounds like the process Kimmo Saunisto describes in his VituixCAD measurement instructions. In which case, you wouldn't flatten the drivers +/- one octave of the passband, correct?
I wonder if there is benefit in flattening the drivers +/- one octave before using the VituixCAD method when using DSP (rather than passive crossovers).
Specifically within the context of the method you described in post #6, are you talking about baffle step compensation or are you moving on to room curve measurements at this point; what I mean is, you're no longer describing the crossover process but how you deal with room curve when using DSP, correct?
When I design an active DSP filter, I use VituixCAD to convert impulse response to frequency response, and to merge near field to far field... exactly as Kimmo advises. Then I flatten the drivers for an octave beyond the crossover point(s). Then I apply idealized crossover filters, usually LR2 or LR4. There is nothing incompatible about using a merged NF/FF response and also applying good basic DSP filter techniques.
I should add that when I have completed the filter as I described above, this is not the end of the process. It is really just the starting point. There is still the process of optimizing all of the various responses; the on-axis, listening window, early reflections, power response, predicted in room response, and the directivity index... And then there is the subjective evaluation and final subjective adjustments (i.e. the voicing).
There is an alternative to flattening the drivers for an octave beyond the passband. We could establish a target curve and then EQ/Filter the driver to match the target curve. VituixCad has this feature in the optimizer function. The target curve for a midrange driver for example might be an 89 dB flat curve, with a 300 Hz LR2 high pass and a 3000 Hz LR4 low pass. We then apply the various active filter elements so as to match the target.
I have done it both ways, and I get the same result acoustically, but the second approach sometimes uses fewer biquads. However I find the second approach more time consuming and more difficult to fine tune in the subjective voicing stage. It is a personal preference, and one method is not really superior to another.
Good luck with your project !
J.
The 'flatening' +/-1 octave is to ease the filter implementation of 'textbook' filters: by having a margin on which behavior of freq response is already known is often enough ( for 4 pole and steeper).
I'm using mostly fir xover ( 48db/octave) and with this +/- 1octave i'm sure the acoustic response is 'textbook' up to ~ -50db and so 'perfect' on the range that mater.
It's not mandatory to follow this: with IIR it's rare to see 'textbook' filters implemented, there is variation over the electrical function to achieve a xover in the 'ballpark' of your target for an acoustic target -which is the important one ( eg a textbook 2pole butterworth can be modified by the use of eq to achieve a 3order butt acoustical (resulting from your 'electrical filter curve'+ eq + driver acoustic behavior).
I hope it's clearer.
Room curve. Well it's usually the last step i implement ( as well as tweaking/ tune/voicing). Bsc is just before as it is location dependent, so the level needed have to be defined when location is choosen.
In a way it is part of the 'in house' curve...
As the inverse to most i don't rely on simulation with the loudspeakers i played with until now : either commercial units i tweaked either experiments to validate some hypothesis, try arrangements...
I don't need the accuracy Virtuix ( or lspcad i've used the most) cad gives. I find this tools extremely useful to learn things, retro engineer design and pure theorical approach, and there is numerous example how good loudspeakers made with them are but for me atm i've got no real use for a full simulation as does Jim or others members.
So i'm basic about my needs. But one thing is sure the more your good at interpreting your measurements the easier it gets to have meaningfull corrections applied.
I'm using mostly fir xover ( 48db/octave) and with this +/- 1octave i'm sure the acoustic response is 'textbook' up to ~ -50db and so 'perfect' on the range that mater.
It's not mandatory to follow this: with IIR it's rare to see 'textbook' filters implemented, there is variation over the electrical function to achieve a xover in the 'ballpark' of your target for an acoustic target -which is the important one ( eg a textbook 2pole butterworth can be modified by the use of eq to achieve a 3order butt acoustical (resulting from your 'electrical filter curve'+ eq + driver acoustic behavior).
I hope it's clearer.
Room curve. Well it's usually the last step i implement ( as well as tweaking/ tune/voicing). Bsc is just before as it is location dependent, so the level needed have to be defined when location is choosen.
In a way it is part of the 'in house' curve...
As the inverse to most i don't rely on simulation with the loudspeakers i played with until now : either commercial units i tweaked either experiments to validate some hypothesis, try arrangements...
I don't need the accuracy Virtuix ( or lspcad i've used the most) cad gives. I find this tools extremely useful to learn things, retro engineer design and pure theorical approach, and there is numerous example how good loudspeakers made with them are but for me atm i've got no real use for a full simulation as does Jim or others members.
So i'm basic about my needs. But one thing is sure the more your good at interpreting your measurements the easier it gets to have meaningfull corrections applied.
Last edited:
Regarding baffle step compensation: When I merge FF and NF responses together, I include the baffle diffraction response in the NF measurement. VituixCad has a feature which allows us to model the baffle diffraction based on its size and shape, or we can use a more generic "spherical" baffle diffraction response. By doing this, when I flatten the driver's response, I am automatically applying the full 6 dB of BSC.
The method is the same for both. Therefore saying you flatten the driver first, is the same as saying you make a passive crossover according to one response curve. It isn't bad but neither is it ideal.. meaning it is you who decide how far you will go.
Any EQ you do before and keep throughout the process will make no difference in the end.
This is what I did in my current test project:
Listening: too bright, which is expected as SPL level is fairly linear from 40Hz to 20kHz and there is this valley between 1kHz and 2kHz.
Back to Rew:
Listening: still too bright, but better
Now, if I take the autoeq predicted from rew for each driver, import them in vituixcad and apply the same xover filters, export the total frequency response from vituixcad into REW again, and autoeq the whole FR against toole curve, now I have a perfect FR.
I guess, this is what Dirac would do but as of now, I cannot import these optimized biquads back in the minidsp.
I guess it's easier to saw wood
- near field measurement of the three drivers.
- import them in Rew, and use autoeq on each +/- 1 octave with an auto calculated target SPL.
- import the biquads in minidsp, apply gain correction so calculated target SPL all matches the same level.
- apply LR4 filters : low pass, band pass, high pass.
Listening: too bright, which is expected as SPL level is fairly linear from 40Hz to 20kHz and there is this valley between 1kHz and 2kHz.
Back to Rew:
- use autoeq against a Olive Toole curve for each driver.
- import the biquads in minidsp, apply gain correction so calculated target SPL all matches the same level.
- apply LR4 filters : low pass, band pass, high pass.
Listening: still too bright, but better
Now, if I take the autoeq predicted from rew for each driver, import them in vituixcad and apply the same xover filters, export the total frequency response from vituixcad into REW again, and autoeq the whole FR against toole curve, now I have a perfect FR.
I guess, this is what Dirac would do but as of now, I cannot import these optimized biquads back in the minidsp.
I guess it's easier to saw wood