How can I adjust the result of the driver linearization

[Mister Cool]

Hello,

I tried to linearize my tweeters using procedure described by Mitch Barnett in his book. I went through all the steps FDW, phase extraction, inversion, convolution, cut'n window, normalize and I was successfull . To successfull The driver measured later on in the listening position without further correction through Macro 2-4 (target curve), what I tried to achieve, was flat in the whole range, but... "to flat" -> I missed the typical smooth high frequence roll-off, which I was used from applying apriopriate Target Curve.
I thoought, I need a method, which decreases the strength of the linearization and adjusts the result towards the preferred Target Curve. I had a crazy idea to use the FD-Function Amplitude Average to create "somethink in between". I used further on the procedure described above, but between the step Cut'n Window and Normalize I added another steps to create an average of the result of Cut'n Window and the regular, flat XO. I checked the minphase box off. Finally I have normalized the outcome.
The new "manipulated" linearized XO looks a little bit flatter and indead the following Log Sweep shows the frequency responce which is less "flattisch" and closer to my wished Target Curve -> with HF roll-off

long speech short

Is there a better way to control the linearization process, which would help to adjust the result of the linearization towards defined target curve, instead of flat frequency response (e.g. convolve with Target Curve, average with Target Curve or what ever...)

The background/idea behind this proceeding is: I try to correct the direct signal/wave only (what happens when I do linearization measurement in near field) instead of correcting the sum of direct wave and reflections, which is measured in the listening position.

I get corrected direct wave and and uncorrected room reflections ("ambience")

I am not saying and I am not sure, if the result of this procedure is better then the usual way through correcting the signal in the full frequency range using Target Curve (unlimited range in Macro 3). I want simply to compare this two approaches and to hear the difference

Thanks in advance for your ideas
Alwin

[Mister Cool]

The longer I think about it, the creazier the ideas

I try to understand, what happens during such a "linearization" process, and what are the commonalities, parallels to other optimization techniques availble in this great toolbox Acourate.

here are my "thoughts" (please correct, if I am wrong)

1. I think e.g. about Beamforming measurement. We use it to focus our meassurements on the direct sound and to supress the impact/share of the reflected sound, which varies with the listening position. But this is also (among others), what happens, when we make our "linearization" measurement in the near field of the driver -> we focus on the direct sound and reduce/suppress the reflected sound.

My conclusion -> both techniques allow to minimize/suppress the impact of the reflected sound on our measurements

2. Both processes: "linearization of the driver step by step" and "running the automated Macros 1-4" give a simmilar possibility to equalize the frequency response to a "linearized" flat response. The first one through the step "inversion", and the second one through the Macro 3 (I pressume, I use a flat line as a Target Curve). So this is the "commonality". And what is the difference?

- The "linearization" uses as base the direct sound measurement and the result/outcome contains only the amplitude domain information/correction (EQs). No time domain correction is included (needed)

- The "automated Macro 1-4" process uses as a base the sum of the direct and reflected sound and the result contains corrections in amplitude and in time domain

Long speach short

If I would like:

- to achieve the (I call it in this way) "target-driven" linearization results but in an automated way (similar with running Macro 1-4)
- to focus on direct sound
- to correct the signal without time domain corrections (in amplitude domain only)

then I could run the Macro 1-4 procedure using nearfield meassurement (log sweep recording in nearfield for each driver) and then "remove" the time domain corrections from the created correction files.
This would be "somehow" equivalent to lineaarized XOs from the first process.

Does it make a sense?
How can I remove the "time domain corrections"? Using TD function "Phase Extraction"?


"There’s more than one way to skin a cat"


Alwin

[UliBru]

Hi Alwin,

there is always the wish to optimize a loudspeaker independent of the room. So there are several approaches but they all may have weak points.
Examples for such measurements are nearfield or close-up measurements, beamforming measurements, open-air measurements, reflection-free chambers.

Close-up measurements may not get aware about edge diffrations or influences horn geometries. With dipole speakers the overall sound also depends on the sound radiated from the back side. I have made bad experiences with nearfield measurements of dipoles.
Some speakers use a principle which also takes care about the room. Think about corner woofers. It makes no sense to measure them in an acho chamber as there is no "corner".

Even with a perfect linearized speaker the sound will differ in different rooms. And so a measurement at the listening position has also its advantages. It is worth to learn about the final natural slope of the higher frequencies as you can get some imagination what the correction is doing at the end.

So we all have to find our way to get a most satisfying result.

[Mister Cool]

Hello Uli,
thank you for your response.
Yes, I am aware about the variaty of different measurement approaches (their pros/cons) and different speaker design conceps (closed, BR, horn, open baffle). And because I am curious to learn and to hear them all, I build them all!!!
I am almost through with all of them and now I am "stacked" with big Open Baffles in U-Frame supported with 4 Subs in DBA setup



These are definitely the "wrong speaker" for the "near field measurement" and generaly I am definitely in favor of corrections based on measurements at the listening position, but nevertheless this doesn't stops me to try, to experiment, to learn and to compare other approaches.
Therefor it would be helpfull for me to know in this context "How can I remove the "time domain corrections"? Using TD function "Phase Extraction"?"

Thanks,
Alwin

PS
I found your great idea with VBA -> that's my next research project

[UliBru]

Room - Special: Convert filters to minphase

You have to be careful in case of linearphase crossovers as the conversion also changes the XO type to minphase.
So it would be necessaey to use minimumphase XOs from the beginning.

[Mister Cool]

only
You have to be careful in case of linearphase crossovers as the conversion also changes the XO type to minphase.
So it would be necessaey to use minimumphase XOs from the beginning.

Thank You. I use NT crosovers 4. order, so I have to consider it.


Here is an picture, which shows what I try to achieve / test

At the moment only Sub (XO1) Midwoofer (XO3) and Tweeter (XO4) are active
Because of the Open Baffle the only driver worth to be "linearized" is the Tweeter (AMT with closed back)

And here are the measurements
red -> log sweep measurement at the listening place using the regular nonlinearized XOs for all drivers
green -> log sweep measurement at the listening place using the regular nonlinearized XOs for the Sub / Midwofer and "half-way" linearized XO4 for the Tweeter
brown -> my target curve

Underneath are the respective XOs (including the "half-way" linearized XO4



For the "daily use" I make measurement based on nonlinearized XOs, and correct the whole frequency range 20-23.000Hz

For the experiment I make measurement based on the "half-way" linearized XO4. Then the Tweeter measurement above 2.300Hz follows very close the target, and I decrease the correction range in Macro 3 accordingly (As-Is)

The goal of the experiment is to "linearize" the Tweeter in this way, that it follows closely the target in the whole range (>1.400Hz) and does not require any further correction of the Tweeter -> Macro 3 range is limited to 1.400Hz (To-Be)

PS.

I presume that, when I use linear crossovers (NT) from the beginning, then there is no need to convert my "adjusted-linearized" XO to minphase

[PhilActiveAudio]

Just to share my experience / testings.

I also followed Mitch linearization method for the 3 drivers: tweeter (a Fostex compression driver), medium (a Radian compression driver + Arai horn) and woofer (a Supravox driver in a Jensen vented box). I measured each driver at 40 cm distance. I measured the room with optimized XO at 350 cm distance to loudspeaker.

The result after 1) linearization, 2) time alignment and 3) macro 1 to 4 is not good compared to 1) time alignment and 2) macro 1 to 4 with almost equivalent parameters. The scene is flatten and less detailed.

I suppose that drivers measurements were done too close to drivers. They do not take into account horn or box effect.
I will measure drivers at 80 or 100 cm and perform a new test in coming weeks.