rePhase, loudspeaker phase linearization, EQ and FIR filtering tool

[Barleywater]

Originally Posted by pos

The longer the measurement signal the more resolution you get in the low frequency.
2^16 should give good results. I have add some strange problems with the 2^14 setting (looks like a bug).
(of course you can also change the length of the sweep by changing the start frequency, but I assume this is not what you are talking about).
I usually use MLS for these measurements, as I have not found any advantage for the sweep (no need for THD or high noise rejection here) and it is easier to the ear...

MLS measurements produce lower S/N, in part this is due to MLS signal being broadband signal at all points in time. Maximal slewing occurs, exercising generation of IMD that forms much of the apparent noise floor seen in MLS results. Additionally, high pass filtering of electronics leads to aliasing artifacts in results. Room reverberation time greatly extends length of IR representing room response, and likewise aliasing artifacts are concern.

To accommodate this I work with sweeps at least 2^17, often 2^18, sometimes ever bigger to check. To get really good results with subs, sweep should start at <10Hz.

S/N isn't so important when synthesizing filters based on IR, but all becomes critical with direct inversion techniques.

Originally Posted by Pano

What I heard (I think). A difference.
For the first few tracks, I didn't like flattened phase. It sounded looser and not as natural as the normal phase of the BR box. The normal (not flat) phase seemed to give a tighter, cleaner bass. It seemed more focused than the flattened phase. I.E. slightly, subtlety, different, but not an improvement. The flattened phase seemed slightly less focused, more diffused, looser.

Then I started to notice the mids sounded different, not as prominent or edgy. You might not think that a bandwidth limited to just under 4K could sound edgy, but it did and was a bit annoying. With the phase flattened at the bottom, the mids tended to shift backward, away from me and were easier to listen to for long periods. The midrange was the area where there seemed to be the biggest change. Bass did seem wider and more stereophonic, too. On some symphonic pieces, bass notes that wrapped all the way around my head with unaltered phase went distinctly only as far as left and right ears with the flattened phase. Odd!

Another odd effect was that even tho the bass seemed wider and more stereophonic with the flattened phase, bass instruments were easier to locate - more stable - and easier to follow in the musical line.

Going back to the full system - lows, mids, highs and crossovers - revealed pretty much the same effects, sometimes more pronounced. With the flattened phase, some midrange sounds - like voices, strings, etc. - tended to move back into the space. The whole space became deeper and wider and ambient clues of the recording space were more obvious. I was also able to listen louder than before. At some point I just left the convolver in and enjoyed the music. I stayed up listening until 3AM, it was so much fun hearing the better soundstage and depth. The low bass still seemed looser with flattened phase, but everything else seemed better focused and open at the same time.

That all seems pretty remarkable and rather hard to believe from a simple phase flattening at the low end of the frequency range. I'll continue to listen to find out if it's all just my imagination.

Pano, seems lid may be off of Pandora's box. As I've said before, with phase and amplitude correction done well, the difference is night and day.

From perspective that most of directional cues and timbrel identification come from harmonics, it seems that when these tell brain that low frequency source is located at X, but low frequency decode of location X doesn't match, realism is reduced. Higher frequency cues decode in brain before low frequency cues.

Square waves <500Hz band passed to remove content above 3.5kHz, still look remarkably like square wave. 100Hz square wave with only harmonics to 1kHz has six components and also looks like square wave with smoothed corners and ripple:

sq100 lp1k.gif


When 37Hz 4th order Butterworth high pass filter is applied to above:

sq100 lp1k hp37 but4 min.gif

Group delay of each harmonic changes waveform, changes sound.

[tomtom]

Thanks for all answer

This is my system Beyond the Ariel - crossover is but 48 LP to but 12 HP at 450 hz

i can even think about what it doing with phase. It measure flat at xo in every angle - no lobing. I can measure gated to ca 200hz. So it can be possible to measure and correct phase deviation. But so far measurement is very dependent on sweep length. Very long sweep produce completely nonsense IR.
My plan is use DCX only to XO and make all notches and phase correction in rephase /as proposed/. Theoreticaly is possible to measure phase in electric domain on dcx out but im sure that acoustical side is quite different.

Thank you one more time for great program

[pos]

Oh yes I remember this wonderful system!!

Using asymetrical electrical slopes is okay as long as you end up with fully complemantary acoustical slopes for both magnitude and phase as a result.
Phase coherency between the two crossed-over drivers is a must.
A good acoustical target would be symetrical Linkwitz-Riley crossover.

To obtain that you need to measure your drivers without crossover, using only a protection cap for the compression for example (that you will leave inplace afterward) and/or using the integrated linear phase filters in HOLM (applied directly to the measurement signal).
Then you need to apply the proper amount of EQ and filtering to get a "perfect" LR slope at least one octave after the crossover point, for both magnitude and phase.
One way to do that is to use asymetrcial slopes (plus a little bit of EQ...), another way is to flatten the driver in and around its bandpass unsing minimum phase EQ (the one you have in the DCX), and then apply LR eletrical slopes that will result in identical acoustical slopes.
One problem with that technique is for HP filters: when using a horn the phase will turn rapidely down low, and flattening the magnitude even 1 or 2 octaves below the target crossover point will not completly compensate for the phase shifts that will occure down low in the passband. To compensate them you would need to EQ magnitude to flat even lower than that...
This is exactly like a Linkwitz transform. It is not allways practical as it can typicaly require EQ in the realm of +60dB one decade below the crossover point... Though it *can* be done with the floatting point engine of the DCX (stacking EQs...), as the subsequent electrical filter will compensate this EQ anyway...

[tomtom]

Thanks for kind words.

Unfortunately i need this asymmetrical XO /i mean acoustic/ layout to get good powerresponse without lobing. The Lambdas twin must be crossed very steep, otherwise they response go too high and start to produce lobes /CTC will be too high/ On other side horn had slightly different radiating pattern than Lambda twins and shallow slope on horn make this transition softer /the acoustical slope is of course steeper than 12db/oct/. I dont know exactly on which HP slope i end with horn /ca 3-4 order/ but it sum with lambdas quite nice. I know there are filters that are asymmetrical and sum correctly /+- some delay my math behind this is lousy/.

My plan was measure system as is /amplitude wise is correct/ and correct only the final acoustic phase sum. Drivers are mounted quasi coaxial so i think it can be possible. /im thinking at this as single driver with phase defect/

[pos]

Your system

Having asymetrical slopes is okay, but you absolutely need complementarity of the magnitude (flat summation) and full coherency of the phase through the crossover.
But of course what happen off axis can be quite different...

If you were doing the filtering in FIR you could try the Horbach-Keele slopes that are specially intented for MTM arrangements like yours.
Why not buy a pair of openDRC?

(no need for FIR filtering for the subs: you can do the HP in FIR and the the LP in IIR, and just use delay to linearize the phase of the sub.... exactly like HOLM does when detecting the offset of a low-passed driver and making its phase shift disapear...)

What compression drivers are you using?

Here is an example of an horbach-keele filter in rephase:

[more10]

I have followed this thread with great interest. I would like to try to build a four way DSP based crossover using convolution and rePhase impulses for the all front loaded horn system I am building. The low bass (30-80Hz) is constructed, I am now building the hypex upper bass (55-400Hz) to be used from 80 to about 300 Hz. Later I will build tractrix horns for 300-2500 and 2500-20k, to be used with compression drivers.

I cannot say I fully understand this technology, therefore I ask your advice before buying any equipment. Will this work?

It's very difficult NOT to have a minimum phase speaker at LF though some back & front loaded horns try very hard.

This worries me a bit. The low bass i have build rotates about 50 degrees between 30 Hz and 80 Hz. Is this a problem?

[pos]

Hi more10

I have been following your construction thread the lansingheritage forum!
What hardware are you planning to use for your crossover?

[more10]

I am going to use an Analog Devices ADSP-BF609 EZ-Board hooking up cheap DACs using I2S.

[pos]

That looks like some serious DSP power in there!
How are you going to implement the convolutions?

[more10]

Blackfin DSP library (-lbfdsp), the convolve_fr16 function.