rePhase, a loudspeaker phase linearization, EQ and FIR filtering tool

TNT

Member
Joined 2003
Paid Member
I think I have never heard that term. Where did you see it? Bandwidth yes, but not phase.... but thinking about it, the corresponding meaning would be the frequency range where the phase is within a certain +/- degree.

When FR deviates, so will the phase. FR and phase are married for eternity (in a minimum phase system).

Your XO system, is it based on IIR or FIR? FIR being the only power that has been wested to separate phase and FR :)

//
 
Can can say rephase is winning in car audio comps …. 2nd 1st place with for rephase filters

Whoop whoop
B9DB7462-5D40-427D-8F84-044E20C02240.jpeg
 
  • Like
Reactions: 2 users
@StigErik

Nice job man… yeah that looks great


I also use minidsp96k and it takes a little creativity, but yes you can get an outstanding results with very limited taps. They give you like exactly enough to get it done…. Barley…

Very nice to see someone else take ahold of it and show it who’s boss :p

That looks great…. Is that the flex or flex8

And just curious what your impulse centering used ended up being and did you use a window of sorts on your fir?

Thanks
 
Disabled Account
Joined 2008
45 degrees off-axis. Bass is exaggerated because the mic is closer to the front subwoofer array. And the phase shifts at higher frequencies because the mic will be farther away from the mid and tweeter. The speaker system is a Magnepan 3.7 with the tweeter on the outside. There is also two Magnepan 1.6 acting as bass panels.

Off-axis.png
 
  • Like
Reactions: 1 user
So here is a video I made on a tune I quickly put together for a video

So it wasn’t refined at all so keep that in mind…

My objective with this video is to show how easy it is to get a fantastic correction in RePhase , but most importantly using a dual channel fft analyzer for measurements…

REW I’m sorry just doesn’t cut the cake.. sure you can get your measurements to line up in REW all day long , there wrong though.

There’s nothing worse than importing bad data into RePhase. That is the worst and I did it for years…

The problem with REW is the vector averages are horrible!! It puts dips where there are none. Furthermore it’s the only way to get a phase average which is necessary to get good results

In my video I demonstrate that the data in smaart is exact… easily to capture , and the results are fantastic and sound great!

With smaart I do believe it puts RePhase at the very top of Fir convolution programs as it’s not automated which we all know leads to errors

I understand why pos won’t go automated now… this is so much better…. It’s dead on balls accurate and the video shows it’s accuracy…

I know some of the heavy hitters like @fluid @mark100 @wesayso and a few others Byrtt and John and the OG @pos will have there methods… I just hope my method helps show someone how easy it is to make a FIR in rephase and helps someone

Enjoy, (excuse my nerdy voice I hate doing videos) and please excuse the dip at 450hz , I forgot to enable a crossover and had a linearizarion on another downstream fir, so it created a phase issue that I later fixed….

 
  • Like
Reactions: 1 user
@JohnPM

Cross correlation??? Idk what that means, but if it’s anything like the magnitude response of a db average with phase , then I’m sure it works much much better!!!

Thank you , yes I have a friend I have been remote tuning his car and he has REW filters I will give it a look!

The vector averages I followed the Swiss bear method to the T and it was catastrophe in a car… I think the reflections made the vector issue worse…. It was unusable….

So thank you I’m very exciting try it…. I’ll see if I can download it tonite…

I have the pro version of REW and have done mult mic captures and I love how it shows the crest with the amplitude… would be really cool if I could setup a multi mic session and hit measure once and get a average with phase … like it could automatically set IR (REW auto IR shift works fantastic btw!) shift it and give a average

I would even be happy with a few steps as long as it can be done I would be thrilled

Yes I’ll give it a look 🥳🥳🥳

Andrew
 
@JohnPM

Cross correlation??? Idk what that means, but if it’s anything like the magnitude response of a db average with phase , then I’m sure it works much much better!!!

Thank you , yes I have a friend I have been remote tuning his car and he has REW filters I will give it a look!

The vector averages I followed the Swiss bear method to the T and it was catastrophe in a car… I think the reflections made the vector issue worse…. It was unusable….

So thank you I’m very exciting try it…. I’ll see if I can download it tonite…

I have the pro version of REW and have done mult mic captures and I love how it shows the crest with the amplitude… would be really cool if I could setup a multi mic session and hit measure once and get a average with phase … like it could automatically set IR (REW auto IR shift works fantastic btw!) shift it and give a average

I would even be happy with a few steps as long as it can be done I would be thrilled

Yes I’ll give it a look 🥳🥳🥳
 
You might find this page helpful in understanding the different types of averages that different software can produce.

https://www.sounddesignlive.com/know-your-audio-analyzer-averages/

This is a comment not a criticism, in your video you are altering the delay of the mics in real-time until they become as similar as possible. The coherence ends up being very good up to 2k. This will give you the cleanest trace when averaged with complex data because you have taken out the major phase differences. If you did the same thing with other software the result would be more similar. Smaart also has coherence and power averaging options which will weight the peaks more than dips and avoid what often happens with vector averaging where dips take over.

I don’t advocate for any particular method other than the one that works best for you in your situation. I do think that the idea of correcting phase through an average has some fundamental problems though.
 
  • Like
Reactions: 1 user
Thanks @fluid

I know your comment isn’t criticism… even if it was negative feedback is the best feedback ;-) (as long as it doesn’t have any element of sarcasm its completely valid)

Yeah this method works and sounds good and the data I get works great , when I measure as you saw (as long as I don’t move the mics lol) is exact and perfect…

No matter how hard i tried in REW I could never get that to happen. The remeasure after a correction always looked like I only did a part of the correction…

I’m sure there’s a way to format the data with smoothing and such to make it better , believe me I tried and it make me feel like I was a bad tuner and not doing something right…

I bought an array of 600$ earthworks mics because I thought it was the mics that were inconsistent… and it was REW the whole time… I can use cheap o mics now and get great results.

So yeah I believe there are many way to average… but it comes down to the data for me…. It has to be clean and averaged to sound good in a car.. that much I know to be true.

As far as the alignment on my mics, yes the find works then manually adjust the time until there all on top to about 1k

Then they drift apart, it’s not possible to align them past that because they have different types of reflections and things have been modulated differently at every other inch.. so it will never be exact.

The whole point tho is the average of them will get you closer then not having anything at all…

Arbitrarily, I wouldn’t do a phase correction above 1k as I can’t garuntee the results, as long as the microphones are around the headspace where you listen, and the seat never moves, and you sit in the exact same spot every time (boy that sure sounds a lot like a car) the phase average will get everything above 1k to be closer then not using anything…to a point…

Coherence sure sounds good and the upper midrange and Hf center information stays dead center so it doesn’t hurt to do it as where the phase was before was a mess anyway because our heads are bigger then most those frequencies…

So my point , it’s not destructive to phase to 20k… it does make it better , and it’s impossible to be perfect (maybe if your head was the size of an ants head a single mic phase to 20k would be exacting and perfect)

I am interested in other methods of averaging, altho I just don’t see the benefits unless maybe it was designed for a binural mic that uses a dummy head. Then some HRTF curves and averages could potentially get you a little closer, but at wavelengths barley an inch long, even then I can’t see how adding a few hundred degrees of phase makes any difference at 10k….

I know my correction that high up isn’t doing anything, I also know it’s not hurting anything either. But it sure looks pretty on the screen :p


I’m very very happy with smaart… it was so much worth the 400$ bucks… I fought REW for years trying to do what smaart can do in seconds…

I read the paper u sent , yes I will look at more as better is always better :)
 
Last edited:
Cross correlation??? Idk what that means
It means that REW will compare the impulse responses of the individual measurements and adjust them so that they all line up as well as possible with the first measurement in the set chosen. They can then be vector averaged with good results. It is a little like an automated form of the manual delay finder tweaking you mentioned in your video.
 
  • Thank You
Reactions: 1 user
I have a Linear Phase width question if someone would be so gracious to answer.

Using 64-bit IEEE floats, the roll offs appear to extend down to over -300dB, but the corresponding valid liner phase width appears to only extend down a third of that.

Should the valid linear phase width be the same as the valid roll off width on a 64-bit band pass filter ?

Thanks much.

ScreenCap.png
 
Last edited:
It means that REW will compare the impulse responses of the individual measurements and adjust them so that they all line up as well as possible with the first measurement in the set chosen. They can then be vector averaged with good results. It is a little like an automated form of the manual delay finder tweaking you mentioned in your video.
Does this automate the process of manual alignment of the impulse responses of different measurements?
If yes, it's not doing what Smaart or Eclipse Software Averager is doing with the power/complex averages.


Blue is with CrossCorrelation, Green with manual alignment of impulse responses and red imported from Averager. I only had a trial of Smaart, but from what i remember the power average from Smaart was within +-1dB of the Averager program.
1684996669711.png

Phase averages are different, too.
1684996786509.png

So for Oabeio i would still recommend to do the averages in Smaart as they better align with what's actually happening in a car.
 
  • Like
Reactions: 1 user