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

I got that from your description which is why I said the response was similar to an array. Compare your long ribbon impulse to my 25 driver impulse from a few pages back similar behaviour. Look at the first millisecond. There will be a path length difference from the ends of the ribbon Vs centre to the mic just like there are in an array. And when you look you have the same unsettled first millisecond.
 
Thanks Fluid. Here is the ribbon tweeter zoomed in.

I calculated the +/- 2.5 foot distance from the mic tangent which comes to @ +/- 4" or 1/3rd of a foot (first circle) which is shorter than the 8/10ths of a ms ripple.

I assume the additional ripples (second circle) must be the floor ceiling line source reflections ???

Thanks for the info. I see some of transferred onto the tweeters STEP.

QSpAdzj.jpg


I got that from your description which is why I said the response was similar to an array. Compare your long ribbon impulse to my 25 driver impulse from a few pages back similar behaviour. Look at the first millisecond. There will be a path length difference from the ends of the ribbon Vs centre to the mic just like there are in an array. And when you look you have the same unsettled first millisecond.
 
I would be surprised if anything less than a millisecond was a reflection rather than the speaker, 1ms is a path length difference of about 30cm.

The portion of the impulse in the left circle is quite different to the right which seems a more defined ripple.

You can try a short gate and look at the frequency response, this is dominated by high frequencies anyway.

In your earlier graphs the first reflection looked to be between 3 and 4ms.
 
I will investigate further.

I also noticed the waveforms of the content of the 2 circles to be different but didn't know what to make of it. I appreciate being able to better interpret the plots.

I use the 1ms~=1ft rule of thumb. 30cm is 11.8 inches.

The 4ms/4ft reflection is most likely the leather couch which I should probably move back (or throw a quilt over) when taking measurements.

I would be surprised if anything less than a millisecond was a reflection rather than the speaker, 1ms is a path length difference of about 30cm.

The portion of the impulse in the left circle is quite different to the right which seems a more defined ripple.

You can try a short gate and look at the frequency response, this is dominated by high frequencies anyway.

In your earlier graphs the first reflection looked to be between 3 and 4ms.
 
I use the 1ms~=1ft rule of thumb. 30cm is 11.8 inches.
Seems odd to quibble over a slightly inaccurate rule of thumb division.

For us metric users 3ms per metre is the ball park figure. Neither is totally accurate.

I hope you took the intended point that unless you have a reflective surface in that area you are looking at speaker response in the sub 1ms region.
 
Seems odd to quibble over a slightly inaccurate rule of thumb division.

For us metric users 3ms per metre is the ball park figure. Neither is totally accurate.

I hope you took the intended point that unless you have a reflective surface in that area you are looking at speaker response in the sub 1ms region.

Sorry Fluid, I was not quibbling or disagreeing at all. Just stating I don't normally think/dream/work in the metric system and none of my tapes are metric but do have a metric machinist ruler, feeler gauges, wrenches and sockets, none of which would work for measuring the 2-channel room distances.

1) I confirmed the 4ms ripple was a leather couch reflection.
2) I confirmed the 8ms ripple was a reflection off the back wall.
3) I tried covering the top and the bottom of the front of the ribbon to make it a quasi-point-source. It made some minor magnitude changes to the 1st ms of the plot confirming your assertions.
4) I tried padding the floor in front of the ribbon and it made no difference confirming it wasn't the floor's line source extension. I didn't test the flat ceiling reflection.

The Ribbon Tweeter is a standard Magnepan Ribbon.

295maggie.promo_.jpg


As for the mic, it is on a standard mic stand, pointing straight at the speaker with a 0 degree correction file. No special shock absorbed mount, just a pressure fit clamp.
 
Last edited:
Sorry Fluid, I was not quibbling or disagreeing at all. Just stating I don't normally think/dream/work in the metric system and none of my tapes are metric but do have a metric machinist ruler, feeler gauges, wrenches and sockets, none of which would work for measuring the 2-channel room distances.


As for the mic, it is on a standard mic stand, pointing straight at the speaker with a 0 degree correction file. No special shock absorbed mount, just a pressure fit clamp.

No problem seems I misunderstood your intention. I did wonder which tape measure you had to mark out 11.8 inches ;)

Troels graveson has a good document showing mic ripple putting a tube over the back can make quite a difference.
 
No problem seems I misunderstood your intention. I did wonder which tape measure you had to mark out 11.8 inches ;)

Troels graveson has a good document showing mic ripple putting a tube over the back can make quite a difference.

The 11.8 inches came from dividing 30cm by 2.54. I wanted to see how far off my 1 foot rule of thumb was to your 30cm.

2.54 and 25.4 are my metric conversion aces in the hole.
 
Member
Joined 2002
Paid Member
Hi, i tried 100x more and trying to understand things.
- Still this howto : Dropbox - REW_rePhase_tuto.pdf - Simplify your life

I will try to explain step by step how i followed the howto, hoping someone can take a look at it ? :

REW :
1. Take vector average of both L & R channel.
2. 1/6 octave & IR window smooth them and then make rePhase target EQ file.xml
3. Disable IR window(FDW) and export L & R vector averaged measurement as .txt
--- From now only explain 1 channel (L)
4. Import L vector average to rePhase
5. Apply EQ filter (.xml)
6. Generate Left 48Khz 32 LPCM mono impulse
7. Import Left 48Khz 32 LPCM mono impulse into REW
8. Disable all IR windows to be sure and do Left vector average * Left 48Khz impulse
9. Add -offset (I did -105dB) to make this around 72dB
10. Create excess phase (again be sure all IR windows are disabled)
11. Save Excess phase as .txt
12. Reset settings in rePhase (or restart it) and load measurement Excess phase Left
13. Load target EQ from REW.
14. Create filter for 44100, 88200, 96000, 172400, 192000 etc... all .bin or whatever to be used.

I read many different explanations of this, but a lot is still unclear to me.
I'am very much in doubt of if i do it right in rePhase, as i only load the Excess phase of channel LEFT and apply my target EQ to this, and then generate the filter.

Thanks in advance dsp folk's :)

Jesper.

Following in Jesper's footsteps!

1).
Here's my 18 measurements. Nine for each channel as outlined in SwissBear's tutorial.
attachment.php


2.) I manually aligned the impulses of all the individual measurement using the
'Set T=0 at cursor' feature with the cursor at the center of the highest peak on each impulse.
attachment.php


3.) I then vector aligned the nine measurements for each channel and applied 1/6 smoothing and FDW of 15 cycles. Left Vector average shown.
attachment.php


4.) Just for information I've included here the impulse of the Left Vector Average.
attachment.php


5.)A screen capture of the Auto EQ of my Left Vector average.
attachment.php


6.) A screen capture of the filters generated by the Auto EQ.
attachment.php


7.) The filters are saved for each channel in .xml format for use in Rephase.

8.) Windowing (FDW) is removed from each vector average but the 1/6 smoothing is left intact before exporting each as text for use in Rephase.

Part II soon!

Regards,
Dan
 

Attachments

  • Pic1.jpg
    Pic1.jpg
    96.1 KB · Views: 667
  • Pic6.jpg
    Pic6.jpg
    169.2 KB · Views: 531
  • Pic5.jpg
    Pic5.jpg
    183 KB · Views: 531
  • Pic4.jpg
    Pic4.jpg
    84.1 KB · Views: 549
  • Pic3.jpg
    Pic3.jpg
    124.4 KB · Views: 559
  • Pic2.jpg
    Pic2.jpg
    90.8 KB · Views: 602
8.) Windowing (FDW) is removed from each vector average but the 1/6 smoothing is left intact before exporting each as text for use in Rephase.

No 8 doesn't make much sense.

Also a flat in room response is unlikely to be the best final target. A speaker with flat anechoic on axis response has a downtilted response when measured in a room.

I realise that you have been following the tutorial quite closely and their graphs show a flat response but that is something that is probably better to deviate from.

Add a tilt in the REW EQ section target response. The LF and HF sections switch at 200Hz so you can create a linear tilt from any frequency or have it hinge a little at 200Hz.

Anywhere from 6 to 9dB slope from 100Hz to 20K is a reasonable ball park figure, which comes down to personal preference.
 
dantwomey :)

The "guide" i wrote as you refer to, was not ment as i guide, just a how i do it, and what do i do wrong for someone to help me correct it. Belive me im a totally newbie going at the steep step's right now at the learning curve :eek:

fluid corrects my no. 10 btw...

Originally by fluid ::
At point 10 when exporting the excess phase it should be Frequency Dependant Windowed otherwise you will get a significant amount of room response in the phase trace and will be trying to micro manage it with rephase.

1. Take vector average of both L & R channel.
2. 1/6 octave & IR window smooth them and then make rePhase target EQ file.xml
3. Disable IR window(FDW) and export L & R vector averaged measurement as .txt
--- From now only explain 1 channel (L)
4. Import L vector average to rePhase
5. Apply EQ filter (.xml)
6. Generate Left 48Khz 32 LPCM mono impulse
7. Import Left 48Khz 32 LPCM mono impulse into REW
8. Disable all IR windows to be sure and do Left vector average * Left 48Khz impulse
9. Add -offset (I did -105dB) to make this around 72dB
10. Create excess phase (Corrected to with FDW window)
11. Save Excess phase as .txt
12. Reset settings in rePhase (or restart it) and load measurement Excess phase Left
13. Load target EQ from REW.
14. Create filter for 44100, 88200, 96000, 172400, 192000 etc... all .bin or whatever to be used.

Also i have learned later that i measure with an umik usb 48khz mic. so i am not at all sure that filters could be created in rePhase other than 48khz.
I am quite sure i have to resample my 48khz'a measurements into say 44100, 88200 & 96000 before making filters in rePhase. And this i also have my doubts about. I think it could be better to resample all measurements and then load them into REW and make time allign, average etc..

I was recommended to use Audacity for resampling.

Hope i do not make to much confussion for you ... sry if i do!
Jesper.
 
Also i have learned later that i measure with an umik usb 48khz mic. so i am not at all sure that filters could be created in rePhase other than 48khz.
I am quite sure i have to resample my 48khz'a measurements into say 44100, 88200 & 96000 before making filters in rePhase. And this i also have my doubts about. I think it could be better to resample all measurements and then load them into REW and make time allign, average etc...
The filters can be created in rephase at any sample rate you like, the REW generated xml files don't rely on the sample rate and the text measurements imported into rephase don't rely on the sample rate either. The only time it might present a problem is if you were trying to use a filter that was higher than the half sampling frequency of the lowest rate you will use.

It is DRC that needs the sample rate to be correct as it works directly on the impulse. Easy to get confused :)
 
Member
Joined 2002
Paid Member
No 8 doesn't make much sense.

Also a flat in room response is unlikely to be the best final target. A speaker with flat anechoic on axis response has a downtilted response when measured in a room.

I realise that you have been following the tutorial quite closely and their graphs show a flat response but that is something that is probably better to deviate from.

Add a tilt in the REW EQ section target response. The LF and HF sections switch at 200Hz so you can create a linear tilt from any frequency or have it hinge a little at 200Hz.

Anywhere from 6 to 9dB slope from 100Hz to 20K is a reasonable ball park figure, which comes down to personal preference.

Better?
attachment.php

attachment.php


Regards,
Dan
 

Attachments

  • Pic7.jpg
    Pic7.jpg
    434.9 KB · Views: 600
  • Pic8.jpg
    Pic8.jpg
    450 KB · Views: 488
Most likely but only you can decide what your preference will be :)

That sort of target that resembles a BK curve with a slight bass boost will probably be preferred if listening loud. When I try it I like it more the louder I listen.

I don't really like to listen that loud most of the time so I tend towards something that is a more linear slope.

To get that change the HF Fall start to 200Hz. I also like a little bit steeper slope between 100 and 300 that can change things quite a lot with only 0.5dB difference. I also see this same trend in a lot of spinorama measurement of good loudspeakers.

It's only a few clicks to generate the different filters and hear for yourself.
 
Member
Joined 2002
Paid Member
Part II
Left Channel only

Left Vector Average imported into Rephase with 1/6 smoothing but no FDW.
attachment.php


Left Vector Average with REW filters applied and generating Impulse.
attachment.php


View of the Impulse generated by Rephase.
attachment.php


Regards,
Dan
 

Attachments

  • Pic9.jpg
    Pic9.jpg
    478.3 KB · Views: 468
  • Pic10.jpg
    Pic10.jpg
    492 KB · Views: 442
  • Pic11.jpg
    Pic11.jpg
    437 KB · Views: 543