Welcome, Guest
Username: Password: Remember me
  • Page:
  • 1
  • 2

TOPIC: rePhase 1.0.0

rePhase 1.0.0 5 years 1 month ago #20115

  • pos
  • pos's Avatar
  • Offline
  • Platinum Boarder
  • Posts: 350
  • Thank you received: 196
  • Karma: 39
rePhase 1.0.0 is out: sourceforge.net/projects/rephase/
1.0.0 2015-06-25
  * New functionalities:
    - Albrecht cosine windows implementation
      Ref: A Family of Cosine-Sum Windows for High-Resolution Measurements
    - multiple memory slots in range settings to be able to quickly go from
      one view to the other and focus on different aspects of the response
      curves
      These slots are preset with (hopefully) useful values but can be
      manually modified and copied.
    - "Load Settings From Clipboard" and "Save Settings To Clipboard" menu
      entries in order to be able to easily share corrections on web forums
    - frequency marker for the last correction point (5 seconds persistence)
    - fader values can now be manually edited to arbitrary values
  * Bug corrections:
    - bug correction in Minimum-Phase Filters tab: the polarity of low-pass
      Linkwitz-Riley filters of order (2*n+1)*2 was reversed
      (eg 12dB/oct, 36dB/oct, 60dB/oct, 84dB/oct, etc.)
      A warning will be emitted when loading correction files from prior
      versions using an odd number of such filters, as the polarity will now
      be correct and reversed compared to the prior bogus correction.
    - bug correction with higher than normal noise floor with even order
      taps (introduced in version 0.9.9 while solving a similar problem
      for odd order number of taps!)
    - bug correction with txt output file with 0.000(...)0 values
      (especially pregnant when using Hann window)
    - correction of the bogus flat top window implementation
    - corner case instabilities corrections (undue octal conversions on some
      value entries)
  * Adjustements:
    - set "32 float txt" as the default output format instead of "32bit
      LPCM wav" in order to avoid  rising the result noise floor because of
      the fixed point format
    - added de-empahasis and pre-emphasis presets in the Paragraphic EQ tab
    - added Linkwitz-Riley linearization orders 11th to 16th (why not?) 
    - reduce default phase EQ range to +/- 45° (was +/- 90°) and removed
      unpractical ranges
    - increase default EQ range to +/- 12dB (was +/- 6dB)
    - added 384 and 352.8kHz sampling rates as drop menu options for ease
      of use (any other value can still be manually entered)
    - got rid of the "Curves" tab for the time being, waiting for the
      capture functiuonality to be implemented in some future version...
https//wavetracing.com | rephase.org
The administrator has disabled public write access.
The following user(s) said Thank You: devteam, dreite, curryman, micfre, samoht, dimike27, Richard, Thunderstruck, sly

rePhase 1.0.0 5 years 1 month ago #20123

  • devteam
  • devteam's Avatar
  • Online
  • Administrator
  • Posts: 8750
  • Thank you received: 1125
  • Karma: 162
Goodness, that's a lot of modifications done on this new update! Don't you guys have a summer in France? ;-)

Great work as usual, I'm sure that a whole lot of DIYers and FIR lovers are forever grateful of all your hard work!

Devteam
miniDSP, building a DSP community one board at a time.
The administrator has disabled public write access.

rePhase 1.0.0 5 years 1 month ago #20126

  • pos
  • pos's Avatar
  • Offline
  • Platinum Boarder
  • Posts: 350
  • Thank you received: 196
  • Karma: 39
Hi Devteam
Most of these modifications have been ready for a long time, but I wanted to push some more new features before the actual 1.0.0 launch.
Unfortunately there was two serious bugs I had to correct ASAP so I removed some beta features and released what should be a "stable" version with some new features :)...
https//wavetracing.com | rephase.org
The administrator has disabled public write access.
The following user(s) said Thank You: Thunderstruck

rePhase 1.0.0 5 years 1 month ago #20192

  • Thunderstruck
  • Thunderstruck's Avatar
  • Offline
  • Senior Boarder
  • Posts: 59
  • Thank you received: 8
  • Karma: 5
Dear POS,

Thank you so much for rephase. I can't live without it.
I can easily understand what to do and how to do it.
I've learned not to be too aggressive in the low Hz ranges.
I upgraded to 3 miniSHARCs so that I could get 6144 taps to do my two lowest channels.
I have a 20 to 63 Hz channel for my 2 18" subs and a 63 to 200 Hz for my 5 12" subs.
I am doing LR8 crossovers and with a Linkwitz-Transform to boost my sealed 18" subs and some flattening of the freq response as well as linear phase box roll off adjustment and linear phase LR8, the 2048 taps at 48kHz just wouldn't match.
I even had trouble matching the downhill low cut side of my 63 to 200 Hz range.
LR4 was much easier to match with the 2048 taps but I wanted it all so I bought 3 miniSHARCs and I run my 6-way mono system at 48kHz on the 20-63, 63-200, 200-632 and 632-2000 all with 6144 taps each.
I run the top two channels, 4 tweeters and a super tweeter, 2000-6324 and 6324-20000, with the 96kHz 4x8 plugin.
FIR filtering the 2000-6324 range with 2048 taps at 96kHz is easy.
The 6324-20000 range gets the remaining 1024 taps at 96kHz but it matches well.
The other 6 channels of the 4x8 96kHz Plugin are not used.
Can you believe that you only get 2048 taps on one channel and ~1024 on another and all the taps are used up on the 96 kHz 4x8 Plugin.
Really you need twice as many taps at 96 kHz as you do at 48kHz.
I went with the higher 96kHz sample rate for my tweeters and super tweeters because rephase seemed to match LR8 crossovers and linear phase adjustments with just 2048 and 1024 taps.
I felt like my highs were a little harsh when I processed them at 48kHz and played my system loudly.
I'm going to download the new software tomorrow now that I see your announcement. I'll let you know what I think about it. I use it exclusively.

POS. What do you think about my strategy? Any comments would be appreciated.
Last Edit: 5 years 1 month ago by Thunderstruck.
The administrator has disabled public write access.

rePhase 1.0.0 5 years 1 month ago #20212

  • pos
  • pos's Avatar
  • Offline
  • Platinum Boarder
  • Posts: 350
  • Thank you received: 196
  • Karma: 39
Hi Thunderstruck
That is an interesting and impressive system!
https//wavetracing.com | rephase.org
The administrator has disabled public write access.
The following user(s) said Thank You: Thunderstruck

rePhase 1.0.0 5 years 1 month ago #20294

  • samoht
  • samoht's Avatar
  • Offline
  • Expert Boarder
  • Posts: 111
  • Thank you received: 9
  • Karma: 1
Hi pos,

thank you so much for updating and enhancing your wonderful rephase tool. It's really a pleasure to work with it.
I'm currently trying to linearize the phase of the filters of my 3-way speakers, a remake of the GrImm LS1.
With the help of my miniDSP 4x10 I've already linearized the amplitude of all drivers at least 1 octave below and above the operating frequency range.
The lower frequencies have been extended by using using Linkwitz transforms.
In order to understand the effects of rephase in detail I built a test environment consisting of my measuring equipment with ARTA Software, the miniDSP 4x10 and the OpenDRC with rephase.
Assuming that all drivers are perfect in frequency response and equal in SPL, I set up just the LP and HP filters using LR24 at the cut off frequencies at 100 Hz and 1200 Hz. A HP LR12 at 24 Hz has been used to simulate the roll off of the sealed box woofer. In reality the roll off frequency of the woofer has been extended from 39 Hz Qtc=0.72 down to 24 Hz Qtc=0.7 by a Linkwitz transform filter.
The signal of the combined 3 Outputs of the miniDSP shows a perfect flat frequenvcy resonpe and a continuously decreasing group delay starting with [email protected] Hz, [email protected] Hz, [email protected] kHz and [email protected] kHz.
Using the same cut off frequencies and type of filters under the "Filters Linearization" tab for "crossover" and "box" provides a very good linearized group delay starting with [email protected] Hz, [email protected] and [email protected] kHz. I'm absolutely happy with this result.
My questions are:
1. Can the frequency range below 100 Hz be optimized even better?
2. Is it correct to replace the behaviour of the woofer with Linkwitz transform just by a HP at the transformed cut off frequency, or has a Linkwitz transform filter to be linearized in a different way?

Best regards and thank again

Thomas

PS: I attached a photo of a collection of my speakers. The ones at the right and left side are the Grimm remakes.
Attachments:
Last Edit: 5 years 1 month ago by samoht.
The administrator has disabled public write access.

rePhase 1.0.0 5 years 1 month ago #20295

  • pos
  • pos's Avatar
  • Offline
  • Platinum Boarder
  • Posts: 350
  • Thank you received: 196
  • Karma: 39
Hi Thomas

Stunning work on the LS1 clones, kudos!samoht wrote:
1. Can the frequency range below 100 Hz be optimized even better?
Yes of course, but then you will have to make sure your measurements are accurate.
It is often easier to look at the phase behavior rather than the group delay.
You can load a magnitude/phase measurement into rePhase and directly address the phase shifts from there.
2. Is it correct to replace the behaviour of the woofer with Linkwitz transform just by a HP at the transformed cut off frequency, or has a Linkwitz transform filter to be linearized in a different way?
Yes, you want to correct the final acoustical HP of the system, so in the case of a (perfect) LT that is going to be the target HP.
https//wavetracing.com | rephase.org
The administrator has disabled public write access.
The following user(s) said Thank You: samoht

rePhase 1.0.0 5 years 4 weeks ago #20393

  • samoht
  • samoht's Avatar
  • Offline
  • Expert Boarder
  • Posts: 111
  • Thank you received: 9
  • Karma: 1
Dear All,


for those of you who are interested in my findings here you see how rephase linearizes the group delay.

The first graphic shows amplitude and phase response of my test filters, consisting of 2 LR24 crossovers and one HPF representing the roll off of the subwoofer.
The second graphic shows the group delay caused by the above mentioned filters and the nice flat group delay after rephase does done ist work.
A really great tool and easy to handle.

Best regards

Thomas
Attachments:
Last Edit: 5 years 4 weeks ago by samoht.
The administrator has disabled public write access.

rePhase 1.0.0 5 years 3 weeks ago #20422

  • samoht
  • samoht's Avatar
  • Offline
  • Expert Boarder
  • Posts: 111
  • Thank you received: 9
  • Karma: 1
Dear Pos,

i did some further testing and found a problem for which i like to get an explanation.
It looks like that the linearization of the box roll-off frequency changes the amplitude in such a way that the frequency resonse shows a 6dB/oct. instead of 12 dB/oct.
Please see the attached screen shot. It starts at 10 Hz. The lower curve shows the roll-off of a box at F = 24 Hz with Qtc = 0,7, which is a butterworth characteristics. So I used a butterworth 12dB/oct in my test setting to simulate the roll-off.
In rehase I used the same figures to linearize the roll-off, i.e. closed Q=0,707 at 24 Hz in the box tab. But this setting leads to a frequency response as shown by the upper curve. It reduces the slope while the phase linearization works well.
Did i made a mistake in my assumption or what went wrong?

Best regards

Thomas
Attachments:
Last Edit: 5 years 3 weeks ago by samoht.
The administrator has disabled public write access.

rePhase 1.0.0 5 years 3 weeks ago #20428

  • pos
  • pos's Avatar
  • Offline
  • Platinum Boarder
  • Posts: 350
  • Thank you received: 196
  • Karma: 39
Hi Thomas,

This would probably be better to open a dedicated thread for your issue, as the minidsp team was kind enough to dedicate an entire sub forum to rephase :)

Anyway, back to your problem.
How many taps and what sampling frequency did you use?
How did you do the convolution process?
What did you measure exactly (output of a crossover unit?)

Can you share your rephase setting file? ("save to clipboard" and paste here between code markups)

If you do not use enough taps for a given correction and sampling frequency, then the magnitude response can indeed be affected (ie linear distortion).
You should be able to monitor this in the rephase interface: once generated the result curve (red) should track the target curve (blue).
https//wavetracing.com | rephase.org
The administrator has disabled public write access.

rePhase 1.0.0 5 years 3 weeks ago #20436

  • samoht
  • samoht's Avatar
  • Offline
  • Expert Boarder
  • Posts: 111
  • Thank you received: 9
  • Karma: 1
Hi Pos,

thanks for the quick response.
Currently I do not have access to my settings, but I can provide more information.
I'm using 6144 taps per channel, max. number allowed for OpenDRC.
Sampling frequency is 48 kHz.
I assume the convolution is done by the OpenDRC dsp. I loaded the bin-file according to the manual.
Yes, I did the measurement at the output of the crossover (minidsp 4x10).
In addition I tried another way. I deleted the parameters in the box window and inserted them in the subsonic window and selected a BUT12 HPF.
I got the same (not correct) result like before..
I made a screenshot for my documentation. Please verify the parameters. The blue and red curve are looking fine, 100% congruent.
Any idea?

Best regards

Thomas
Attachments:
Last Edit: 5 years 3 weeks ago by samoht.
The administrator has disabled public write access.

rePhase 1.0.0 5 years 3 weeks ago #20440

  • pos
  • pos's Avatar
  • Offline
  • Platinum Boarder
  • Posts: 350
  • Thank you received: 196
  • Karma: 39
Ok, I understand.
I was able to replicate your issue with HOLM: this is a measurement problem, the correction is probably just fine.

Contrary to a typical minimum-phase impulse, the corrected impulse extends on both side of the impulse peak, and in fact expends further on the left part because of the 120Hz and 1400Hz corrections.
I don't know if Arta tries to automatically set its start/stop window markers, but you should put them further apart, especially the start one, or tell him not to apply any windowing (if you can do such a thing) and consider the entire measured impulse.

Here is an example of two windowing of the same impulse in HOLM. This impulse should be similar to the one you measured: I repoduced your correction and added the miniumum-phase 24Hz HP filter.
The shorter the (left) window, the shallower the slope...
Attachments:
https//wavetracing.com | rephase.org
The administrator has disabled public write access.

rePhase 1.0.0 5 years 3 weeks ago #20455

  • samoht
  • samoht's Avatar
  • Offline
  • Expert Boarder
  • Posts: 111
  • Thank you received: 9
  • Karma: 1
Hi Pos,

Thanks again for your support. Your findings sound logical.
In ARTA the user has to define the window by two independant markers. And in order to achieve a phase resolution with just a few 360 degree crossongs the start marker should be set close to the .impulse. This is what I did and very probably I cut the correction impulse as you explained.
This evening I will repeat the measurements.

Best regards and thanks again

Thomas
The administrator has disabled public write access.

rePhase 1.0.0 5 years 3 weeks ago #20466

  • samoht
  • samoht's Avatar
  • Offline
  • Expert Boarder
  • Posts: 111
  • Thank you received: 9
  • Karma: 1
Hi Pos,

Good news. Your explanation is right. I moved the start marker from close to 2000 samples ahead of the corrected impulse and could see, that both frequency response curves are getting more and more congruent. Perfect.
One strange thing remains: with the moving of the start marker, the group delay indicated by ARTA remains flat but changes the absolute value. Starting with 2 ms it ends up with approx. 42 ms when the marker was 2000 samples ahead of the impulse. I assume that this is again a measurement issue, but nevertheless, the indicated group delay figure is worthless.

Best regards

Thomas
The administrator has disabled public write access.

rePhase 1.0.0 5 years 3 weeks ago #20468

  • curryman
  • curryman's Avatar
  • Offline
  • Platinum Boarder
  • Posts: 792
  • Thank you received: 181
  • Karma: 101
2000 samples are 41,6ms at 48kHz samplerate ;)

Guess the group delay is calculated from the start of the window, however AFAIK this can be corrected :unsure:

kind regards, Daniel
The administrator has disabled public write access.
  • Page:
  • 1
  • 2
Moderators: devteam