TOPIC: Re: rePhase Intro - A great FIR tool !

rePhase is the Brainchild of Thomas (Aka Pos), a Frenchman with great ideas! Initially started with the idea of providing phase linearization with the OpenDRC, this freeware application keep evolving with new features on a monthly basis! A great work with a dedicated developer so we can only recommend that you have a look at it and make sure you thank/increase his Karma for his good work!

Please consult the following posts/thread for more info:
- sourceforge.net/projects/rephase/?source=navbar
- diyaudio forum: www.diyaudio.com/forums/multi-way/221434...-filtering-tool.html
- minidsp forum: www.minidsp.com/forum/3-suggestion-box/6054

Let's dedicate this forum to questions you may have or simple feedback of how useful the software was to you!

Devteam

Hi Devteam,

you are absolutely right, I've downloaded rePhase and started testing immediately. It's such a fantastic great tool although I need more time to understand all possible functions and settings. It would be helpful to have at least a basic documentation.

Thomas

Hi samoht,

I must admit I am not fond of writing documentation
I understand the need for some documentation though, and might (should!) get at it in the future...

In the meantime, and as Devteam has opened a whole forum section for rePhase (thank you Devteam!), you can open a post here and ask any question you might have, and I (and other users maybe) will make my best to answer them

That is the purpose of the section and any question or suggestion is welcome!

Hi Pos,

Thanks for your willingness to provide further explanations of your great tool.

First question is about the section "Filters Linearization".
I assume this section is used to linearize the phase of the filters of a physical already existing box.
In "crossover" i only can find LR filters of different shape but no Butterworth for example. Why?
What is the use of this crossover?
In "box" a frequency should be entered. Is this frequency the lower cut-off frequency?
What is the difference between "box" and " subsonic", which includes butterworth filters?
Do the 3 entry lines in the "box" section represent the 3 speakers in a 3-way box?
When rePhase calculates the impulse, i suppose this is done always for the whole frequency range from 10 Hz to 20 kHz, or can i set individual frequency ranges for woofer, midrange and tweeter in order to reduce the number of taps?

Section "Linear Filters" i believe having understood, although i'm asking again why Butterworth filters are missing. Are Butterworth filters not linear phase ones?

That's for the first step. Sorry but i have to understand the different functions before using them.

Best regards

Thomas

Hi Samoht,

>First question is about the section "Filters Linearization". I assume this section is used to linearize the phase of the filters of a physical already existing box.

Yes, it compensates for the overall phase response of a standard textbook Linkwitz-Riley crossover alignment, which is the combined sum of the output LPF and HPF sections. You can build up to 8 crossover points (9-way system) in Rephase, by adding more turnover points on the graph. The overall sum of combined sections’ phase response will be shown on the graph, but this is, of course, giving the “inverse” of the actual phase response for correction purposes – by convolution with your audio signal. Once this specially convolved pre-corrected signal goes through your actual crossover and is subject to those natural unavoidable phase shifts, it will then emerge (in separate freq bands, of course) which should theoretically recombine (sum) back together with a net result of zero relative group delay, ie. Linear phase. Viola! (Similar techniques are also used by the Thuneau Phase Arbitrator)
Caveat number 1: the Rephase impulse, by it’s nature of trying to “undo” phase delays which smear energy out after the original impulse t=0 point , must therefore introduce some audio frequency energy earlier than the original impulse t=0 point, by introducing processing time BEFORE t=0, thus forcing the original impulse point to occur some time after t=0, and introducing an overall delay to the audio stream. This won’t bother you much for hi-fi music only playback, it may bother you slightly for video sync, and it will bother you a lot for live audio reproduction if you are a musician playing an instrument or singing through a mic with this type of DSP processing delay on the signal.
Caveat number 2: the recombined sum of the freq bands after the crossover and Rephase pre-processing will theoretically have linear phase response, but only as far as would have been otherwise spoiled by the crossover process itself (which is all you’ve so far corrected for) and if the crossover outputs are going back into a digital audio editing system (for multiband processing) this might be perfect behaviour, but if the crossover outputs are feeding into real-world loudspeaker drivers (woofers, midranges, tweeters) there will definitely be further serious phase shifts and non-linearities due to those drivers as well. So don’t believe that Rephase-ing a crossover alone will let you hear “magic” phase linear audio in your listening room, because you won’t, until you have found a way of exactly correcting for the phase distortion of each individual driver as well. But this same shortcoming would also be true of any stand-alone phase linear crossover, such as Dolby Lake machine, if you had one. Textbook phase linear crossover behaviour can be approximated using Rephase, but it won’t give you phase linear behaviour with any connected loudspeaker driver unless you have done the homework to correct for those errors also. Fortunately, Rephase’s included Paragraphic Phase EQ may go a long way towards helping you correct for those errors once you’ve got a clear graph of the before / after results to compare everything.

>In "crossover" i only can find LR filters of different shape but no Butterworth for example. Why? What is the use of this crossover?

It’s a free beta software release, and that’s all that’s on offer so far. Perhaps Pos may one day include Butterworth and other filters to try. Most speaker designers use Linkwitz-Riley 24dB crossover alignment anyway – especially with analogue crossovers - it is widely agreed to be the best compromise among the traditional crossover types for numerous reasons.

>In "box" a frequency should be entered. Is this frequency the lower cut-off frequency?

It’s the vent frequency of the box tuning for vented systems, or the Thiele-Small alignment of the cabinet (the box rolloff) for closed boxes. It gets you fairly close to your target curve, but may require trial and error fine-tuning looking at your result graphs.

>What is the difference between "box" and " subsonic", which includes butterworth filters?

Subsonic is for a subsonic rolloff filter (to prevent over-excursion) if you have one of those in your setup that you also wish to phase-compensate for as well, as part of Rephase correction.

>Do the 3 entry lines in the "box" section represent the 3 speakers in a 3-way box?

No, but they would allow for designs with multiple ports at different freqs (if you haven’t calculated your overall vent freq), or vented midrange drivers, or other combinations, etc. but mostly it allows more scope for fine-tuning the results on the graph by very slight amounts.

>When rePhase calculates the impulse, i suppose this is done always for the whole frequency range from 10 Hz to 20 kHz, or can i set individual frequency ranges for woofer, midrange and tweeter in order to reduce the number of taps?

The graph displays 10Hz-20kHz but the audio covers everything from DC to half the Nyquist frequency, which could be up to 96kHz audio bandwidth since Rephase supports 192kHz sampling rates. The number of taps will affect the low frequency cutoff below which processing won’t work properly, but the tpas are typically a limitation of the hardware DSP device (eg. OpenDRC 6144 taps, MiniSharc 2048 taps) rather than Rephase software itself, which handles up to 1,048,576 taps.
You can’t set individual ranges with Rephase, but you could use it three times to create three impulse files for each drive unit, woofer / mid / tweeter, in certain applications. The “Filters Linearization” process however, corrects the overall sum of crossover outputs, not the individual HPF or LPF or bandpass sections individually per se, so won’t be much help in that instance.

>Section "Linear Filters" i believe having understood, although i'm asking again why Butterworth filters are missing?

Butterworth filters are not missing!
Under “Linear Filtes” section of Rephase you can select “1st order” which is Butterworth 6dB/oct with the-3dB point at any chosen frequency.
You can also select “2nd order” with user selectable Q or centre frequency. The choice of Q determines the shape of the slope and allows for most well known filter types.
Butterworth Q=0.7071 by definition (1 over square root of 2)
Linkwitz-Riley Q=0.5 by definition (product of two cascaded Butterworths, ie. 0.7071 x 0.7071)
Bessel Q=0.58
Chebyshev Q=1.0 etc.
Higher order filters can be simply constructed from combinations of 1st and 2nd order.
You should be able to create all familiar textbook filters using these controls, and even more advanced filter slopes, eg. like 4th order Legendre by cascading two 2nd order filters with Q=2.10 and Q=0.597 respectively. Almost anything’s possible with up to 8 flexible filters to play with!

> Are Butterworth filters not linear phase ones?

No, a Butterworth filter isn’t linear phase at all. Except for a textbook 1st order 6dB/octave Butterworth crossover (two matched Butterworth filters HPF and LPF at same –3dB freq) which does sum together naturally to give flat overall amplitude response and flat overall phase response. (Although neither HPF nor LPF section outputs are linear phase, on the own, only when summed together do you get net zero phase shift. Rephase wouldn’t therefore be required for this crossover alignment.) About 17 years ago, I built my own custom 4 way Dynaudio speakers whose design uses a 1st order Butterworth crossover fundamentally. There are various pitfalls to such designs though, such as the very shallow slopes being difficult to handle for most drive units - not enough rejection outside passband, tending to limits on power handling, etc. but that’s more of a speaker design issue.

Regards Richard.

Pos, thought I'll save you the effort...!

Hi Richard,

Thank's for your explanation. This is very helpful for me. I see now, how rePhase works. Great and thank's again.

Best regards

Thomas

Richard wrote: Thank you Richard!

Concerning the Butterworth/Bessel and other traditional filters types, I am going to open another topic to explain why they do not (and will not ) appear in the Phase Linearization tab (expect for the subsonics) or Linear Filter tab.

Hi,

Filter linearization works really good. Have done this for my active LR24 crossover.

Additionally I like to linearize the group delay where dissortions are caused by PEQs and time delay used to compensate the physical
Offset between drivers. I tried several settings within the EQ menus, but wasn't successful. Is it possible at all?
What is the purpose of the EQ menus, to set EQs (output magnitude and Phase) or to correct phase and group delay caused by other PEQs?
Has anyone worked on this subject? Any feedback would be appreciated.

Best regards

Thomas

Hi,

No ideas?

BR

Thomas

Thomas,

Physical offset is a pure delay and doesn't require phase linearization....per se.
You could address this either by actually physically aligning the drivers in your design or adding a pure time delay in your crossover platform.

If your crossover system is analog and some type of all-pass filter is used, you could utilize re-Phase to linearize the phase response of that portion.....in addition to your crossover phase response.

Or, maybe I'm not understanding your query?

Cheers,

Dave.

Hi Dave,

Thanks for your feedback. You got the point. You are saying that time delay does not create phase dissortion and therefore group delay has not to be corrected. That sounds good to me and solves half of my problem.

Thanks again

Thomas

Hi Thomas

As Dave said delays and phase linearization are two different things.
If your delays are set correctly inside your crossover (to compensate for physical driver offsets) then there is nothing to compensate outside of the global phase shift from the filters themselves.
On the other hand if the delays are not set correctly, then nothing can be done in practice to solve that on the stereo signal.

Regarding EQs, I am not sure to understand what you are implying exactly.

The amplitude EQ section in rePhase is there to let the user modify the amplitude response of the system (or specific driver when building a multiway crossover).
The user can choose between minimum-phase and linear-phase EQs. Most of the time minimum-phase EQ is what should be used, as a driver can be considered a minimum-phase device in its usable range: any EQ that flatten the amplitude also flatten the phase around that point.

The phase EQ section is there to help the user fine tune its phase linearization process after filter phase linearizations has been applied.

Hope this helps
samoht wrote:

Hi Pos,

Thanks for your explanation. That's I was looking for.
I had the wrong thinking that rephase is used only to correct all(!!!) phase distortions created by what soever other DSP/Xovers, including those caused by PEQs.

Now I understand from your explanation, that I should use the EQs in rephase rather than others in order to achieve a linear phase response. This means that rephase becomes part of the overall digital correction of a speaker, including amplitude flattening.

The allocation of tasks will be probably:
- classical digital DSP with IIR filters creates the crossover filters, for ex. LR24 and eventually subsonic filters.
- DSP with FIR filters linearizes the phase shift caused by xover, speaker's rolloff depending on type of box and flatten the frequency response by means of EQs.
The combination of both leads to a (nearly) time-correct speaker system.
If this is not correct, please let me know.

Best regards and thanks again

Thomas

Hi Thomas
It can be used in different situation: either doing the whole crossover, including filters, EQ, and box/driver phase and amplitude linearization (one impulse per channel, running on a miniSHARC for example) or only doing phase linearization on the stereo signal in front of another crossover (2 impulses, running on an openDRC for example)

But just to make it clear once again: IIR (minimal phase) PEQs do not introduce phase distortion: they fix them most of the time (flat amplitude = flat phase in a minimal phase system).
Sorry, I think I was not very clear.
You should use minimal phase EQ 99% of the time, for the reason explained above.
IIR EQs that can be found in most crossovers *are* minimal phase, so there is no reason not to use them. By the way it is almost impossible to setup a correct crossover without using EQs, as you almost always need to adjust the natural response of the driver to get the acoustical slopes you are looking for. If your acoustical filters are not complementary in amplitude and coherent in phase throughout the crossover, and your drivers not time aligned, there is no way to correct that using rephase or any other tool acting on the stereo signal in front of the crossover...

So if you want to combine an IIR crossover and a stereo convolution engine (using rephase or any FIR design software) you should use all the EQ, filtering, and delay capabilities of your IIR crossover to achieve that (amplitude complementarity, phase coherence, time alignement), and then use the convolution engine to adjust the global amplitude and/or phase curves at your will (phase linearization, ampltide target curve, fine adjustements, DRC, etc...)

Hello Thomas,

I'm planning to use rePhase & your FIR/IIR with miniSHARC suggestions for the implementation of a 4-Way Active System.
My current questions:
- What is rePhase's capability or your suggestion for Driver Offset alignment?
- Not having the Math Background... Which FIR x-over slopes are possible when using rePhase with miniSHARC at 96 ks/s?

Thanks a lot in advance for your advice!

Greetings,
Winfried