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

fluid

Member
2009-01-24 2:20 pm
can someone take a peek Into to this and help me figure what they mean by a 9 pole fir and 10 pole fir, I’m not exactly understanding because I learned fir filters have no poles and is why there stable
Seems like they mean the number of channels that can be processed

"A dedicated firmware release enables the user to process up to 9 digital channels with crossover FIR filters"
 
  • Like
Reactions: 1 user
Seems like they mean the number of channels that can be processed

"A dedicated firmware release enables the user to process up to 9 digital channels with crossover FIR filters"

aah I see now…. I wonder so much what there pre programmed fir look like and windowing , stop band ripple, they used etc etc….. I might get one and give it a go….. just for fun . Expensive experiment. But it’s an interesting unit

thanks! That actually makes more sense. There manual is goofy, especially with the % symbols and such…….
 
MiniDSP have similar units, both for car and home use. Can often be used with Dirac
I know, I’ve been using minidsp and rephase in my car since 2015. I’ve recently got good enough at it that I no longer use the assistance of Dirac. Now it’s all rephase and I love it…..

But I just like to try things out for fun.
and what’s crazy the car dsp market blows away the home dsp market….. except for pc based. But that’s a pc running dsp not a dsp
 
What's beyond rePhase guys? Moving the speakers outside? LOL. Need more tweaks!

i love rephase….. I wish there was something slightly more automated… like audiolense it’s an attempt but it’s all fir

i think we want something that can do mixed fir/iir , Multichannel, can manage all speakers, measurements, corrections from one platform….. and looks at left and right coherence. We don’t want to spit out separate left and right filters and do nothing about the impulse matching as much as possible.

rephase does do all of this, but it sure is a lot of bouncing between holm, and REW and back……..
 
  • Like
Reactions: 1 user
So I have a question….

For my driver tuning ive changed from doing it at the speaker to at the listening position…. (It’s a car so only one seat is valid and that’s fine)

i turn off all crossovers first
And instead of doing auto eq on the magnitude, I have been doing moving mic averages with pink PN. Then, after everything is flat throughout the natural response using peq in my 2x4hds.

then I take one measurement (directly between my ears) and based off that measurement alone I don’t look at the response or try and do anything with that.
As my averages real time did that. The only thing I do with the single point measurement is:

1. take single point measurement 1/6th smoothing and a a FDW and window gets shorter on low frequency drivers between 15cy and 4cy

2. After ir windows and smoothie I extract excess phase

3. export to rephase

4. remove excess phase and engage fir crossover of choice and send to dsp fir bank

So far so good….. after I do all the drivers and sun them flat it sounds good. Then I go to my OpenDrc and use auto eq for the summed response and to make my target shape. And I follow the guides and do the multi point measurements and vectors and all the steps in the Swiss bear and pda0 guides to the T

overall sound is good, but responce after using MmM and RTA isn’t as flat as I would like…..

how can I use MMM properly a d completely get out of auto eq and still get proper excess phase charts ???
 
Hi!
Basically I'm using RePhase as equalizer (nothing fancy it would seem, just generate a filter after adjusting the gain EQ phase)

My hardware for this is OpenDRC

I have several questions

1)
Centering part of RePhase UI allows to specify delay in ms instead of "middle" or "energy"

Middle doesn't seem to work at all (the optimization step never finishes, goes into thousands of iterations and keeps going for many hours with no end in sight)

energy works, filter is created but UI printout claims that expected delay is quite hefty - 52ms (ouch, I guess)

specifying some very low ms value (2ms) results in apparently successful creation of filter with no apparent deviation (both thick red line for gain and punctuated red line for phase match the predicted curves perfectly)

What is the tradeoff from specifying a very low delay value (1-2ms) in the "centering" part of UI specifically in context of using minimum-phase gain EQ for headphones?

Would the resultant EQ be somehow deficient / introduce unusual noise / ring / anomalies relative to the energy-centered, "heftier" 52 ms version?

2)

Does it make sense quality-wise and latency-wise to break the EQ down into "problem" part that PEQ GUI of open DRC doesn't handle (due to Q<0.5) and "do the problem part" in FIR and then add the missing "normal" peaks (Q>1) via OpenDRC's PEQ interface (thus essentially chaining filter-as-EQ and "proper built in EQ")?
 
Hi!
Basically I'm using RePhase as equalizer (nothing fancy it would seem, just generate a filter after adjusting the gain EQ phase)

My hardware for this is OpenDRC

I have several questions

1)
Centering part of RePhase UI allows to specify delay in ms instead of "middle" or "energy"

Middle doesn't seem to work at all (the optimization step never finishes, goes into thousands of iterations and keeps going for many hours with no end in sight)

energy works, filter is created but UI printout claims that expected delay is quite hefty - 52ms (ouch, I guess)

specifying some very low ms value (2ms) results in apparently successful creation of filter with no apparent deviation (both thick red line for gain and punctuated red line for phase match the predicted curves perfectly)

What is the tradeoff from specifying a very low delay value (1-2ms) in the "centering" part of UI specifically in context of using minimum-phase gain EQ for headphones?

Would the resultant EQ be somehow deficient / introduce unusual noise / ring / anomalies relative to the energy-centered, "heftier" 52 ms version?

2)

Does it make sense quality-wise and latency-wise to break the EQ down into "problem" part that PEQ GUI of open DRC doesn't handle (due to Q<0.5) and "do the problem part" in FIR and then add the missing "normal" peaks (Q>1) via OpenDRC's PEQ interface (thus essentially chaining filter-as-EQ and "proper built in EQ")?

To center in ms , the filter taps need to be at least the length or half (??? What I’ve noticed) of the amount of delay…. Like you can’t do 10ms on a 200tap fir , but could on a 1024tap fir for example

moving the centering forward is for getting better centering on low frequencies. You can use “less taps” and get good phase and amplitude on your fir

you can’t capitalize “ms” in the box or put spaces

i think Just try all lower case like centering ten milliseconds would be typed ; 10ms

if your using an OpenDrc and doing a correction on everything I would go for use center of impulse or exact centering. Use the max taps if it won’t generate clean. And try starting with a rectangular window first and try other windowing schemes after rectangular works. (I always use rectangular or Hahn)

and you should see 64ms of delay using the 6144taps. So 52ms isn’t bad.

as far as other errors, idk , pos would know
 
Last edited:
@Oabeieo there are taps to spare it would seem given that I'm not doing much other than EQ, so using them all up to get a fast and reasonably accurate EQ is okay by me.
The filter apparently generates fine with very low values in ms (like 1 or 2) and as far as I understand the GUI is very congruent with expected curve and phase

@dgerg

1) how do I specify in samples?
2) are there any drawbacks at all in building a pure-minimum-phase-EQ filter this way and if no, why does "energy" centering and "middle" centering (managed to make it work) produce filters with so much more delay?
 
Yeah what degreg said made sense but i’ve never tried that. I was always under the impression the fir needed taps to make whatever you towing it to whether it’s minimum phase or linear phase. I don’t see how a minimum phase correction could take less taps… so I’m gonna go out on a limb and assume I’m misunderstood on that.

why don’t you just do it in bu-quad iir section of the OpenDrc? Are you just trying to ditch the recursion?

it kinda sounds like your install of the software might be freaking out a little bit
Or somethings not being explained properly that I can understand.

maybe one of the heavy hitters can chime in…
 

fluid

Member
2009-01-24 2:20 pm
Yeah what degreg said made sense but i’ve never tried that. I was always under the impression the fir needed taps to make whatever you towing it to whether it’s minimum phase or linear phase. I don’t see how a minimum phase correction could take less taps…
It does not take less taps as that will will affect the frequency resolution and shape of the filter.

A minimum phase filter doesn't need to have the impulse centred or offset as the impulse response is not symmetric like a linear phase or symmetric'ish like it would be in a mixed phase filter.

When there is no phase/time correction being applied the peak can be right at the start. The actual latency that you get through any device or computer will depend on how the convolver is programmed and the length of any buffers used too.
 
  • Like
Reactions: 2 users
It does not take less taps as that will will affect the frequency resolution and shape of the filter.

A minimum phase filter doesn't need to have the impulse centred or offset as the impulse response is not symmetric like a linear phase or symmetric'ish like it would be in a mixed phase filter.

When there is no phase/time correction being applied the peak can be right at the start. The actual latency that you get through any device or computer will depend on how the convolver is programmed and the length of any buffers used too.

aah filled in the dots perfectly.

thank you for that!

so I wasn’t overthinking it, but definitely didn’t know that. So what do you think his issue is , an install issue with rephase?

because an OpenDrc and a minimum phase fir with rephase is kinda a no brainer.
 
Last edited:
  • Like
Reactions: 1 user

fluid

Member
2009-01-24 2:20 pm
So what do you think his issue is , an install issue with rephase?
Bad choice of settings I think. Enter 0 with closest perfect impulse to get the lowest latency.

I tested this with Jriver and it gives me the shelving as depicted with effectively no latency, for MiniDSP would need to be a 48K bin file at some other number of taps.

MPTest.png

MPTestGraph.png
 
I don't think I have a "problem" per se - with settings like 1ms the filter builds and does not exhibit any "coarse" anomalies (one of the problems with headphones that are good, is that equalizing them to be even better can get placebo-tricky very fast ;-) )

I am mostly curious whether specifying a delay/latency and very low one like that (I'll try "0 with closest perfect impulse" today) has any substantial drawbacks with my application (a "no frills minimum-phase EQ).

I suspected it might have drawbacks given that "default settings" produce a filter with delay in 50ms range and decided to seek more knowledgeable people.

P.S.:
the reason I'm not using biquad PEQ facilities provided by miniDSP's UI itself is that the oratory EQ for these headphones involved Q values way below 0.5, which rePhase handles okay but MiniDSP biquad graphical equalizer facility does not handle at all.
 
  • Like
Reactions: 1 user
It does not take less taps as that will will affect the frequency resolution and shape of the filter.

A minimum phase filter doesn't need to have the impulse centred or offset as the impulse response is not symmetric like a linear phase or symmetric'ish like it would be in a mixed phase filter.

When there is no phase/time correction being applied the peak can be right at the start. The actual latency that you get through any device or computer will depend on how the convolver is programmed and the length of any buffers used too.
Fluid, to remove this problem one just use the same amount of taps when generating, yes?