Rephase needs a feature , two features
1 an overlay measurement.
2. Minim phase , phase EQ (to simulate the bumps in a response
I’ll ask him , but if others can agree maybe we can ask together
1 an overlay measurement.
2. Minim phase , phase EQ (to simulate the bumps in a response
I’ll ask him , but if others can agree maybe we can ask together
Same, except if you can have it be that way for the early sound arriving at the listening position. I find this has an immediately obvious effect that anyone should be able to hear.
The key is being able to window the response correctly and use whatever tools to change the response based on experience and knowledge to avoid trying to correct things that should be left alone or tackled another way. This is why there is such mixed results, most people are looking for the audio equivalent of a diet pill, a quick simple fix without requiring any effort or work.
It doesn't need either if you want to be able end up removing excess phase i.e. return the response to minimum phase.
Create an excess phase version of the response in REW and export it, now use rephase to flatten the phase where you want to, that filter applied to the original will make it close to a minimum phase response.
Rephase isn't developed in the same way REW is. Asking for changes to rephase is much more likely to end in disappointment to you.
@fluid
Okay so , I have explored that, and the measured phase and excess phase look almost identical (few few minor changes)
Is there really that much excess phase?
If that works then I may just leave it, same shape of phase except the excess what maybe 100 cycles higher on all the wraps and shapes …. Otherwise identical
Okay so , I have explored that, and the measured phase and excess phase look almost identical (few few minor changes)
Is there really that much excess phase?
If that works then I may just leave it, same shape of phase except the excess what maybe 100 cycles higher on all the wraps and shapes …. Otherwise identical
I can't think how it could be but this is getting a long way from the thread topic. Send me a PM with your measurements and I'll take a look.
I was casually reading this thread again and wanted to address the idea that the phase (or time delay) of a system should follow the minimum phase of the system, particularly at low frequency.
If you consider that a system has impulse h(t), then the response to the input can be expressed as o(t) = h(t)i(t) where "' indicated convolution. If h(t) us just a unit impulse it represents a zero phase all pass filter and o(t)=i(t). If h(t) is a linear phase all pass, it is a simple delay and o(t) = i(t-td). In either case, other than the delay, there is no time or amplitude distortion in the output. Now, let h(t) be the impulse of a minimum phase high pass response with, say, 20 Hz cut off. Such a filter will have a frequency dependent time delay from some where significantly above the cut off frequency down to DC. And as the frequency rises the delay will tend to zero. Thus, in general, for a wide band input with content below the cut off frequency the o(t) will suffer both amplitude and time distortion relative to the input. But suppose that the input doesn't contain any content below about 40 Hz. For a typical 2nd or 4th order, 20 Hz Butterworth high pass this would mean all content was in the flat band region. So the output would not suffer any amplitude distortion. However, the frequency dependent delay region of such a minimum phase high pass typically extends well into the flat band region. Thus, while there would be no amplitude distortion there would still be significant time domain distortion in the output. On the other hand, if the HP filter were zero or linear phase such a signal would suffer neither amplitude or time distortion (other than the possible delay). One may feel that the minimum phase HP response "sounds" better but it is not a more accurate reproduction of the input. Furthermore, if the HP system has zero or linear phase well below the cut off (below any possible low frequency content in the input) then even in the event that the input has content into the stop band, the output would only suffer the same amplitude distortion as the minimum phase filter but be void of time domain distortion.
Who doesn't love a sensible mathematical explanation 🙂 I have been seduced by many great sounding explanations only to be disappointed in reality. To me listening to already recorded source material and linearizing the high pass just sounds wrong, as in broken. This could be personal preference, adaption, circle of confusion or anything else.
My advice to anyone is to ignore my or anyone else's opinion try for yourself and make your own mind up.
The most preferable sound I found through testing uses a correction that does things no self respecting engineer would choose to do because it goes against rules that have very sensible explanations. I tried really hard to make it sound better by doing it "right" and I couldn't. So now I just accept it for what it is, the best solution for me with my speaker in my room that I have found so far.
So fluid ,
I’m with you on this 100%….
So the system high pass … so I’ve just been doing a tilt up at 100hz , because I saw a screenshot from Swiss bear that showed that and his sub played much lower
So let me ask you this ….
And I’ll take the time to say it proper so it’s not easy to type this out, I’ll try and be as articulate as possible, and it’s not off topic I don’t think it is all related and man you have done good bits to share… so thank you
so let’s imagine I don’t have a car system for a min (because for some reason ppl think that changes things and it doesn’t really)
Let’s Do imagine you have unequal path lengths, a left , a right and a sub behind you…
You have fir on every channel and a fir as a gloabal 2ch over all of it…So nothing is impossible in this system fully flexible
So with no crossover filters on :
So the left speaker Rolls off at 100hz with a 48db slope because the room HP
The right speaker rollls off at 80hz at 24db because the room HP
The sub behind you rolls off at 65hz with a 24db slope because the room LP but comes back up at 100hz (a huge dip at 80hz) and plays to DC
The the front speakers if you ignore linearizarion on the high pass , the phase starts it’s rise way way before the sub,
The sub can play to dc flat if you want it to…..but you put a BW6 at 18hz so it has more electrical damping….
So summing these together???
Even tho the fronts roll off higher naturally, the cones are still moving as if they’re not rolling off at all…. So you make the phase flat at the driver on the sub and the fronts and ignore the high pass of all 3 sections…
Then at LP , you TA and make the sums flat in amplitude and in phase and the sub no longer sounds like it’s behind you and it sounds like it’s coming from the fronts….. that’s great that’s what we wanted right….
So…… if I did the minimum phase adherence, I would have to do that on the sums to keep that bass to the front because the fronts and sub have been forced to have flat phase (or matching phase rather we could use whatever as long as they match)
If I did the minimum phase match on everything then summed it I would have a giant mismatch in the stop band of all the fronts with themselfs and the sub…..
So the rules must be broken some to get things to match….. and I get what you mean on breaking rules because it sounds better, I’m with you so much on that , altho I understand we want to keep as much correct as we can
So in that scenario, what would you do?
Would you force the sub to have a crossover with the fronts, the 80hz dip is only on the sub and left front, the right front has no dip….
Or just too complicated and can’t think it through?
I’m with you on this 100%….
So the system high pass … so I’ve just been doing a tilt up at 100hz , because I saw a screenshot from Swiss bear that showed that and his sub played much lower
So let me ask you this ….
And I’ll take the time to say it proper so it’s not easy to type this out, I’ll try and be as articulate as possible, and it’s not off topic I don’t think it is all related and man you have done good bits to share… so thank you
so let’s imagine I don’t have a car system for a min (because for some reason ppl think that changes things and it doesn’t really)
Let’s Do imagine you have unequal path lengths, a left , a right and a sub behind you…
You have fir on every channel and a fir as a gloabal 2ch over all of it…So nothing is impossible in this system fully flexible
So with no crossover filters on :
So the left speaker Rolls off at 100hz with a 48db slope because the room HP
The right speaker rollls off at 80hz at 24db because the room HP
The sub behind you rolls off at 65hz with a 24db slope because the room LP but comes back up at 100hz (a huge dip at 80hz) and plays to DC
The the front speakers if you ignore linearizarion on the high pass , the phase starts it’s rise way way before the sub,
The sub can play to dc flat if you want it to…..but you put a BW6 at 18hz so it has more electrical damping….
So summing these together???
Even tho the fronts roll off higher naturally, the cones are still moving as if they’re not rolling off at all…. So you make the phase flat at the driver on the sub and the fronts and ignore the high pass of all 3 sections…
Then at LP , you TA and make the sums flat in amplitude and in phase and the sub no longer sounds like it’s behind you and it sounds like it’s coming from the fronts….. that’s great that’s what we wanted right….
So…… if I did the minimum phase adherence, I would have to do that on the sums to keep that bass to the front because the fronts and sub have been forced to have flat phase (or matching phase rather we could use whatever as long as they match)
If I did the minimum phase match on everything then summed it I would have a giant mismatch in the stop band of all the fronts with themselfs and the sub…..
So the rules must be broken some to get things to match….. and I get what you mean on breaking rules because it sounds better, I’m with you so much on that , altho I understand we want to keep as much correct as we can
So in that scenario, what would you do?
Would you force the sub to have a crossover with the fronts, the 80hz dip is only on the sub and left front, the right front has no dip….
Or just too complicated and can’t think it through?
Last edited:
Sorry that is a complicated explanation that I can't give you an answer to. Try the options you have and pick the one you like, if you like the result it doesn't matter how you got there. Then you can analyse it to try and work out why you like it or why you didn't like the other options.
It really does change everything due to size of the room and less ability to use "room treatment". It basically is a reflection nightmare. I've played with car audio extensively and know one cannot work around or even avoid those reflections, you should accept them being there.
I don't think fluid was talking about the use of minimum phase crossovers though, my guess is the minimum phase he talks about is the phase that follows the measured SPL output at the listening position. So if the bass finally rolls off, one could EQ phase flat, or phase could follow the roll off of that measured output, all at the listening position. I basically do the same thing, as this had my preference in listening tests. As I'm out to please myself, I've accepted this as my preferred solution. I've listened to the system with and without this kind of phase manipulation, the difference certainly is large enough for me to be able to make a 'prefered' choice.
This is one step further than using minimum phase crossovers, it took time and dedication to make it sound right. I didn't get to this point by simply "unwrapping an acoustic measurement of the entire speaker", that's for sure. It took quite a few measurements across the room, additional room treatment, lots of listening and learning when, what and why to correct and most certainly: what to leave as is. Timing is everything 🙂.
This is what the timing of my system looks like, as measured at the listening position, a Stereo measurement, 2x main speakers and 2x subs (home audio):
I believe you're talking about the use of linear phase crossovers and time alignment to make it sound like the sub position is up front, I see nothing wrong about that if it does the trick.
If you have a dip at 80 Hz in sub and left front, you could try make up for it with right front (if it can take the stress, has the headroom) and fill that dip somewhat. Do remove some energy from that left woofer too at that frequency, EQ-ing a null simply does not work.
Timed right it should work nicely to make the listener unaware of the dip in listening sessions with music. 80 Hz represents a mighty long wavelength compared to the usual size of a car. As long as nothing rattles you should be alright.
This talk about "minimum phase filters" make it sound that there is something special about them. The name implies that phase follows magnitude - thats all really... no?
//
//
That’s exactly what I did, and thank you for detailed reply
I actually meant phase works the same in a car or small room as a big room.
And I know he was talking about the phase of the speaker not the individual drivers…the natural high pass
I was asking mostly about dealing with the natural HP of all the drivers in the bass range where we aren’t “supposed “ to use fir to correct….. follow me …. I was trying to expand the conversation to all the bass drivers…. I mucked it up sorry
And I was curious if he is measuring his speaker at LP… because in my room I make the sub all the way flat in phase and frequency at the driver. And then if I measure at LP it has a incline…. That i don’t remove…
So I also wonder if correction on the HP of the speaker doesn’t sound good because it’s being measured with the room. When I make my system phase flat at LP , I’m with fluid it doesn’t sound right… bass is hollow and the 1st harmonic on a lot of frequencies sound just wrong…..
But I get you on my crossover, let’s not talk about that yeah you explain it perfectly. Thank you
Yes nothing special implied or suggested and it not suggested the filters be minimum phase but the sum of all the filters speaker and room's early sound.
Hi John, That all makes sense to me, when considering both amplitude and phase.
My question/concern with whether a system high-pass should be minimum phase or not, stems solely from the potential for pre-ring when using a linear phase system high pass.
As we know, step response clearly ramp-dives negative before t=0 whenever using a lin-phase high pass.
May i ask what are your thoughts with regard to step response then? Does that equate to pre-ring potential? (leaving audibility aside)
@mark100, I think the best way to answer that is with an example. But first I don't think a music system's response to a step is all that relevant. This is because a step has a lot of DC content and speakers are not designed to reproduce DC. Speaker are high pass (or band pass if you like) systems where the band pass should be wider that the content of the input signal. So if I can bore everyone with an example, below is a figure which on the left shows the input step in step function, delayed by 1 msec in green and the output of a 30 Hz linear phase high pass filter with B4 amplitude in red. You can see that the output shows the preringing before t = 1. msec. (If I delayed it longer you would see the preringing extends further to the left).
On the right, before sending the input step to the linear phase high pass I passed it through a 50 Hz 84 db, high pass LR type minimum phase filter. This retains the causal nature of the signal but removes nearly all content below 50 Hz, less than 1 octave above the cut off of the linear phase filter. The resulting input to the linear phase filter is plotted in green. The output of the linear phase high pass is then plotted in red. You can not see the green trace because it is identical and under the red trace. Thus, with the low frequency content of the step removed below 50 Hz there is no ringing in the output of the linear phase, 30 Hz high pass. The obvious point is Input = Output.
So for me it's why should I car about how the system response to a signal it is not designed to reproduce faithfully in the first place?
But after I did the work with the work with the SoundEasy developer and played around with many linear phase system including applying global FIR linearization to some of my old systems I just concluded it was a fun academic exercise but it didn't result in better sound. Sometimes there were audible differences but different didn't equate to better.
On the right, before sending the input step to the linear phase high pass I passed it through a 50 Hz 84 db, high pass LR type minimum phase filter. This retains the causal nature of the signal but removes nearly all content below 50 Hz, less than 1 octave above the cut off of the linear phase filter. The resulting input to the linear phase filter is plotted in green. The output of the linear phase high pass is then plotted in red. You can not see the green trace because it is identical and under the red trace. Thus, with the low frequency content of the step removed below 50 Hz there is no ringing in the output of the linear phase, 30 Hz high pass. The obvious point is Input = Output.
So for me it's why should I car about how the system response to a signal it is not designed to reproduce faithfully in the first place?
But after I did the work with the work with the SoundEasy developer and played around with many linear phase system including applying global FIR linearization to some of my old systems I just concluded it was a fun academic exercise but it didn't result in better sound. Sometimes there were audible differences but different didn't equate to better.
Thank you John k !
This post has been an eye-opener for me.
At first, I thought your example was all well and good, but what if I actually had frequency content down to the 30Hz linear phase high pass, that was being cut off by the adding the preceding 50Hz min phase high-pass.
So I replicated your experiment and lowered the minimum phase high-pass frequency......
and found even then, that step response does not take a negative dive before t=0. (15Hz min phase hpf shown below).
I have simply not understood how dominant low frequency down to DC, is on step response "pre-ring"...
I think it's the same lack of that recognition that started this thread....when I was using REW's sweep generator starting at 20Hz (rather than DC) , and getting step response measurement artifacts.
This summer, I plan on doing my most extensive listening/measuring tests to date, on system high-pass. I have a new large synergy build that has sub drivers in it, and is good to 30Hz with high SPL. All tests to be made outside.
So far, I've viewed the system high-pass trade-off, to be that of potential pre-ring vs group delay.
But now, I'm wondering is pre-ring is even real....or is it simply mathematical representations having to deal with finite bandwidth (and discrete Fourier math).
Anyway, should be fun. Thanks again.
This post has been an eye-opener for me.
At first, I thought your example was all well and good, but what if I actually had frequency content down to the 30Hz linear phase high pass, that was being cut off by the adding the preceding 50Hz min phase high-pass.
So I replicated your experiment and lowered the minimum phase high-pass frequency......
and found even then, that step response does not take a negative dive before t=0. (15Hz min phase hpf shown below).
I have simply not understood how dominant low frequency down to DC, is on step response "pre-ring"...
I think it's the same lack of that recognition that started this thread....when I was using REW's sweep generator starting at 20Hz (rather than DC) , and getting step response measurement artifacts.
This summer, I plan on doing my most extensive listening/measuring tests to date, on system high-pass. I have a new large synergy build that has sub drivers in it, and is good to 30Hz with high SPL. All tests to be made outside.
So far, I've viewed the system high-pass trade-off, to be that of potential pre-ring vs group delay.
But now, I'm wondering is pre-ring is even real....or is it simply mathematical representations having to deal with finite bandwidth (and discrete Fourier math).
Anyway, should be fun. Thanks again.
Since there has been discussion about just correcting the system response to a minimum phase HP I though I'd look at the same filtered step input response when the 30 Hz HP was minimum phase. Green is the input and red is the output of the 30 Hz minimum phase HP. As expected, the initial rise is the same but at longer time the output diverges from the input. Does such a system sound better or worse? In either case you still have to consider what I refer to as the source to listener transfer function which is dependent on both source and listener position and the room modal response. Personally I find that the room dominates.
Which is why my suggestion is minimum phase approximate at the listening position. To me this sounds better than either forcing the phase at the listening position flat or leaving it alone entirely. It is not an absolute preference and is still track dependant but the overwhelming majority of the tracks I have tried it with had my preference with minimum phase at the listening position. Adjusting low frequency phase in room is where to me the effects of phase changes to be most audible. But it is just that a suggestion and my opinion, albeit based on a lot of practical testing.
That's fine which every way, minimum phase or linear phase. Break it down. Say the raw speaker transfer function is S(w), w = frequency, then to correct it's phase to either a minimum phase or linear phase we introduce global fir, FIR(w). Next we have the room to listener transfer function, RL(w). Assume RL(w) is minimum phase. Lastly we need room eq to compensate as best for RL(w), cal it REQ(w) which would ideally be 1/RL(w). So we have at the listener,
L(w) = S(w) x FIR(w) x RL(w) x REW(w) = S(w) x FIR(w) x RL(w) /RL(w) = SW(w) x FIR(w). The only argument here is whether FIR(w) corrects to a minimum phase or linear phase HP. You "prefer" minimum phase. I state the linear phase is more accurate reproduction. Two different sides of the same coin.
But next I like to look at what the global FIR does in the crossover region. It's probably been discussed here already but I'll give a couple of examples for clarity. I consider two similar systems. Both have a 30 Hz HP response with a 2k Hz crossover. One system used an LR4 crossover, the other a B3 crossover. Global FIR is used to correct the phase to a minimum phase 30 Hz HP system (for Fluid, 😉 ). Then the question is what does impulse response of these systems look like at the design point (or what ever reference point you define) and what is the impulse of the HP and LP filters.
For the LR4 the result looks like this. The impulse is delayed 1 msec and as shown, the effect of the global FIR is to reduce the HP and LP section to near linear phase behavior with symmetric ringing to each side of 1 msec. This is no surprise since the LR4 HP and LP have the same phase and application of the global FIR to each filter reduces them both to linear phase through the crossover region. The system impulse show a typical minimum phase behavior with slight undershoot after the pulse due tot he low frequency cut off of the system.
But what of the B3? Somewhat of a different story. The system impulse is identical as would be expected since the system has the same minimum phase HP response. But the HP and LP impulses are significantly different from the LR4 result and lack symmetry. The fact is the both LP and HP section show exaggerated, asymmetric ringing. This is because that while the summed response has basically zero GD through the crossover region (linear phase), unlike the LR4 case, the individual HP and LP filter actually have regions of negative GD and are not even remotely linear phase over any frequency range.
If you listen in an anechoic chamber, on the design axis, none of this matters but in a normal room where there are many reflections from all different angles where the HP + LP sum doesn't cancel preringing it obviously has to color the sound. Thus my personal conclusion after playing around with such systems, it's just not worth it. It creates more problems than it solves and just added complexity and additional processing to the system with no sense of improved sound quality. I just haven't been able to ever convince myself that if I start with a system that sounds good, global FIR phase correction makes it sound better.
Anyway, it's fun to play around with this stuff but I don't think you it should be taken to seriously.
Just my conclusion from my experiences. Tanks for putting up with the ranting of an old man.
Hello John,
thanks a lot for your excellent explanations which are well digestable for many people not so deep into filter theory...
I'd have a specific practical question to ask because I'm planning to buy a miniDSP Flex Eight:
This system features 2048 FIR Filtering taps and multiple IIR EQs in the input section, then a 4-Way xover and then multiple IIR EQs in the four branches. What's your recommendation/suggestion in light of the findings in this thread? I'm asking because I'm not sure to derive the best approach for a practical system yet (probably my fault ... 🙄).
My take so far is e.g.:
1. IIR linearize the drivers in each branch to about 1 octave above and below x-over frequency (employing LW transform if necessary)
2. Apply LR4 overall subsonic IIR HP, 3 LR4 IIR x-overs, adjust delay per branch for driver offsets
3. measure overall Freq.response and Step or Impulse response, adjust branch delays if needed.
4. And now? Shall one (iteratively) do global IIR or global FIR smoothing (e.g. with Rephase) and tweak branch delays for best FR/Impulse/Step response?
Shall one decide if to use global IIR or global FIR smoothing by listening tests or which criteria help deciding ahead of the smoothing work?
Thanks for your evaluation and advice! I hope for my "practical intervention" to not be off-topic 😉
Regards,
Winfried
thanks a lot for your excellent explanations which are well digestable for many people not so deep into filter theory...
I'd have a specific practical question to ask because I'm planning to buy a miniDSP Flex Eight:
This system features 2048 FIR Filtering taps and multiple IIR EQs in the input section, then a 4-Way xover and then multiple IIR EQs in the four branches. What's your recommendation/suggestion in light of the findings in this thread? I'm asking because I'm not sure to derive the best approach for a practical system yet (probably my fault ... 🙄).
My take so far is e.g.:
1. IIR linearize the drivers in each branch to about 1 octave above and below x-over frequency (employing LW transform if necessary)
2. Apply LR4 overall subsonic IIR HP, 3 LR4 IIR x-overs, adjust delay per branch for driver offsets
3. measure overall Freq.response and Step or Impulse response, adjust branch delays if needed.
4. And now? Shall one (iteratively) do global IIR or global FIR smoothing (e.g. with Rephase) and tweak branch delays for best FR/Impulse/Step response?
Shall one decide if to use global IIR or global FIR smoothing by listening tests or which criteria help deciding ahead of the smoothing work?
Thanks for your evaluation and advice! I hope for my "practical intervention" to not be off-topic 😉
Regards,
Winfried
Last edited: