VituixCAD supports FIR filters and has IR export for convolver.
I would use IIR for crossover and linearize total phase with FIR common to all output channels. This has some practical limitations e.g. typical miniDSP application does not have FIR for input channel (for outputs only).
I've also tested linear phase vs. phase distorted with headphones and few speakers. Audibility with headphones is not so sure without high levels that you can sense harder kicking on the skin around the ears. Advantage of FIR was not worth with loudspeakers having dynamic PA drivers. So I continue with IIR though my active systems have FIR.
I would use IIR for crossover and linearize total phase with FIR common to all output channels. This has some practical limitations e.g. typical miniDSP application does not have FIR for input channel (for outputs only).
I've also tested linear phase vs. phase distorted with headphones and few speakers. Audibility with headphones is not so sure without high levels that you can sense harder kicking on the skin around the ears. Advantage of FIR was not worth with loudspeakers having dynamic PA drivers. So I continue with IIR though my active systems have FIR.
NO ! FIR is the most powerful tool though but linear-phase can be done in IIR as well.
It can even be done in purely analog fashion, although it is not easy and it does usually come at the price of either complicated circuitry (older Meyer designs or PSI Audio for instance) or not so perfect lobing behaviour. And it is definitley easier to correct for every slight deviation from linear with DSP.
Regards
Charles
Hi, as diyaudnut said, rePhase can do all you asked about above. It was on the rePhase thread that i learned about compensating for drivers' natural roll-offs before tying them together with crossovers.
FirDesigner is an excellent product too, with great automation. FIR DESIGNER - Audio FIR filter design tool for speakers | Eclipse Audio It's my go-to software now that I feel more comfortable using automation judiciously (rePhase is entirely manual)
Your linked project looks fantastic...really nice work and great documentation..thx for sharing it.
After you mentioned making critical comparisons of different type filters on headphones, I decided to do that this morning since it's been a long time since I've used anything other than my speakers for evaluations.
I split the audio stream into two 4-way outputs. Crossover frequencies were 100, 650, and 6300Hz for both (these are what I use on my speakers) Then I summed the outputs after the crossovers. No other EQ, processing etc.
So, 2 summed outputs; using 4th order linear-phase crossovers on one, and 4th order LRs on the other. A Linea ASC48 4x8 processor allowed for immediate comparisons.
I started with a 50Hz square wave as a signal, and verified on a scope that the summed linear-phase signal was indeed nice and square. The LRs summed were of course wonky looking. I was expecting to hear a tonal difference, thinking harmonic timing has to be off with the LRs. I did NOT hear any tonal differences, at any freq square wave...which i found a bit surprising.
Then I switched to music. Quickly noticed that transients were more open and dynamic. It was clear enough, I stopped to double check levels on all the individual passbands post crossover, thinking that might be part of the cause. With a scope again, I carefully set all output channels to be spot on equal, using sine waves in the middle of their passbands. Some small mV adjustments were needed.
I heard the same thing, an easily noticeable improvement on transients. The lower the fundamental freq, the greater the improvement, and the greater the comparative 'opening-up' of the sound.
Whew ! After hearing no tonality difference with square waves, I was wondering how much wishful hearing have I been I doing with my speakers haha.
Aah yes, good ole 'older Meyer' ! It was some Meyer speakers and Bob McCarthy's writings that started me down the whole "phase matters" trail. In school for Smaart, instructors talked about the Meyer 'signature phase trace', its gentle slope and being the same on all their boxes. And like you say, all done analog. And like you also say, complicated and not easy.....especially the lower that phase stays flat in freq.
Below is a transfer of one of my UPA-1p's....as you can see phase lets go at around 500Hz. I do think taking phase flat all the way down to the bottom like Meyer's Bluehorn touts, is a very new development in commercial speakers. Could it even be done analog, pragmatically ?
Attachments
Last edited:
I don't find this surprising. If someone plays the same chord repeatedly (on the piano, for example) with the same technique (attack, etc) then I hear it the same way. There's no way the relative phase between the notes is the same each time they play it, but that doesn't impact my perception at all.
If you haven't done this blind, I would suggest doing so (ABX testing). Then assuming you can still distinguish the differences (I suspect you will be able to), it would be interesting to see if you can identify which crossovers lead to phase distortions that are audible. I suspect that if you mimic a two way with a crossover at 6300 hz you'll never hear a difference. Mimic a two way with LR4 at 650, and you'll probably hear something. Make it LR2, and the difference will likely be very small, and you may not be able to say you prefer one over the other. I'm not really sure what to expect for the 100 hz crossover. There's not much for transients down there, but it's a lot of group delay. I have been able to identify phase distortion for a 200 hz or 250 hz crossover in the past. It may be that the phase distortion of a 100 hz crossover is discernible on headphones, but not on speakers in a room, due to room reflections. It's important to remember that results obtained on headphones are not necessarily relevant to speakers.
Hi,
I suppose what you mean with lineair phase is zero acoustic phase. This is in the passive analog form possible but only with a with a limited bandwidth and most of the time this is not possible with a direct radiator system (like the photo from Meyer).
A direct radiator, when acoustically small, the output is (almost) 90 degrees phase behind it`s input. This is due the falling radiation impedance. Most analyzers measuring zero phase when the magnitude response is flat, but this is not correct. The TDS technique (TEF) will show the real acoustic phase.
Zero or nearly zero acoustic phase is possible when placing a more resistive loading on the driving loudspeaker. The only possible way i am aware of is using a horn. A big plus is of this is also making the system more sensitive.
gr. Marcello
This 4-way (at home) with FIR preset is quite close to minimum phase. I will compare to IIR preset tomorrow though there are some (other) obvious audible differences due to different filter orders and EQs which make difference to power response etc. Minimum phase response in SPL is lime/green and flat zero excess group delay in GD&Phase is blue.
An externally hosted image should be here but it was not working when we last tested it.
Last edited:
Which analyzers are you referring to? All FFT based analyzers I know of show real acoustic phase, it's easier to do that than to calculate derived phase. There were some stepped sine analyzers that would only do Hilbert Phase, but I don't think those are made that way anymore. And of course "RTA" types don't show phase at all
I presume that's due to the way that the guys at miniDSP program their software. The SigmaDSP line of chips has the ability to place an FIR filter before the rest of the IIR filtering blocks.
I think I described this here before, but I did a demo/experiment at an audio club here with speakers playing music. Same speaker (coaxial point source), same (passive) crossover, but with and without FIR phase correction. I showed the ability to properly radiate good looking square wave sweeps via mic and oscilloscope first then played a number of music tracks, mostly acoustic recordings. The crowd wasn't bowled-over, to say the least. Not many said they could hear the difference and of those who said they did, about half preferred the non-phase corrected (possible digi-phobia there, though the signal went through the same FIR hardware in both cases, just with different files loaded).
Not terribly scientific, but the demo (and my own experiences) would suggest that it's not a generally significant effect, though there may be people who are sensitive to it to some degree.
Not terribly scientific, but the demo (and my own experiences) would suggest that it's not a generally significant effect, though there may be people who are sensitive to it to some degree.
I've read the audibility of flat phase response disappears when the complex phase response of the room is added to the mix. Under certain conditions flat phase response can sometimes be heard through headphones. I think it might have been Earl Geddes who wrote that. He has some useful info to add to this discussion.
Hi,
All system that rely on the hilbert transform for calculating the phase response from the magnitude response will show the wrong zero phase. The old MLS systems work that way. And i suppose other systems that work with a steady state signal like pinknoise will also. The measurement system should measure the magnitude AND phase on its own.
A point source, direct radiator have most of time flat magnitude response in the midband area. Its flat because it`s been compensate with it`s velocity controll (mass related), while still feeling a changing radiation resistance acoustically. This is the reason you can`t measure/show this with an measurement of a electronic device (equalizer). There are no frequency dependent resistors in electronics.
I shall put some measuring phase traces together from different programms to compare against the TEF.
gr. Marcel
Depends on what you are talking of. If you include the phase response caused by the low-end rolloff then it isn't possible in any analog fashion. But I doubt whether it would make sense at all because it would come with some unevitable latency. The K+H O500 Studio Monitor had such a (user selectable) feature for instance but it added latency. The new Meyer monitoring system isn't phase linear down to its lower cutoff frequency (which wouldn't make sense either as mentioned). But the Ssub/low-mid crossover is phase linear, which would already be quite a challenge to do in analog, especially when steep slopes are called for.
Here you can find some curves:
New Meyer Sound Bluehorn Monitor System Provides Phase Accuracy from 25 Hz to 20 kHz on Large-Format Loudspeakers | audioXpress
I definitely wouldn't mind one for home listening if it weren't for the costs.
Regards
Charles
1) So pre-ringing isn't a problem anymore?
2) Same old problem, applying a 1 dimensional correction to a 3 dimensional problem. Just as amplitude response of a speaker varies with off axis position, so does phase.
3) Bodzio's Ultimate EQ has been around for what, 8 years now. With it you can linearize phase to below the cut off without audible delay in HT applications.
Nothing to see here.
2) Same old problem, applying a 1 dimensional correction to a 3 dimensional problem. Just as amplitude response of a speaker varies with off axis position, so does phase.
3) Bodzio's Ultimate EQ has been around for what, 8 years now. With it you can linearize phase to below the cut off without audible delay in HT applications.
Nothing to see here.
I have to correct my statement above: The measurement I was talking of came from a different system than the new Meyer monitoring.
Depending on how exactly phase-linearity is achieved, the off-axis pre-ringing doesn't have to be more severe than the directivity-dependant time-domain deviations of conventional crossovers.
There are not that many studies dealing with the audibility of phase deviations of ordinary crossovers. The group-delay distortion wich is the result of non-linear phase response is actually decreasing the "information capacity" of the whole reproduction chain. How significant itactually is, is hard to tell. A recent study by someone from Genelec was talking of GD distortion in the order of 1 ms as audibility threshold. For some individuals and source material it was as low as 500 us.
I am using an analog crossover BTW with GD distortion of 420 us, the shape of the GD curve is a gentle downward slope towards higher frequencies without peaks and Dips (acoustic x-over frequency is ca 700 Hz).
I would word the 2nd Statement of John a little different though: It is 1 dimensional correction of a three dimensional error.
The "problem" as such one wants to solve is actually only a one-dimensional one. I boldly claim that if phase linearity really is important, it is mainly so on the listening axis, off-axis the importance of linear power response is higher than that of the time-domain response.
Regards
Charles
Depending on how exactly phase-linearity is achieved, the off-axis pre-ringing doesn't have to be more severe than the directivity-dependant time-domain deviations of conventional crossovers.
There are not that many studies dealing with the audibility of phase deviations of ordinary crossovers. The group-delay distortion wich is the result of non-linear phase response is actually decreasing the "information capacity" of the whole reproduction chain. How significant itactually is, is hard to tell. A recent study by someone from Genelec was talking of GD distortion in the order of 1 ms as audibility threshold. For some individuals and source material it was as low as 500 us.
I am using an analog crossover BTW with GD distortion of 420 us, the shape of the GD curve is a gentle downward slope towards higher frequencies without peaks and Dips (acoustic x-over frequency is ca 700 Hz).
I would word the 2nd Statement of John a little different though: It is 1 dimensional correction of a three dimensional error.
The "problem" as such one wants to solve is actually only a one-dimensional one. I boldly claim that if phase linearity really is important, it is mainly so on the listening axis, off-axis the importance of linear power response is higher than that of the time-domain response.
Regards
Charles
And phase response looks so much better off axis without that one dimensional correction...
Test is done and the result is: no need to change from active or passive IIR to linear phase with FIR. I was hoping a bit stronger kick and tiny positive change also to higher frequencies but the result was something between none and too small to bother.
Excess group delay of tested 4-way is about 2ms @ 100Hz with IIR which seems to be good enough for me, though it is the longest delay for a very long time in my constructions. I've heard quite many speakers with excess GD of 3...5ms @ 100Hz and they have been weak without exception. Weak is that you don't feel e.g. piano and drum in your face until sound pressure is very high.
This wasn't scientific trial at all but it looks that flat zero excess group delay is not top-5 method to improve anything if delay is "short enough" with IIR.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.