Seems like you're thinking of the acoustic and electrical aspects of loudspeaker phase as independent. Like magnitude, they're coupled. For example, if magnitude was decoupled it wouldn't be possible to EQ a dipole's SPL flat. Same goes for phase.
Actually, I own SoundEasy v16. HOLMImpulse is easier to work with, but the methodology for phase measurement is independent of the tool. Any decent one will do impulse extraction and compute phase based on the time of flight indicated by the impulse. Bear in mind a one sample error in locating the impulse corresponds to a 180 degree phase shift at the Nyquist frequency; subsample interpolation in locating the impulse is one area where HOLMImpulse makes SoundEasy look like a toy. You're correct preamp cal is important. I've never seen a mic cal which included phase. You're left to guess where the mic's acoustic center is, but I've yet to hit a case where exact positioning of the mic within millimeters mattered.
If the drivers aren't time aligned you'll see the phase rotate. In which case you measure the drivers one by one from the listening position, note the time of flight to each driver's acoustic center, and adjust accordingly. Like DRC, time alignment can only be correct in one place. So there's latitude in how to set it up. My box speakers are pretty well engineered so the tweeter only needed a 146us delay to bring it into phase match. The difference isn't very audible; I score maybe 10% better than random on blind A/B.
Be interesting if the 979 supports frequency dependent phase shift. Usually all that's supported is various forms of uniform shift. It's not like frequency dependent phase correction is hard to implement---all it requires is an complex * complex multiply rather than a real * complex multiply in an FFT based equalizer like PLParEQX3---but I program for my day job so writing VST plugins is kinda low on the priority list. The FIR synthesis for it isn't terribly complicated either.
Current plan is the four woofer amp channels will be tomchr LME49811+STD03 variants, probably with a single STD01 output pair, with cap multiplied rails around 25V. The four mid and tweeter channels will probably be a similar LME49710+STD01 design with 15V regulated rails.
Actually, I own SoundEasy v16. HOLMImpulse is easier to work with, but the methodology for phase measurement is independent of the tool. Any decent one will do impulse extraction and compute phase based on the time of flight indicated by the impulse. Bear in mind a one sample error in locating the impulse corresponds to a 180 degree phase shift at the Nyquist frequency; subsample interpolation in locating the impulse is one area where HOLMImpulse makes SoundEasy look like a toy. You're correct preamp cal is important. I've never seen a mic cal which included phase. You're left to guess where the mic's acoustic center is, but I've yet to hit a case where exact positioning of the mic within millimeters mattered.
If the drivers aren't time aligned you'll see the phase rotate. In which case you measure the drivers one by one from the listening position, note the time of flight to each driver's acoustic center, and adjust accordingly. Like DRC, time alignment can only be correct in one place. So there's latitude in how to set it up. My box speakers are pretty well engineered so the tweeter only needed a 146us delay to bring it into phase match. The difference isn't very audible; I score maybe 10% better than random on blind A/B.
Be interesting if the 979 supports frequency dependent phase shift. Usually all that's supported is various forms of uniform shift. It's not like frequency dependent phase correction is hard to implement---all it requires is an complex * complex multiply rather than a real * complex multiply in an FFT based equalizer like PLParEQX3---but I program for my day job so writing VST plugins is kinda low on the priority list. The FIR synthesis for it isn't terribly complicated either.
Current plan is the four woofer amp channels will be tomchr LME49811+STD03 variants, probably with a single STD01 output pair, with cap multiplied rails around 25V. The four mid and tweeter channels will probably be a similar LME49710+STD01 design with 15V regulated rails.
Last edited:
I thought it was the toothache smiley. Or maybe my mind is not as macabre as yours. =)
I guess I'm not being clear - I want to treat the final acoustic response of the crossed-over driver because that includes both the electrical filter and the driver's inherent acoustic response, and I'm guessing that the linear phase software only treats the electrical part, thus isn't dealing with the whole picture. But it works for you?
If working with phase is just a matter of removing the TOF, that should be easy. I just need to 'get my hands dirty', I guess.
I started using Waves LinEQ last night - not much to report yet. I also tried the Voxengo PHA-979, and it does not have frequency dependent phase shifting - so it is out.
The LME49830 was the chip I used - there weren't any boards available at the time. Do yourself a favor and buy professionally made boards! Grounding is definitely an issue with a project like this, with multiple amp stages. Similarly, I suspect that 150W is overkill - and that 50W is probably enough, particularly for the mid and tweeter...
Member
Joined 2009
Thanks cuibono, those are some good insights.
I understand that this might not be an issue in your design but I'm interested in the audibility of driver offset. Would you mind running some tests with the drivers offset forward and back? Phase plots(and maybe group delay) would be appreciated.
I understand that this might not be an issue in your design but I'm interested in the audibility of driver offset. Would you mind running some tests with the drivers offset forward and back? Phase plots(and maybe group delay) would be appreciated.
The software addresses what you tell it to. Arbitrator, for example, corrects both crossovers (electrical) and mechanical resonances (acoustic). It doesn't actually distinguish between the two, though the UI sort of implies otherwise. My bet is combining Arbitrator with judicious choice of warped and linear phase equalization will get most anything sorted out without the needing the ability to make pure phase corrections. So far it's worked OK for my two way box speakers and dipoles, but I haven't done a four way dipole yet. Or even a two way with drivers chosen for dipole use.
If you think about it you'll see 1mW into a pair of 86dB efficient drivers delivers a conversational level 50dB at a typical home audio listening distance of 3m (56dB @ 1m - 9dB at 3m + 3dB for two channels). Definitions vary somewhat, but the threshold for permanent hearing damage with long term exposure is usually taken as 80dB with OHSA not recommending more than 75dB and the EU preferring 70dB. For illustration's sake, taking the EU standards and applying a worst case 20dB crest factor gives a maximum SPL of 90dB, or a total of 10W per side. Music is reasonably well approximated by a 1/f power spectral density, so tweeter levels are usually 10 to 20dB below total power per side depending on crossover and music type. That's 10-100uW typical, though I've found some passages where tweeter and woofer power is equal in a two way.
The difficulty with nude dipoles is the woofers run 10+dB lower than rated efficiency. From Linkwitz's SPL spreadsheet I believe 20W per woofer should be sufficient for the 95+dB efficient pro drivers I plan to use. But there's not a lot of data to go on. I've a pair of Parasound A23s I'll use to measure the woofer power requirements and choose the woofer transformer accordingly. For the tweeter and mid amps it's more about scaling down so as not to blow the Neo3 and Neo8s I'll be using. Conveniently, the 78xx/79xx regulator series current limit at about the right level to protect the Neo3.
The last board I did was 40 layers, 235 power planes, and 123,000 nets, about 17,000 of which were impedance controlled within 1% and timing matched within 1ps from DC to 4GHz. So I'm fairly comfortable laying out star ground and star supply and keeping things tidy. The challenge is more in finding the optimal cost versus performance point for the design. So far I don't see any reason it won't fit in two layers but for cheap boards I'm used to working with SPGS or POOL stackups which offer better shielding than two layer. Be interesting to see how the floods go.
Jeez - I won't worry about you then - I will point out one other benefit of the GC amps - built in overvoltage and short protection. Just saying.
Otherwise, all good points. Luckily I still have Allocator around.
I've started playing with Waves LinEQ - first off I have to say that the way it defines Q is very unusual, and a pain in the ***. It also has limited dynamic range - looking at pure electrical function plots, it becomes obvious, but it many not be an issue sonically. It is limited in a number of other ways. I've started porting my EQ settings into it, and it is going to take a while.
But I did one interesting thing today - I took the Alpha6a driver, and its EQ settings, from the beginning of this thread, and ported them into Waves. Fairly simple to do one driver - and then I gave it a listen. WHOA! I did not expect what I heard! If this is a phase issue, phase matters! It was super apparent that when using regular, nonlinear phase crossovers, the sense of depth of the soundstage, and the relative placement of instruments is seriously degraded. I had no idea what I was missing - listening to a single driver with linear phase EQ/XO presented a very deep soundstage with very clearly defined locations of instruments - particularly the acoustic bass and human voice (probably related to the fact that dipoles have a lot of low end EQ). I hope this experience translates over to the full range system!
Very interesting!
Curious to see the change in phase between the two sets of equalization.Yep. Unfortunately I've not found any which meet my <0.05% THD linearity requirement at typical output powers. I was hoping running the 3886 class XD would work out, but it's limited by its input stages and hence turns in a mediocre ~0.5% THD below 1mW regardless of the crossover displacer.
The effects of xover type on power response are easy to see, just take any program that allows the drivers to sum in quadrature (90 degrees apart, acoustically) and this gives a rough indication of the dip in the xover regions power response, assuming omnidirectional sources.
We used to do this with CALSOD back in the 80s, nothing new there. Another way to look at it is that LR types by definition are in-phase so the combined output HAS to be less at most other observation points (again assuming idealized omnis).
I found it interesting that you noted the vocals were more forward with higher orders.
At one time PCABX had a great listening test on-site, where an all pass phase response equivalent to a 2 kHz xover was superimposed, and different orders could be listened to. So no change in amplitude, only phase (and of course, group delay). I easily found (over headphones) that the higher the xover order, the "sharper" it sounded.
I've been a proponent of LR2s for years, if the drivers can pull it off.
Thanks for sharing your test results.
Dave
We used to do this with CALSOD back in the 80s, nothing new there. Another way to look at it is that LR types by definition are in-phase so the combined output HAS to be less at most other observation points (again assuming idealized omnis).
I found it interesting that you noted the vocals were more forward with higher orders.
At one time PCABX had a great listening test on-site, where an all pass phase response equivalent to a 2 kHz xover was superimposed, and different orders could be listened to. So no change in amplitude, only phase (and of course, group delay). I easily found (over headphones) that the higher the xover order, the "sharper" it sounded.
I've been a proponent of LR2s for years, if the drivers can pull it off.
Thanks for sharing your test results.
Dave
Last edited:
If the driver are omnidirectional any crossover that sums in quadrature has flat power repsonse if the levels of the sources is the same.
You misrepresent what I wrote. If the xover is an LR type and sums flat on axis, then the power response can be approximated by summing in quadrature.
The approximation, using one one measurement point in space, assumes omni sources.
CALSOD and other programs have been doing it for years, and you know this.
Context John, context.
I also find it mildly disturbing that you think you're teaching me something that I proved to myself on a VAX in WATFIV in 1988.
Last edited:
The thing that has to be remembered is that phase is ALWAYS a relative measurement. Relative to some position. SO instead of worrying about where the AC is or what the mic cal files has for phase all that is needed is to be consistent. I always measure drivers and remove approximately the excess phase as defined from the distant from the mic tip to the mounting flange of the driver. Assuming the drivers are flush mounted on a baffle then what I do is remove the excess phase defined by the distance from mic tip to baffle surface with the mic on the driver's axis. SO if the mic is 1M from the baffle I remove as close to 1M 's worth of excess phase. So in a code like SoundEasy, which give the acoustic distance from the ref I move the start of the FFt window as close to but not more than 1 M's worth of acoustic distance. In this way the phase is always the driver minimum phase plus and excess phase form the AC to the baffle surface, plus and phase shift which may be associated with the mic. IMO, as long as you use the same mic, this is the most consistent way to measure phase. You don't have to worry about where the AC is or inter-sample interpolation, whether the mic phase is meaningful, ..... You have a consistent measurement of phase, relative to the mounting flange, for all drivers. Codes like HOLM impulse which may be better at inter-sample interpolation just make something simple seem hard. It is a total waste of time trying to define the location of the AC because no matter where you determine it to be, you are just going to add it back in any simulation. And, no matter where you determine it to be it is just an estimate because you really don't know that the minimum phase response is.
Oh. Ok, or you can just integrate the power and be done with it. Gee's Dave, don't go off on me here. Go back and read you post. The "context" certaily isn't clear. My comment lead you to post a clarificaton. That is a good thing for the masses.
Another thing, taking an LR amplitude response and summing in quadrature isn't even a good approximation because an LR can have perfectly flat power response. It depends on the separation between drivers in terms of wave lengths at the x-o frequency.
Last edited:
Thats almost an apology! I'll take it.
Well, you can't integrate the power if you only have one measurement in space.
Anyway, the fact of the matter is in-phase types will have the power response dip in practice as most of them are built (ie listener on teh driver axis, non coicident).
I now more favour lower order quadrature types with the xovers offset for this and other reasons. The minor on-axis ripples are inconsequential, you get better integration of differing dispersions and the power resonse is flatter. I'm not too surprised by Hans findings and perhaps its not too big a surprise why the old timers favoured quadrature xovers more than we do today, when they built these by ear and with some TDS backup.
I agree that an LR can have a perfectly flat power response. Make teh drivers coincident! Another more practical method is to use woofer/midwoofer drivers with some narrowing dispersion and point the woofer away from you, on axis. Ie have the on axis response approximate the power response for the woofer + low pass. Thats one of several options.
I agree the omni approximation is iffy, but its handy and simple. I was working with Jeff to get him to add the driver directivity model to his power response calc in PCD, its at least one step up.
Last edited:
You'd have to explain the workflow you're using in more detail for me to grok what the issue is here. My experience is HOLMImpulse's impulse locating is a quicker and easier way of finding the digital delay needed to bring two drivers into time alignment than the manual window positioning approach used by SoundEasy. Computing the time delay between driver ACs does implicitly involve knowing their location, but with most digital crossovers there's no need to know the physical acoustic center locations or feed information back into a simulation. You just change the delay setting on whichever channel's handiest.
I've also found the more fully removed time of flight phase rotation is the easier it is to to work with phase corrections in Frequency Arbitrator.
Thanks everyone for commenting - I appreciate it when things are a little over my head. Looks like I'll be doing some phase measurements soon. Thanks for making it easier John!
As I said earlier, I've been working with Waves LinEQ crossover/EQ software lately, which is 'linear phase'. I (painstakingly) redid my transfer functions from my usual software, and have been listening. I've got to say its is a big difference - I guess in the past I wasn't doing a good enough job of things to hear any difference, but at this point, I think moving the filter out of the computer is not going to happen. The difference is too big. I'm going to post in more detail over in the Violet DSP thread (post #91?), but in comparison, the audibility of the different LR filters was minuscule compared to moving to 'linear phase' (I use quotes because I haven't actually measured the phase yet).
As I said earlier, I've been working with Waves LinEQ crossover/EQ software lately, which is 'linear phase'. I (painstakingly) redid my transfer functions from my usual software, and have been listening. I've got to say its is a big difference - I guess in the past I wasn't doing a good enough job of things to hear any difference, but at this point, I think moving the filter out of the computer is not going to happen. The difference is too big. I'm going to post in more detail over in the Violet DSP thread (post #91?), but in comparison, the audibility of the different LR filters was minuscule compared to moving to 'linear phase' (I use quotes because I haven't actually measured the phase yet).
Thats almost an apology! I'll take it.
Well, you can't integrate the power if you only have one measurement in space.
Anyway, the fact of the matter is in-phase types will have the power response dip in practice as most of them are built (ie listener on teh driver axis, non coicident).
Fitst statement, true. But it can be easily simulated. How do you think I produced all those plots at my web site?
Second, yes and know. We uses the term coincident loosely. I think we should say "acousticly coincident" as it totally depends of on the ratio of separation/wave length. In any event going back to the original post and setup,
An externally hosted image should be here but it was not working when we last tested it.
I ran the sims. Hers is what the power response looks like for LR2, B3 and LR4, left to right, for two 6" drivers, 6" c to c spacing and a 700 Hz x-o.
An externally hosted image should be here but it was not working when we last tested it.
The red line is the uncorrelated power, or, for the LR types, the power assuming quadrature. You can see that assuming quadrature is not too accurate for the LR types.
Well, if I am building a flat baffle speaker the measurement approach I described is all that you need. Driver phase is relative to the baffle surface so the driver offset is already in the phase. All that is needed is to design the crossover to sum flat. I.e. if it is supposed to emulate an LR crossover adjust the slopes of the roll offs to that the phase difference is 0 degrees at the x-o point. SoundEasy will plot that in the system optimizer and even optimize on this if you like (though I don't tend to use that option). I really don't like the term "time align" because it is meaningless. But if you would rather adjust the driver offset to achieve crossover alignment than filter slopes then all that is required is to design the HP and LP sections and adjust the offset so the sum correctly.
The real point is that once it is understood the phase is a relative measurement (relative to some reference point) it is easy to do anything you like. If you don't want to see all those wraps in the phase, remove as much delay from the measurement as you like. If you remove 1M worth of delay from the tweeter and 1.03 M delay from the woofer all that needs be remembered (or recorded) is that the woofer phase is relative to a point 0.3M behind the tweeters. All that trying to define the AC does is attempt to define the point where the phase is the minimum phase, and we really don't know what the minimum phase of the driver is, we only approximate it.
Last edited:
How does one remove the time of flight?
In HOLM the default is to place the impulse peak at 0 time. Is that removing the extra delay? The offset can be locked if one wants to compare driver phase to another driver.
That's how I do it when looking at phase. Lock on the midrange, then it's a matter of working on the crossover and/or delay to get the phase plots to line up.
Is that the sort of thing you mean?
In HOLM the default is to place the impulse peak at 0 time. Is that removing the extra delay? The offset can be locked if one wants to compare driver phase to another driver.
That's how I do it when looking at phase. Lock on the midrange, then it's a matter of working on the crossover and/or delay to get the phase plots to line up.
Is that the sort of thing you mean?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.