OK, so I looked at the B&O Uni-Phase approach. Interestingly, it appears to be very similar to what the first-order filter in a three way setup. Coincidentally or not, the mid in the xover below seems to act like a "filler", similar to the B&O. See the attached plots. The first one is with the mid driver turned off. See the dip as similar to the B&O without the filler. Also see the dotted yellow line of the mid driver which sits right in between the tweeter and the woofer, just like that of the B&O. The second plot is when the mid turned on. Now the system response no longer has the dip. Also the phase of the system is close to zero (of course with ripples from room reflection and diffractions).
This is the first mocked-up, so baffle will probably need to be modified for better diffraction control and better acoustic center-aligned. Also there are quite a bit of room reflections with all the peaks and troughs.
The advantage of the B&O approach as I said above is the use of LR2 which does not stress the drivers as much.
With mid driver turned off:
Mid_Muted | Andy VJ | Flickr
With mid driver turned on:
Mid_On | Andy VJ | Flickr
This is the first mocked-up, so baffle will probably need to be modified for better diffraction control and better acoustic center-aligned. Also there are quite a bit of room reflections with all the peaks and troughs.
The advantage of the B&O approach as I said above is the use of LR2 which does not stress the drivers as much.
With mid driver turned off:
Mid_Muted | Andy VJ | Flickr
With mid driver turned on:
Mid_On | Andy VJ | Flickr
Last edited:
Waiting for the images, it would be interesting if you could find evidence regarding this. It doesn't sit well with me, as it seems to imply something special about single component filters, but doesn't consider the load they work with, or the result for that matter.
Here is a more ideal plot using ideal drivers in a three way setup using first order filter so it's easier to see without any of the real world imperfections. It's basically a similar concept to B&O approach albeit using first order. The mid driver behaves in similar way as "filler" in the B&O filter.
See the first attached plot with the mid turned off. The second plot is with the mid on. Notice the phase of the mid is running in between the tweeter and the woofer phase as similar to the B&O filter.
Here's my comments on the term "transient perfect". I think if you somehow are able to achieve a 0 phase shift within the human audio band, then by definition, you have "transient perfect" within that audio band. Of course it's a lot easier said than done.
I think the words "transient perfect" can be interpreted differently by different people and people may end up saying different things.
See the first attached plot with the mid turned off. The second plot is with the mid on. Notice the phase of the mid is running in between the tweeter and the woofer phase as similar to the B&O filter.
Here's my comments on the term "transient perfect". I think if you somehow are able to achieve a 0 phase shift within the human audio band, then by definition, you have "transient perfect" within that audio band. Of course it's a lot easier said than done.
I think the words "transient perfect" can be interpreted differently by different people and people may end up saying different things.
Attachments
Last edited:
OK, your point? You have a misunderstanding of first order filter apparently. They are actually more complicated than you think of first order as a cap and a coil so maybe that's why your confusion came from.
First order electric turns to Second order acoustic (or LR2) easily, with suitable well behaving drivers, typically paper cones with minimal breakup and weak motors (old school) . With dsp or hardcore analog xo that can be forced too.
An example of hardcore by "Jon Marsh" Kurosawa Koncept System
The midrange plays critical role, but also woofer and tweeter slopes must behave by the book.
An example of hardcore by "Jon Marsh" Kurosawa Koncept System
The midrange plays critical role, but also woofer and tweeter slopes must behave by the book.
Last edited:
miniDSP is not the only dsp gear having fundamental design flaw in FIR implementation. FIR should be available for input channels to make minimum phase response with minimum amount of work and taps. FIR for output channels could be available for experimental purposes but it's quite useless and uneconomical.
One nice solution with DSPs is the use of the subtractive-delay approch. But this can only be done if the DSP is freely programmable (like using Sigma Studio for ADI's Sigma processors). With the usual GUIs supplied with the cheap DSP solutions this is not possible unfortunately.
Regards
Charles
Regards
Charles
Probably yes. Quite simple (2ch) dsp gear/app. with FIR in digital domain would work no matter what kind of XO speakers have. But analog sources cause some complexity or require analog IOs.
I have designed few speakers close to minimum phase with FourAudio PPA dsp plate amps. That also assumes (wrongly) that designer is interested in linear phase XO slopes with FIR only and does not have FIR in input channels. It's doable but needs more design time and has some limits with taps and XO slopes.
Hi guys, my experience runs counter to this.
I've been playing with FIR heavily for about 4 years now, where I can do either global on inputs, or driver-by-driver on outputs. Always on 4-ways.
Setup was 3 OpenDRC-DIs, and a 4th OpenDRC-DA8 that I sometimes used as another DI.
I think it is a mistake to apply phase linearization globally, post crossover(s).
And I share the opinion it is hard to impossible to hear the effect of doing so, especially if the crossovers being linearized are low order to begin with.
I think the correct use of FIR is to linearize phase and amplitude driver by driver, before tying together with a crossover, and then use complementary linear phase xovers.
As we know, drivers are minimum phase, and once summed they are not.
And any correction post summation is valid to a single point in space only.
We know crossovers not minimum phase either, unless they are linear phase FIR crossovers. Again, any correction post crossover is valid to a single point in space only.
Of course the same 'valid to a single point in space' applies to any two or more sources, no matter how well the phase and magnitude alignment, global or driver-by-driver.
But I just keep getting better polars (and better sound) going driver-by-driver on outputs, than global on input.
I rationalize that global seems like crooked trying to straighten crooked, and does so over a narrower range than driver-by-driver (admittedly pure conjecture).
Anyway, my experience is that it is usually too late to obtain the benefits of phase linearization applied to IIR crossovers already in place.
I think if such a good job was done with existing crossovers, that drivers' phase traces perfectly overlapped, then it might work out close to, or maybe even exactly the same.
But such good IIR crossovers are difficult and rarely the case (for me anyway).
Whereas tying together drivers that are first flattened, with linear phase crossovers, is pretty much guaranteed to give excellent results.
I've come to love driver-by-driver linearization, along with steep linear-phase crossovers.
With steep xovers, the frequency range subject to lobing is greatly minimized. So the 'valid to a single point in space' problem gets measurably reduced.
(small rant...I think folks who would shy from steep linear phase xovers because they've read about pre-ringing, should just give a listen. Maybe pre-ringing is sometimes audible by a few folks, but everybody can always hear lobing. Steep linear phase xovers are a no contest win imo)
@John K.... really appreciate your clear well written comments, which make full sense .
When you guys were doing your listening tests, were you using FIR globally on input or some other method?
My favorite way of comparing IIR vs FIR, is to do the driver-by-driver corrections, and then just swap between IIR and linear phase xovers, all else being the same.
BTW, if anybody does want to do global, I still have the 3 openDRC DIs. Will sell for $225 a piece inc shipped conUS. (I've switched to q-sys for processing.)
I found minidsp2x4hd does not have enough FIR taps to linearize high order filters at bass crossover (120 hz) ... but I would expect it could do that for a mid/tweeter cross.
I am able to amplitude and phase correct my tangband w8-1808 fullrange on a open baffle quite well using only fir taps ... not using any IIR’s now.
I'm interested in how well a fullrange open baffle can be made to perform, are you able to post measurements and/or details of the EQ used and details of the baffle size and shape?
I can post some...but I lost the REW and FIR files so you need to await my rebuild of those in next few days.
I will post here shortly what I do.
The approach with open baffle is very effective IMO vs any crossover at power amp output...biamping with gain/phase correction and steep cross gives you the entire voiceband on one phase coherent driver.
I will post here shortly what I do.
The approach with open baffle is very effective IMO vs any crossover at power amp output...biamping with gain/phase correction and steep cross gives you the entire voiceband on one phase coherent driver.
Approach is roughly as below.
Withe the minidsp set to no crossover and zero filtering, scan just the FR with REW several times around the listening position and average the results. I also set REW to white noise and record many samples on peak hold by waving mic around listening position and store that. These measurements are consulted to set the amplitude response when FIR is created. I don’t import those to set filters.
Next I measure at one meter directly on axis of fullrange, and adjust the measured result with something like 1/6 smoothing and I reduce the IR window to a low value like 10...that eliminates reflection artifacts from the measurement. I export this and open it in rephrase.
Using the measurements of paragraph 1, I use the amplitude filter panel of rephrase to set the desired on axis response of the driver (I target flat, but insert your preference at this stage) This step repeats with new measurements, you iterate inside rephrase, generate filters, measure and adjust. It is somewhat cumbersome to reload the fir as you must do for each new measurement.
Once you are happy with amplitude response, go back into rephrase and use the phase linearisation panel...just like last paragraph except this time you are minimizing phase “lumps” .
Eventually it’s as good as you can get it. Generating your final fir requires you note the timing delay produced, which is there each time you generate a FIR. If you do not generate an identical depth FIR for the driver on the other side of the crossover, then the drivers will be out of phase by that many milliseconds.
This is as far as I have taken it, using 1\4 the total available FIR taps for each fullrange, and reserving the balance for the bass channels.
But the process repeats for bass drivers, at least in theory. The lower the frequency, the more FIR taps you need to accomplish anything.
Anyway, when you are done, you should have a FR and a bass driver with amplitude and phase correction, aligned in time...the perfect prerequisites for a steep slope LR cross to add perfectly with minimal group delay.
I suspect when I am done, I will conclude it’s better to use all the fir taps only on the FR...because my cross is so low a frequency.
Last unique thing about open baffles...you can’t really correct for any time mismatch between drivers that’s caused by physical misalignment front to back in their mounting (because that just makes the rear wave twice as bad). In my case that’s 5 cm...rendered not significant by selecting a low cross freq. but that would be important in a mid to tweeter cross.
Withe the minidsp set to no crossover and zero filtering, scan just the FR with REW several times around the listening position and average the results. I also set REW to white noise and record many samples on peak hold by waving mic around listening position and store that. These measurements are consulted to set the amplitude response when FIR is created. I don’t import those to set filters.
Next I measure at one meter directly on axis of fullrange, and adjust the measured result with something like 1/6 smoothing and I reduce the IR window to a low value like 10...that eliminates reflection artifacts from the measurement. I export this and open it in rephrase.
Using the measurements of paragraph 1, I use the amplitude filter panel of rephrase to set the desired on axis response of the driver (I target flat, but insert your preference at this stage) This step repeats with new measurements, you iterate inside rephrase, generate filters, measure and adjust. It is somewhat cumbersome to reload the fir as you must do for each new measurement.
Once you are happy with amplitude response, go back into rephrase and use the phase linearisation panel...just like last paragraph except this time you are minimizing phase “lumps” .
Eventually it’s as good as you can get it. Generating your final fir requires you note the timing delay produced, which is there each time you generate a FIR. If you do not generate an identical depth FIR for the driver on the other side of the crossover, then the drivers will be out of phase by that many milliseconds.
This is as far as I have taken it, using 1\4 the total available FIR taps for each fullrange, and reserving the balance for the bass channels.
But the process repeats for bass drivers, at least in theory. The lower the frequency, the more FIR taps you need to accomplish anything.
Anyway, when you are done, you should have a FR and a bass driver with amplitude and phase correction, aligned in time...the perfect prerequisites for a steep slope LR cross to add perfectly with minimal group delay.
I suspect when I am done, I will conclude it’s better to use all the fir taps only on the FR...because my cross is so low a frequency.
Last unique thing about open baffles...you can’t really correct for any time mismatch between drivers that’s caused by physical misalignment front to back in their mounting (because that just makes the rear wave twice as bad). In my case that’s 5 cm...rendered not significant by selecting a low cross freq. but that would be important in a mid to tweeter cross.
Ok, I exaggerated telling that FIR in output channels is useless. It's not necessarily useless if you have so much dsp capacity that you don't have to design some compromise you don't want to.
I've done both IIR EQ + linear phase XO slopes and linearized magnitude & phase responses and then linear phase XO slopes. The first one was selected to final product due to better sound. Capacity limits forced to use slopes which caused worse directivity features, so I wasn't too happy with the situation.
All radiators are not exactly minimum phase.
FIR and linear acoustical phase does not change unfortunate fact that certain phase coherence is not achieved in full space with multi-way so that argument does not work against other than linear phase. Quite opposite because acoustically linear phase individual BP radiator is not minimum phase alone ~ kinda unnatural if sound separates from the other radiators.
But all this is too theoretical.
Typical cone and dome drivers are minimum phase. However, that does not mean the phase response is independent of position. The amplitude response changes with position and so does the phase. As was stated, any multiple driver speaker system will sum correctly only at the design point. Move off axis in any direction and the response changes. In a vertically aligned system, moving off to one side or the other tends to have only a small effect if the crossover frequencies are chosen correctly since most of the directionality issues are buried in the stop band of the crossover filters.
I'm interested in the off-axis behavior of the examples andy2 did in the beginning of the thread, also how deeps the nulls are when inverting a tweeter.
I have a feeling the drivers don't add so well together to achieve the flat phase curves, and that would make the response a bit odd when moving off axis, especially up/down?
I have a feeling the drivers don't add so well together to achieve the flat phase curves, and that would make the response a bit odd when moving off axis, especially up/down?
Thanks for your explanation. I'm not sure what you mean by the above? What shape and size is the baffle and where are the drivers placed on it?
In a closed box speaker, you can use a minidsp to delay the signal to a tweeter so that it time aligns with the driver on the other side of its crossover. With an open baffle delaying one driver to time align with another could match the timing of the front signal, but only at the expense of making the rear facing signal twice as far out of phase.
In my case the FR is mounted 5 cm further forward than the woofers, and that timing mismatch could only be properly fixed by a new baffle which aligns driver position more closely. Not important at the low frequencies I am crossing at, but if you want to employ an open baffle tweeter, that mismatch could become important.
In my case the FR is mounted 5 cm further forward than the woofers, and that timing mismatch could only be properly fixed by a new baffle which aligns driver position more closely. Not important at the low frequencies I am crossing at, but if you want to employ an open baffle tweeter, that mismatch could become important.