They will measure similarly, on the design axis, but not off axis. Of course they won't sound the same. Using FIR to correct phase does nothing about the issues I have brought up regarding driver overlap, excursion, polar response, etc. But you can take any off the shelf speaker, measure the response on axis and construct an FIR, unity gain filter to correct the phase response. Nothing changes in the speaker. It's just that the input signal is preprocessed so what come out is zero (or linear) phase. It's not different than any other active equalization except you are working with phase as opposed to amplitude. And it's really easy to do with programs like the Ultimate Equalizer or even SoundEasy. That capability has been around for 10 years now. With FIR EQ you are not changing the way the speaker reproduces the input, you are changing the input.
Last edited:
Please tell me, I just can't understand how input signal FIR before xo works with multiway speakers.
The overlapping range is often several octaves and each driver has different phase roll depending on slopes being hp/lp and Q, often non-ideal. Is it really the sum that matters? Doesn't this lead to unpredictable off-axis response/phase in both hor and vert?
The overlapping range is often several octaves and each driver has different phase roll depending on slopes being hp/lp and Q, often non-ideal. Is it really the sum that matters? Doesn't this lead to unpredictable off-axis response/phase in both hor and vert?
It works like any other EQ before the crossover. The only difference being the fact that it is much more powerful when it comes to the correction of temporal properties compared to other types of EQ.
The fact that it doesn't influence inter-driver phase response (and subsequent off-axis behaviour) is exactly why John proposed its usage for comparing transient perfect to non transient perfect behaviour without changing any other properties !!!
Regards
Charles
Juhazi, I totally have the same opinion/experience.
I don't even consider input FIR to be FIR tuning....
It's like John and Charles said, it's just an EQ that handles phase too...
(and if I may add, ironically barely helps folks hear the benefits of linear phase.)
And I think FIR has gained a lackluster reputation because most of the experiments with it (at least that I've read about) were just that...
input FIR laid onto existing speakers.
Might as well go full bore and get full room correction ala Dirac or something similar.
A truly better speaker tuning process demands driver by driver correction, linear phase crossovers, and multiamping each driver section....again imo/ime.
Nothing I haven't been saying over and over Lol
Hi John, nice cogent post.
I think, like I just posted to Juhazi, that the way you defined FIR EQ is the way most people have come to think about it, because that was the way most people tried it.
And then we heard honest opinions that they didn't hear much of a change. Which has become kinda accepted lore re FIR tuning. (And is an assessment I share about input FIR..not much of a change).
I do wish more people would try driver-by-driver correction/linearization, linear phase xovers, and multiamping. I'd love to see if the general consensus changes.
Maybe the reason I hear more benefit doing that is because all my speakers are 4-ways, and phase issues compound more.
I've often wondered about that...if 2-ways would really benefit at all..or rather where is the tip point in terms of how many ways....
With FIR phase correction you are correcting the phase of the sum of the filters, not the individual filters. You start with LP + HP = O, where O has unity gain and some arbitrary phase. Then define I = 1/O. I will also have unity gain but the inverse phase of O. I x O = 1, unity gain, zero phase.
I x LP + I x HP = I x (LP + HP) = I x O = 1. Doing this changes noting about the speaker, and I X LP and I x HP are not linear phase. This is very different than designing linear phase HP and LP filters that sum flat.
Think of it another way. Rather than phase think of it as time correction. The phase response of a speaker can be equated to a time delay. If the phase is linear the delay is constant. If phase is nonlinear the delay is frequency dependent. Suppose for a given speaker a 200 Hz tone is delayed 2 sec and a 2k Hz tone 1 sec. If you construct a filter that it delays the 2k signal 1 sec and doesn't delay the 200 Hz signal then an input passed through that filter is fed to the speaker the 200 Hz and the 2k Hz tones will both be delayed by 2 sec so, while delayed, they will be back in sync.
Last edited:
This is not the best way to design loudspeakers. Better to design a minimum phase filter first and then linearize phase. Comptutationally way more efficient.
Is this done to facilitate working without a simulator or do you mean something else?
Totally agree minimum phase is the procedure for driver-by-driver correction/linearization.
First, minimum phase amplitude correction that simultaneously corrects phase within the passband, that everyone commonly practices.
And second, minimum phase out of band flattening to achieve a degree of phase linearization through summation regions.
Both the best and easiest way I know to do that for each driver, is to embed all those minimum phase EQs into a FIR file for each driver. Essentially unlimited EQs, incredible flexibility.
So imo, best practice is to have a FIR file for each output, if for no other reason than doing top notch minimum phase work.
And since I already want/have a FIR file for each output, it only makes since to also embed the linear phase crossovers into those separate files.
Re computational efficiency....
I first ask myself, how much computation goes into an all minimum phase dsp capable of matching FIR's minimum phase capability......and then adding in a second stage of dsp on top for FIR phase linearization....
….compared to...
Simply do it all in separate FIR files to begin with, recognizing that FIR's computational requirements for the various passbands halve with every octave increase.
And I mean, who cares about computation efficiency anyway ?
I think the real issue is acquiring a processor with a sufficient number of FIR channels, and the necessary I/O.
Hehe, I learn valuable lessons from posts like this:
What I tend to do these days is tiny little tweeks to things like speaker position or cross over frequency and relish the difference.....for a while....
What I tend to do these days is tiny little tweeks to things like speaker position or cross over frequency and relish the difference.....for a while....
Last edited:
Let me address this because it makes little sense to do this in steps. Let D be the driver's raw, unfiltered response. Let A be the desired acoustic target. Let F be the filter response. So F x D = A. Or, F = A / D. Having computed F it is straight forward to implement use using various convolution algorithms for FIR filtering. Notice I didn't say anything about A other than it was some acoustic target. It can be minimum phase or linear phase. Makes no difference. But done this way F will include all amplitude and phase correction in a single filter designed from a simple, and single calculation. All that remains is setting how many taps will be used for each filter along with the sampling rate. I don't remember but I believe when we set up the Ultimate Equalizer 10 years ago each filter was assigned 8192 taps with 8 channels (8 separate filters) suitable for 2 4-way speaker, or 4 2-ways. or any other permutation of 8 channels you would like. Easily handled by an average Windows XP computer of the day. One reason to keep the number of tap constant for all filters was that they all required the same processing time and it just made things a lot simpler. Like Mark says, with a dedicated processor capable of handling the load, you may as well use all of it. The only time you might worry about computational efficiency is if you have to sync the audio output with video.
Last edited:
Embedded systems I've been played with use output FIR only so chances to get minimum phase result 20-20k were depending on features of radiators and numbers of ways. I have not succeeded so far with output FIR due to limits in the system.
---
In theory input FIR + output IIR is exact match with output FIR with some assumptions related to phase differences between radiators and XO slopes.
So actually interesting question is WHY output FIR has produced different or even better result. Is it due to poor measurement method or system incapable to measure timing/phase differences properly or laziness with phase matching with IIR or what? Something is different in filter implementation assuming that it's not just subjective imagination.
If you use IIR you are limited to cascading minimum phase filters and drivers. You can not, for example, achieve an acoustic LR4 LP on a midrange driver. But with FIR you can define the acoustic target to be a minimum phase LR4 response and construct an FIR filter that, when cascaded with the driver's response, will yield an LR4 acoustic response.
I think that is the difference. With FIR you can design the filter to provide what ever acoustic output (target) you like. With IIR you are just cascading biquad stages with varying Q along with biquad notch, peak, shelf stages to get the best match to the target you can.
I think that is the difference. With FIR you can design the filter to provide what ever acoustic output (target) you like. With IIR you are just cascading biquad stages with varying Q along with biquad notch, peak, shelf stages to get the best match to the target you can.
I don't agree with that. If you know how to do it you can of course use the same principles to reach acoustic target functions with IIR as you would with analog methods (except at higher frequencies due to the warping of the frequency axis). But if you use a GUI that is restricted to just cascading filter blocks then it is of course not possible.
Regards
Charles
I always thought (because minidsp 2x4HD only allows output FIR) that use of FIR was simply a better way to equalize independently powered drivers prior to applying a crossover between them, because FIR can align both amplitude and phase prior to enabling a crossover. To get that cross to add perfectly, I presumed phase must match prior to applying the crossover?
Is there a hidden advantage to IIR that I am missing?
Is there a hidden advantage to IIR that I am missing?
Yeah Charles, I agree with you.
It seems to me, if you use the procedure I keep recommending for individual driver-by-driver tuning before considering xovers, you can achieve any desired acoustic target with either IIR or FIR.
The key is to flatten out of band amplitude as best you can thru the intended summation region, knowing that it will be 'undone' when the crossover is added.
The out of band flattening and the xover sum to the filter you would eventually reach thru trial and error, trying to reach a measured acoustic target of processing plus driver.
The out of band flattening just gets you there a whole lot easier and faster, and has a logical elegance to it imho. (learned it from Pos/rePhase)
And here's where I agree with kimmosto in theory.
If you do the above, and then add in IIR crossovers, and then add input phase linearization of those IIR xovers ….. that should very closely replicate doing the above with output FIR using linear phase crossovers.
But in practice, doing the above driver-by-driver tuning is hamstrung if not impossible, by the difficulty in finding an IIR processor with the needed number of filters per channel to do both in and out of band flattening.
Output FIR solves that with it's unlimited embedded IIR capacity..
If someone is using PC processing, they have ample filters and taps, and should be able to go either route.
But since both routes require driver-by-driver tuning best accomplished via FIR, and processing constraints are not an issue, it seems pointless to not simply use linear phase crossovers on outputs.
Either way, the key is first-class in and out of band driver flattening prior to crossover.
If we are using non-PC hardware for processing, our experiments get tougher.
FWIW, I've found QSC's q-sys and a Core110f to be a godsend for experimentation.