I've been designing some multi-way OB speakers for a while, and looking to push them past the finish line. I use a calibrated mic with SoundEasy for measurement, and do all my XO and EQ in the computer - specifically, with Thuneau's Allocator.
I believe Allocator is an IIR type digital filter - it acts similarly to analog components, being able to define the frequency, order and Q. I've gotten my design to the point where I am very happy with it - and then I realized FIR filters offer a whole other world of filtering. Particularly, I was looking at Acourate, and the ability to completely flatten and define the bandpass of a driver (for a given point in space). That could be very interesting.
So here is the conundrum - I had planned on reproducing my IIR filters from Allocator 'outside the box' with analog, active line level circuits. Removing the computer from the equation is very desirable in terms of ease of use. But I'm wondering if FIR filters are a big enough improvement to keep the computer involved?
In another thread, DSPgeek suggested that FIR filters are most useful for the mids and highs, while IIR filters are good enough for the bass (for a couple of reasons). Anyone else here with experience with both types of filters? I suppose if the mid and high drivers were inherently flat, there would be no need for FIR filters.
I believe Allocator is an IIR type digital filter - it acts similarly to analog components, being able to define the frequency, order and Q. I've gotten my design to the point where I am very happy with it - and then I realized FIR filters offer a whole other world of filtering. Particularly, I was looking at Acourate, and the ability to completely flatten and define the bandpass of a driver (for a given point in space). That could be very interesting.
So here is the conundrum - I had planned on reproducing my IIR filters from Allocator 'outside the box' with analog, active line level circuits. Removing the computer from the equation is very desirable in terms of ease of use. But I'm wondering if FIR filters are a big enough improvement to keep the computer involved?
In another thread, DSPgeek suggested that FIR filters are most useful for the mids and highs, while IIR filters are good enough for the bass (for a couple of reasons). Anyone else here with experience with both types of filters? I suppose if the mid and high drivers were inherently flat, there would be no need for FIR filters.
Speak of the devil
<SFX: cloud of sulfurous smoke>
You rang?
Actually, what I said was that IIRs are good for room EQ, since rooms tend to show minimum-phase response characteristics. HOWEVER, this does not mean EQ will make any lousy room sound wonderful. You still need to work on making a useful listening environment before doing anything else, just as you want to build a good speaker before trying EQ, otherwise you'd might as well join Bose.
And after two years of listening at various Burning Mans, I'm more convinced than ever that linear phase in the low->mid crossover is essential to get the last bit of fidelity from a system. At BM1, vocals through just about every full-range system were compelling enough that I had to drop a conversation and wander into the listening room, which I didn't expect at all because most of my experience is with large multiway systems. At BM2, Iain McNeill showed a system with linear phase, then (with the same frequency response!) 8th order allpass phase response. Percussion, particularly snare drums, through the former sounded more solid and less vague, less polite, than through the latter. I heard little difference with crossover frequencies above 2K, but I may have not known what to listen for.
I would be careful of using extreme FIR slopes, however, since they do tend to generate ringing. Of course, the ringing is negated by the converse ringing from the other driver, but that only applies on axis since differential time delays could make the off axis sum actually have worse ringing that from one filter alone. Moreover, steeper slopes also mean longer delays, which can become a factor if you want to watch a movie.
<SFX: cloud of sulfurous smoke>
You rang?
Actually, what I said was that IIRs are good for room EQ, since rooms tend to show minimum-phase response characteristics. HOWEVER, this does not mean EQ will make any lousy room sound wonderful. You still need to work on making a useful listening environment before doing anything else, just as you want to build a good speaker before trying EQ, otherwise you'd might as well join Bose.
And after two years of listening at various Burning Mans, I'm more convinced than ever that linear phase in the low->mid crossover is essential to get the last bit of fidelity from a system. At BM1, vocals through just about every full-range system were compelling enough that I had to drop a conversation and wander into the listening room, which I didn't expect at all because most of my experience is with large multiway systems. At BM2, Iain McNeill showed a system with linear phase, then (with the same frequency response!) 8th order allpass phase response. Percussion, particularly snare drums, through the former sounded more solid and less vague, less polite, than through the latter. I heard little difference with crossover frequencies above 2K, but I may have not known what to listen for.
I would be careful of using extreme FIR slopes, however, since they do tend to generate ringing. Of course, the ringing is negated by the converse ringing from the other driver, but that only applies on axis since differential time delays could make the off axis sum actually have worse ringing that from one filter alone. Moreover, steeper slopes also mean longer delays, which can become a factor if you want to watch a movie.
If you're looking at FIR filters because of the linear-phase crossovers, you should already have that with the Thuneau package. Frequency Allocator does the crossovers and EQ with IIR filters and Phase Arbitrator unwraps the phase with inverse allpass filters. Some would argue that that's a better approach because it doesn't have the pre-ringing in the time domain that constant-delay (linear phase) FIR filters introduce. It's maybe a moot point if the FIR filters aren't too steep but the thing is you already have software that does what you're looking for -- square wave in, square wave out. The bad news, you can't do either approach with analog filters, it's gotta be a digital box.
http://www.thuneau.com/arbitrator.htm
http://www.thuneau.com/arbitrator.htm
Last edited:
Sometimes it amazes me what I don't know (guess I forgot)
The other thing that is impressive with FIR filters (or just Acourate?) is the ability to define a bandpass and flatten the response within it, via convolution of an impulse response. Does that impress anyone else, or is linear phase the main claim to fame for FIR?
Yep, that's why equalizing loudspeaker drivers is best done using broad, gentle minimum-phase filters rather than straight inversion. You just never know what's up with the data in compromised measurement conditions (living room pseudo-anechoic measurement). Small ripples are best left alone.
On the topic of DSP I find impulse response data very useful. Make one good set of measurements and you can play around with those tweaking the crossover and never need to measure again.
Regarding FIR filters and off-axis ringing, there was a perceptual study in a recent AES journal which should prove quite valuable to anyone interested in FIR filters for loudspeaker crossovers. While the article costs money you can download the more extensive Master's Thesis which it is based on, here: http://lib.tkk.fi/Dipl/list.html (Korhola, Henri: Perceptual Study of Loudspeaker Crossover Filters.)
I agree with breeze. My experience is that driver eq should be done with IIR filters, and avoid using EQ's with very high Q. I also like to EQ things reasonably flat one octave or more outside the passband. When it comes to XO, I prefer FIR, and as being pointed out - use rather low Q on those. I've found that higher "order" (or Q) FIR filters dont not necessarly give you much more attenuation an octave outside the XO point, but only a steeper initial rolloff - and a lot of ringing, which sounds BAD. I see no need to use Q values above 2. 1 to 1.5 is usually more than enough.
Thanks everyone for your opinions!
Breez, thanks for the article - its is very interesting. Here are a couple of quotes from it:
"Rough safety limits according to both test methods would be to keep the order of a linear phase FIR crossover filter under 600 at higher frequencies (1 and 3 kHz) to prevent from the ringing phenomenon producing audible errors. At low frequencies, such as 100 Hz and 300 Hz, the order may be up to thousands, and still no audible errors will occur."
"The results and analysis of the listening test show that phase errors can be heard differently with different signals. The approximate JND limits for group delay errors with the L-R crossover filter seem to follow the rule of thumb (1.6 ms, [22]), when listening to 10 Hz square wave signal.
For real life signals, such as the castanets, the JND limits seem to be higher, from 3-5 ms at high frequencies (1 and 3 kHz) to over 10 ms at low frequency (300 Hz)."
Breez, thanks for the article - its is very interesting. Here are a couple of quotes from it:
"Rough safety limits according to both test methods would be to keep the order of a linear phase FIR crossover filter under 600 at higher frequencies (1 and 3 kHz) to prevent from the ringing phenomenon producing audible errors. At low frequencies, such as 100 Hz and 300 Hz, the order may be up to thousands, and still no audible errors will occur."
"The results and analysis of the listening test show that phase errors can be heard differently with different signals. The approximate JND limits for group delay errors with the L-R crossover filter seem to follow the rule of thumb (1.6 ms, [22]), when listening to 10 Hz square wave signal.
For real life signals, such as the castanets, the JND limits seem to be higher, from 3-5 ms at high frequencies (1 and 3 kHz) to over 10 ms at low frequency (300 Hz)."
Thanks! His off-axis curves are quite familiar, but I hadn't seen any blind studies until now. Very useful indeed, especially when one considers room energy is composed of many such off-axis responses.
About the FIR filter order safety limits, I don't remember a mention of what sampling rate was used in the study, but glancing at some of the impulse response plots, 44.1 KHz or 48 KHz seems probable. This is a pretty important point because the length of the ringing is determined depends on length of the filter vs. sampling rate.
Hi,
my personal findings and preferences (EDIT: I see I'm not alone here)
1) Use gentle FIR(==convolution) techiques to trim the individual ways to proper acoustic minimum phase targets (my preference is zero deg locked-phase designs, like Linkwitz-Riley, phase-matched Bessels and various others), averaged over a set of radiation angles (which requires predominantly mechanically aligned acoustic centers -- not compensated by delays alone). This essentially could also be done with analog/IIR means. With "gentle" I mean not flattening out every minute amplitude and phase ripple as this tends to give ringing in regions where the driver's natural response is not minimum phase.
2) Correct the overall (and smoothed, again) allpass transfer function with its phase inverse, additionally compensating the phase roll-off at the bottom end of the response and below. This is of most benefit if we have higher order bass alignement at higher cutoff frequencies, say a bass-reflex design at 50Hz or so. Both can be most easliy done with convolution but IIR solutions are also possible albeit a bit more complex (cascades of peaking -- in the group delay response -- allpass filters). This phase correcting convolution can finally be factored into the indiviual filters of the seperate ways (assuming they are realized as FIR, of course) to avoid multiple convolution stages. This will avoid/minimize any ringing to a great extend, especially there is effectively no off-axis ringing as the x-over still is minimum phase.
It boils down to the fact that the basic design and construction must be good enough right to start with, otherwise no tricky convolution can fix a basically flawed design.
IHMO the notion FIR vs. IIR is not leading to anywhere, since FIR/convolution is a wrapper term for a myriad of filters and other more complex manipulations (e.g. echo cancelling) and any analog/IIR responses can be "emulated" with long enough FIR filters.
For my two sets of speakers (one active set and one passive) I implemented overall phase compensation to linear phase including the bass rolloff as for the time being. The effects are just as described by others; a "tigther sound" especially on percussion and better "bass speed" (no more delayed bass).
- Klaus
my personal findings and preferences (EDIT: I see I'm not alone here)
1) Use gentle FIR(==convolution) techiques to trim the individual ways to proper acoustic minimum phase targets (my preference is zero deg locked-phase designs, like Linkwitz-Riley, phase-matched Bessels and various others), averaged over a set of radiation angles (which requires predominantly mechanically aligned acoustic centers -- not compensated by delays alone). This essentially could also be done with analog/IIR means. With "gentle" I mean not flattening out every minute amplitude and phase ripple as this tends to give ringing in regions where the driver's natural response is not minimum phase.
2) Correct the overall (and smoothed, again) allpass transfer function with its phase inverse, additionally compensating the phase roll-off at the bottom end of the response and below. This is of most benefit if we have higher order bass alignement at higher cutoff frequencies, say a bass-reflex design at 50Hz or so. Both can be most easliy done with convolution but IIR solutions are also possible albeit a bit more complex (cascades of peaking -- in the group delay response -- allpass filters). This phase correcting convolution can finally be factored into the indiviual filters of the seperate ways (assuming they are realized as FIR, of course) to avoid multiple convolution stages. This will avoid/minimize any ringing to a great extend, especially there is effectively no off-axis ringing as the x-over still is minimum phase.
It boils down to the fact that the basic design and construction must be good enough right to start with, otherwise no tricky convolution can fix a basically flawed design.
IHMO the notion FIR vs. IIR is not leading to anywhere, since FIR/convolution is a wrapper term for a myriad of filters and other more complex manipulations (e.g. echo cancelling) and any analog/IIR responses can be "emulated" with long enough FIR filters.
For my two sets of speakers (one active set and one passive) I implemented overall phase compensation to linear phase including the bass rolloff as for the time being. The effects are just as described by others; a "tigther sound" especially on percussion and better "bass speed" (no more delayed bass).
- Klaus
Last edited:
So I tried the 'Phase Arbitrator' part of Allocator, and the results seemed pretty subtle. I noticed singers stood out slightly, and sounded a little more tonally smooth - ie, more present/coherent/realistic. With percussive stuff, I didn't hear anything in the bass, but high frequency transients were similar to vocals - a little more present/coherent, which gave the impression of faster speed. But overall the effect was almost at the limit of my ability to detect it. I doubt I could detect it very repeatedly if I wasn't doing the switching.
I'd say my attempt at making things phase linear was an improvement, but a very small one. I'm not sure if the improvement was enough to keep a computer in the picture.
I also measured before and after frequency response - as could be expected, altering the phase caused small differences around the crossover points (<2dB), so I can't conclude that I'm hearing a difference in phase distortion, or difference in frequency response.
I'd say my attempt at making things phase linear was an improvement, but a very small one. I'm not sure if the improvement was enough to keep a computer in the picture.
I also measured before and after frequency response - as could be expected, altering the phase caused small differences around the crossover points (<2dB), so I can't conclude that I'm hearing a difference in phase distortion, or difference in frequency response.
DDE (Datasat Digital Entertainment - the old DTS Cinema company) is about to show a new audio processor at CEDIA. The new audio processor includes active crossovers and a room optimization technology called Dirac Live. The crossovers are implemented as minimum phase IIR filters. However the room optimization is done using 20,000 tap 'mixed phase' FIR filters. to quote the product cutsheet ...
"Why mixed-phase filters? Because loudspeakers measured in rooms are mixed-phase and consequently only a mixed-phase correction can restore the impulse response. Minimum-phase filters or linear-phase filters only consider the magnitude response and fails to improve (and sometimes instead corrupt) the actual impulse response of the system. Simply put, such filters add or remove energy at the right frequency but at the wrong time, which means that the music sounds processed."
"Why mixed-phase filters? Because loudspeakers measured in rooms are mixed-phase and consequently only a mixed-phase correction can restore the impulse response. Minimum-phase filters or linear-phase filters only consider the magnitude response and fails to improve (and sometimes instead corrupt) the actual impulse response of the system. Simply put, such filters add or remove energy at the right frequency but at the wrong time, which means that the music sounds processed."
This is odd (I suspect a measurment artifact). When only the overall phase is changed there cannot be any amplitude change nor any difference in off-axis radiation etc, since the whole signal is phase-predistorted of sort, not the individual driver feeds.
- Klaus
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.