Thanks Wesayso for the link. This seems to provide all needed explanations for having a look, doing some experiments and satisfy my curiosity for the topic.
Let's say it will be summer experiments ;-)
And I fully understand advices from others about the "one location" and "can't solve bad design or bas usage issues"
JMF
Let's say it will be summer experiments ;-)
And I fully understand advices from others about the "one location" and "can't solve bad design or bas usage issues"
JMF
Can I ask how steep filters are we talking about when the ringing starts to become the slightest problem if we are a bit picky?
Hi JMF, i think the Grimm whitepaper has some of the best general advice I've encountered when it comes to using DSP.
But i do feel it sacrifices general advice in support the the LSI speaker project, and looses a sizeable hunk of its objectivity.
I'll set aside the paper's opening statement.....
"Psychoacoustics tells us thatbelow about 300Hz the ear will no longer clearly discern direct sound, first reflections or reverberant sound.",
and the subsequent assertion the baffle size needs to match that frequency transition to omni.
.....other than to say is shows shows some pretty strong opinions from the start.
The off-axis and diffraction examples on page4 are super, along with the impulse analysis.
It's a real shame though, that the FIR correction that was used to compare on-axis tuning and show the off-axis train wrecks,
is pretty clearly the strongest form of correction to a single spot that is possible to make.......
Because the 'stupid perfect' on-axis freq and impulse response as the start of page5 can only be accomplished by pure impulse inversion (or electrical sims).
Impulse inversion is pure nuts because it's overkill correction to a specific spot. (the very thing the paper's general advice condemns, haha)
Iow, the train tracks were loaded with junk on the tracks that couldn't help but illustrate derailment. Maybe some real world clean track examples would have been better 😉
And that's a shame, because the subsequent general advice in 2.3 and the beginning of 2.4 on page6 is so constructive, ime/imo.
I also take issue with one part of the general advice however "For the time being I would strongly recommend designing the correction manually. This rules out FIR as the main workhorse. For each bump or dip one corrects, one should know exactly where it comes from, and make
sure that it isn’t better corrected for acoustically"
The part of the quote's advice that begs we know what we are doing, and do so manually, rocks.
However, I've found FIR is excellent for designing correction manually.
I mean, use RePhase....it's entirely manual.
Make all the IIR / minimum-phase EQ adjustments you want, and simply embed them into the FIR file. (That's probably the principal use of FIR in self-powered prosound boxes, that cannot tolerate much latency.)
Make the FIR file entirely IIR with impulse peak at start, make it linear-phase with impulse peak centered, or maximal/mixed phase with impulse peak past center...it doesn't matter....which is why FIR can be so cool.
Does it all....with high numeric processing precision.
Will try to address linear-phase "steep" xovers vs "sharp" filters next post...they are very different animals...another area i think the Grimm paper got overzealous with in support of their design...
Hi, good lead in to where i quit on prior post.
Thanks ya!
My experience is to separate the terms "steep" and "sharp".
Because steep most often applies to xover slope dB/oct, or # orders of 6dB/oct.
And i think "sharp" most often applies to pass band filters like parametrics .....their Q, or bandwidth spec.
They are totally different animals in terms of invoking pre-ringing. ime.
High Q passband filters suck no matter how implemented, FIR or IIR.
So i avoid them period, .... both pre-ringing (FIR) and post ringing (IIR) are non-issues when you just say no.
And then you can't goof up and phase linearize what doesn't exist . 😉
In contrast, high slope, high order,...... ie "steep" xovers...when complementary linear phase....totally work ime.
No pre-ringing worth mentioning, unless the acoustic design is pretty crappy.
I think pre-ringing from high Q passband filters is like jolts on a roller coaster. Sharp, quick, turns/twists/bumps.....hurt me anyway.
Whereas, steep xovers smooth out the sharp jolts, like a killer smooth high-G force modern roller coaster.
96 dB/oct LR lin-phase has such little pre-ringing ime, is isn't worth talking about, and certainly not in comparison to the easily obtained sonic benefits available.
Last edited:
you guys brought up the cardioid design but no talk of implementation...
I'll throw it out there...passive cardioid designing is limited to software which Diyers dont have... Active cardioid....I guess this can be design in VirtuixCad but I haven't had the time to explore yet....There are some closed back 10" woofers by one of the pro audio names (beyma maybe?)...slap those on the sides of your enclosures and walla??? Obviously not so simple, my first devils advocate thought was similar initial FR main driver vs directivity driver is likely desirable....but what about after that? Take measurements of the main driver in the desired cancellation area and invert the response and ship it out of the side woofers??? Call it a day? Err wait...match the on axis of main driver and directivity drivers as close as possible....sync the side woofer output with arrival time from the main woofer....invert polarity of the signal of the side woofer...Cardioid city?
I'll throw it out there...passive cardioid designing is limited to software which Diyers dont have... Active cardioid....I guess this can be design in VirtuixCad but I haven't had the time to explore yet....There are some closed back 10" woofers by one of the pro audio names (beyma maybe?)...slap those on the sides of your enclosures and walla??? Obviously not so simple, my first devils advocate thought was similar initial FR main driver vs directivity driver is likely desirable....but what about after that? Take measurements of the main driver in the desired cancellation area and invert the response and ship it out of the side woofers??? Call it a day? Err wait...match the on axis of main driver and directivity drivers as close as possible....sync the side woofer output with arrival time from the main woofer....invert polarity of the signal of the side woofer...Cardioid city?
Last edited:
You seem to be taking this literally. Of course you could use that tool if you wanted. In a theoretical discussion of this nature, however, the mention of FIR makes the implication that either...
1. you plan to make a non-minimum phase change, or..
2. you plan to make an arbitrary phase change.
Hope i haven't misread this thread....i'm under the impression it's been more of a real-world pragmatic discussion, than a theoretical discussion.
I think it is an unfortunate mistake if folks automatically associate FIR with linear phase (or the two implications you listed).
Like said earlier, i'm fairly certain FIR's greatest use in prosound speakers is for IIR minimum-phase EQ work, (that isn't feasible with the limited filter counts of IIR processors).
I also feel pretty certain any cardioid processing with FIR, is via embedded IIR xovers / EQs and delays.
So disassociating FIR as necessarily implying linear phase or arbitrary phase is very important in grasping today's processing alternatives, imho.
For me, FIR does two primary things in one file
1.traditional speaker tuning via minimum-phase EQs,
2. complementary linear-phase xovers.
This is an unbeatable combination, for both SQ and ease of repeatable speaker tuning, .......in my experiences to date.
If the latency of the linear-phase xovers cannot be tolerated, i use FIR to implement IIR xovers.
I have a couple of very nice traditional IIR processors available and could go that route.... but they are just too much work to use, especially so as i've become better at when and how to use FIR.
But the context we mentioned was in the linked paper.
We could talk about FIR..... or about 'an FIR based device'
Perhaps (in this specific case) it is you that have taken liberties by considering a device with FIR based processing capabilities and many additional features... and labelled it as just "FIR"? It doesn't mean that the things you have said about it are wrong, but you're using the term in a 'familiar' sense, rather than a technical one.
With regards to your concerns about minimising the importance.. I think anybody else that understands these concepts will know exactly when and will be willing to use either of these processes when required.
Hi Allen, i should probably address this to avoid possible confusion of others..
Answer is nope. I was not considering a device with FIR based processing capabilities and many additional features.......
My comments have been blind to any specific DSP processors, splitting them only with regard to whether they are FIR capable or not.
Any processor or PC that can handle a FIR file works for any min-phase, lin-phase, or max/mixed phase technique that I've mentioned in this thread. No additional processing features are needed, or have been considered.
And rePhase works for constructing any and all of the FIR file alternatives mentioned.
Quite generic stuff really...
Thanks mark100 for all your input, thankfully I mistakenly used the less appropriate word or otherwise I wouldn't have known the differences between steep and sharp. 🙂
My pleasure Ultima Thule,
Those are just my observations for steep and sharp...can't say i've seen the distinctions made elsewhere...so who knows, those distinctions may not be valid ?🙂
Complementary linear-phase xovers are the only use of linear phase that i've found to be close to riskless.
Sticking with Linkwitz-Riley, I've found everything from LR 12 dB/oct to LR 192dB octave xovers to have impulse responses that look close enough to call the same.
Very little on-axis pre-ringing at all, and surprisingly the same for all orders. Summed brick wall impulse looks almost identical to summed low order.
So steep just doesn't seem to matter.
I'm keep thinking the slope/first derivative of the phase curve doesn't effect ringing so much. Whereas, sharp peaks or kinks/second derivative, appear to.
Been having a deep look at what happens to the complementary linear-phase xovers electrical summation when one side is delayed.
The idea has been to try to use delay to synthesize off-axis geometric timing shifts.
Iow, breakup full complementary coupling, and look for pre-ringing in the impulse responses. That's what the ringing theory says should happen i think.
Can already say below 1000Hz, using delays all the way up to reasonable full center-to-center driver distances, there is no reason for concern with pre-ringing.
Can't triangulate against on-axis, worse than full c-2-c, huh ?🙂
The lower the freq, even less the reason for concern, no matter how steep.
Ironically, above 2000Hz i need to measure more. Their may be a case for lower order, the higher freq goes.
It's been alot of fun, using measuring IIR and lin-phase LR's from 2nd order to 192dB/oct. And with varying amounts of delay.
I'm learning how much I didn't know about off-axis impulses haha.
Those are just my observations for steep and sharp...can't say i've seen the distinctions made elsewhere...so who knows, those distinctions may not be valid ?🙂
Complementary linear-phase xovers are the only use of linear phase that i've found to be close to riskless.
Sticking with Linkwitz-Riley, I've found everything from LR 12 dB/oct to LR 192dB octave xovers to have impulse responses that look close enough to call the same.
Very little on-axis pre-ringing at all, and surprisingly the same for all orders. Summed brick wall impulse looks almost identical to summed low order.
So steep just doesn't seem to matter.
I'm keep thinking the slope/first derivative of the phase curve doesn't effect ringing so much. Whereas, sharp peaks or kinks/second derivative, appear to.
Been having a deep look at what happens to the complementary linear-phase xovers electrical summation when one side is delayed.
The idea has been to try to use delay to synthesize off-axis geometric timing shifts.
Iow, breakup full complementary coupling, and look for pre-ringing in the impulse responses. That's what the ringing theory says should happen i think.
Can already say below 1000Hz, using delays all the way up to reasonable full center-to-center driver distances, there is no reason for concern with pre-ringing.
Can't triangulate against on-axis, worse than full c-2-c, huh ?🙂
The lower the freq, even less the reason for concern, no matter how steep.
Ironically, above 2000Hz i need to measure more. Their may be a case for lower order, the higher freq goes.
It's been alot of fun, using measuring IIR and lin-phase LR's from 2nd order to 192dB/oct. And with varying amounts of delay.
I'm learning how much I didn't know about off-axis impulses haha.
check out this thread concerning the SEAS driver in Grimm's LS1 and my question in post #152 under
https://www.diyaudio.com/community/threads/corrosion-seas-magnesium-drivers.300775/
https://www.diyaudio.com/community/threads/corrosion-seas-magnesium-drivers.300775/