Thanks, and yes, audio bandpass as far as math goes is a necessity.
What I don't get, is why do we care to either DC or infinity....when it comes to making good audible sound.
Also don't understand about speculating on what we can tolerate...
Build the dang thing and listen.....🙂
Fully agree, which is why I consider measured on and off axis together to determine valid filters.
Yeah, different ways of thinking i guess.
Personally, if I don't simplify, I never really understand anything.
When driving early automobiles, one had to attend to choke, mixture, spark advance, manual start by hand crank, manual clutch and gear selection, tire temperature and pressure, and probably a number of other things I've failed to mention. When driving modern automobiles, one only needs to push a button to start the engine, select "D", and proceed. Driving has been simplified. But all of those other tasks still need to be addressed; they're just addressed by automatic systems. Modern drivers have the luxury of assuming that they have been addressed appropriately.
Does a person who masters such a simplified system really have any understanding whatsoever of all of the other processes that operate "under the hood"?
Apologize if that comment was in any way taken offensively....no such intention at all.
What I meant, is when we get too far into technical/theoretical/mathematical extremes, that are far below well known levels of effecting audible discernability, what are we doing? Doesn't seem to me we helping applied science.
What I meant, is when we get too far into technical/theoretical/mathematical extremes, that are far below well known levels of effecting audible discernability, what are we doing? Doesn't seem to me we helping applied science.
Funny you should use something engine related as an example. I've been down on the dock all last week working on engines.
Despite my 72years, I'm still an avid standup jetski fanatic. Old school stuff....racing two stokes, where I dial in displacement, compression, ignition timing, carburation, cylinder port timing, exhaust pipe tuning, yada.
It's all quite similar to speaker tuning in a remarkable number of ways...especially exhaust pipe tuning, because that is based on sonic waves, and timing from pipe lengths and volume, pipe expansions and contractions, and engine rpm..
Makes audio look kind easy really. I still feel like a total novice at it.
Heck, i even got a dyno and an airflow bench to help understand how to better build two-stokes. And then became an even a bigger novice lol.
So yep, as one who strives to be a master at seeing things simplified....
...i do also understand what it takes to "look under the hood".
Excellent analogy. To extend it: Using the various filter design tools without understanding what's going on "under the hood" results in filters that are "adequate", just as accepting an automobile (or jetski) in stock condition results in adequate performance. But if one wants to improve performance, then one needs to understand those under-the-hood things.
Thx. ! Standup jetskis have been my #2 passion to audio (which started at like age 6)
And ok, but still have to ask, what audio filtering details are not left to be addressed/corrected when we have flat mag and phase,
on -axis and extending smoothly thru the desired off-axis pattern?
Like said, audio seem pretty straight forward...if you can stand latency, and are willing to use today's tools
Wonderful deep dive perspectives.
Anyone can freely test in Rephase any combination of FIR or IRR filters.
Many moons ago I fell in the trap that FIR could actually cure cancer and make both theoretical perfect phase and amplitude (frequency response). Back then I did not know anything about the value of proper measurements and simulation. I was never satisfied with Sonics….And I was stubborn enough to continue trying to solve live play via Jriver using streaming services. Never got it working properly and swore that I would never again use USB soundcard for crossover. I kept that promise to myself and then nearly a decade later we still don’t have freely available high quality stand alone multiway FIR DSP that go anywhere near what 100 USD PC could a decade ago. Guess market is just too small….
I’m no master of DSP but could I give one advise to myself it would be be:
Forget everything about FIR till You learn basics of crossover ie:
Proper measurements. Skip the eyeballing measurements and build simple turntable before you even think of building speaker (build turntable big enough ie diameter minimum 80 cm.
You need only to make measurements once and confirm reality. If not true solve till true.
Proper simulation. Learn proper simulation tool. VituixCAD is all you will need. Single most difficult task but uttermost important IME is splicing near field and farfield.
Reference for sanity check. Have some good speakers or headphone for reference. Ears are deceiving.
Practice and have fun.
Anyone can freely test in Rephase any combination of FIR or IRR filters.
Many moons ago I fell in the trap that FIR could actually cure cancer and make both theoretical perfect phase and amplitude (frequency response). Back then I did not know anything about the value of proper measurements and simulation. I was never satisfied with Sonics….And I was stubborn enough to continue trying to solve live play via Jriver using streaming services. Never got it working properly and swore that I would never again use USB soundcard for crossover. I kept that promise to myself and then nearly a decade later we still don’t have freely available high quality stand alone multiway FIR DSP that go anywhere near what 100 USD PC could a decade ago. Guess market is just too small….
I’m no master of DSP but could I give one advise to myself it would be be:
Forget everything about FIR till You learn basics of crossover ie:
Proper measurements. Skip the eyeballing measurements and build simple turntable before you even think of building speaker (build turntable big enough ie diameter minimum 80 cm.
You need only to make measurements once and confirm reality. If not true solve till true.
Proper simulation. Learn proper simulation tool. VituixCAD is all you will need. Single most difficult task but uttermost important IME is splicing near field and farfield.
Reference for sanity check. Have some good speakers or headphone for reference. Ears are deceiving.
Practice and have fun.
As with many things, it's easy to decide upon the objective; the real artistry is in how to achieve it. The person who understands how all of the parts work, and just as importantly, how they interact, is ahead of the game. And if the operation can be described mathematically, so much the better.
For decades, people designed woofer systems empirically -- build it, see how it performs, try to figure out what went wrong, repeat. I have books from my father's era that describe rules-of-thumb for how to do this.
Then Neville Thiele and Richard Small figured out how to describe woofer response mathematically. Woofer design almost instantly not only got easier, but better.
The same is true with filter design in general. Sure, one can experiment with various filters and come up with acceptable reponse. But it is easier, and potentially better, to start with the mathematics and fine-tune from there.
Take 3, my same question again...
... still have to ask, what audio filtering details are not left to be addressed/corrected when we have flat mag and phase,
on -axis and extending smoothly thru the desired off-axis pattern?
Please, no more tangential explanations... directly, what is left on the table with filters, past achieving measured flat mag and phase, acceptable both and off axis??
How do I make that better??
I mean, Why the heck should I care about anything else? ...do mathematics really help?...
Fine tuning for me is dialing in acoustic measurements ...what I have to listen to.
Why not just measure whatever results we think our mathematical models are giving?
What counts, our theory, our modeling, or our measured acoustical results?
Why not head straight to measurements?
Sorry this gets frustrating.....
feels like thinking vs doing...
... still have to ask, what audio filtering details are not left to be addressed/corrected when we have flat mag and phase,
on -axis and extending smoothly thru the desired off-axis pattern?
Please, no more tangential explanations... directly, what is left on the table with filters, past achieving measured flat mag and phase, acceptable both and off axis??
How do I make that better??
I mean, Why the heck should I care about anything else? ...do mathematics really help?...
Fine tuning for me is dialing in acoustic measurements ...what I have to listen to.
Why not just measure whatever results we think our mathematical models are giving?
What counts, our theory, our modeling, or our measured acoustical results?
Why not head straight to measurements?
Sorry this gets frustrating.....
feels like thinking vs doing...
I already tried to answer that directly:
"The designer's problem is to decide where and how to compromise between frequency and time responses. 'Can I tolerate pre-ring?' 'How much pre-ring?' 'Can I tolerate delay?' 'How much delay?' 'Can I tolerate deviation from perfectly flat magnitude response?' 'How much deviation?' 'Can I tolerate nonlinear phase response?' 'How much nonlinearity?' And so on."
You just acknowledged at least one of the compromises: "... acceptable both and off axis??" You cannot achieve perfect flat magnitude and phase response at all frequencies, both on and off axis, so there are compromises to be made. Understanding what the problems are, and the compromises implied by the potential solutions, is the designer's job.
I don't know how else to answer this. It has become an argument between empirical design and analytical design. Empirical design can result in good solutions; analytical design generally makes solutions easier and more complete.
Mark, non est tantum facile. But to answer your question, most FIRs, especially da 'linear phase' ones which da naive FIR filter designers spew out, actually make phase of the SYSTEM worse.
What you want is a MINIMUM PHASE filter, which is 'easier' to get with an IIR. The importance of MINIMUM PHASE response is that if you correct it with MINIMUM PHASE EQ, both the amplitude & phase will be improved.
This is certainly NOT the case with naive FIRs (the majority). MINIMUM PHASE has other good stuff .. eg it has the SHORTEST IMPULSE RESPONSE possible for that amplitude response so is best even if you use an EVIL FIR 😵
Simple Arbitrary IIRs goes into this in detail. It also says what you need to do to improve a non-minimum phase system like a multi-way speaker.
BTW, I'm treating this as a 1-D signal path. Speakers are 1-D in but 3-D out. I'm ignoring the important off-axis issues that gberchin & others have pointed out. Sorry I can't send you a copy cos I'm a REAL BEACH BUM and have lost all my own stuff from several HD crashes.
For almost 100 years since Kellogg and Rice, we have had speakers with all sorts of maladities, thermal compression, suspension non-linearities, motor non-linearities, enclosures 'issues' and crossovers that might be adequate with a few volts. Once some power gets applied, the whole thing falls apart. Woofer fs changes, the inductance vs stroke and amperage changes. Crossover coils can saturate, etc, etc.
There is zero point in trying to take a speaker that suffers from thermal, suspension, motor issues, poor design, and trying to 'linearize' it with FIR, I get that.
From what I have gathered from the inferences so far :
Hypex amps are inadequate, antiquated junk.
FIR is for people that wear pocket protectors and speak in algebra.
You can't use FIR without understanding the math behind it.
Did I miss something?
There is zero point in trying to take a speaker that suffers from thermal, suspension, motor issues, poor design, and trying to 'linearize' it with FIR, I get that.
From what I have gathered from the inferences so far :
Hypex amps are inadequate, antiquated junk.
FIR is for people that wear pocket protectors and speak in algebra.
You can't use FIR without understanding the math behind it.
Did I miss something?
Yes to your list of speaker sins.
But actually, only ONE of your sins is significantly audible. There are other important audible sins but I won't go into that. The secret of good speaker design is to address the audible faults first.
But how do you figure out which ones are audible? The answer is DOUBLE BLIND LISTENING TESTS. My post #72 includes stuff checked against DBLTs 🙂
Exactly when did expertise become shameful?
Rhetorical question. No need to answer. Just expressing my frustration.
Yes - as @lrisbo has already pointed out the computational delay would have to be lower than one sample period. So the delay due to the coefficients is going to be dominant.
The whole convolution - string of MACs - has to be completed before the result is available for output. If the computational delay were greater than one sample period that would indicate that there's not enough processing resource to handle that number of taps. Meaning the filter won't function as the calculation can't be completed in the available time. It would need to have its clock speed ramped up to allow the convolution time to be less than the sample period. Or the number of taps would need to be reduced.
Well, if it's indeed one-sample delay and no buffering, then that only makes it easier for a user to approximately work out what delay they can expect for their filter(s).
No buffering would be the ideal situation but some software coders prefer to use buffers for convenience. Or it may be that different algorithms (e.g. using FFT to speed up convolution when the number of taps is large) require buffering.
I think buffering is included to maintain the 'flow' as processing usually happens in blocks (frames) containing a small number of samples (say 256 or so).
Nature has rules and they're called mathematics.
It becomes helpful only if you understand the underlying meaning : When you know better, you're able to do better. The people who know better have approached the art / science with enough respect and spent time & effort learning things which is why they get rewarded suitably.
Nature has rules and they're called mathematics.
Last edited:
People have come to take the convenient assumption that loudspeaker drivers in a system are mostly linear phase as a rule. This is, in my experience, a mistake. Do you really check if your raw driver measured in a cab is linear phase in ins intended passband and can you really trust your measurements?
As others have pointed out, if you fail with FIR there’s a strong possibility your measurements are contaminated with noise. Most ‘home made’ diy type gated and smoothed measurements are made in noisy environments with boundaries - only because it’s almost impossible for the enthusiasts not to, so there is a real danger you end up correcting for noise in your measurements.
Of course, to parly overcome this, you can do impulse response measurements outdoors to partly overcome this, like Mike100, but I bet he has to work pretty hard to weed out whatever junk that hits his mike.
We use an automated IIR/FIR algorithm that spits out anything with noise, but those measurements need to be made in an anechoic chamber with a specific resolution to provide working results.
It’s great to read the theoretical discussions here, for me one needs to be able to actually do something with the math and we are coming into an age were pro FIR processors are becoming affordable and equipped with reasonable a GUY to boot.
Last edited: