Cheapest Full Range 20hz - 20khz Speakers?

[andreasmaaan]

Is the keyword here, system?
I don't believe that, in general, the use of linear-phase xovers makes the system linear-phase.

And I'm sure you know this, but a step function that does not return to zero does not represent a realistic example of a speaker system.

Well ok, but none of that changes the degree of pre-ringing introduced by the filter, nor the relationship between vertical axis and pre-ringing cancellation.

My purpose was to distil, not to mislead

Here's the same again, this time with a min-phase 2nd order roll-off at 40Hz.

On-axis (a small amount of pre-ringing is already visible here since the simulated woofer is no longer perfectly linear-phase at the XO point):

+15cm off-axis:

 

[dc655321]

My purpose was to distil, not to mislead

I did not assume otherwise

Well ok, but none of that changes the degree of pre-ringing introduced by the filter, nor the relationship between vertical axis and pre-ringing cancellation.

The demonstration is clear. I'm trying to figure out how you got there, and if this is a general aspect.

One of us is missing something though (and I would not bet against you!).
What tool(s) are you using to simulate?

 

[andreasmaaan]

I did not assume otherwise

Haha, thanks

The demonstration is clear. I'm trying to figure out how you got there, and if this is a general aspect.

It’s generally true that ringing will cancel on-axis, but not off-axis. What changes (depending on XO frequency, shape and slope, and C2C distance) is the extent of pre-ringing at a given off-axis angle.

I chose the values in the example somewhat arbitrarily. 1kHz is not an unusual place for a crossover, and LR8 is relatively shallow, so I guess I was choosing values that most would agree were not extreme.

One of us is missing something though (and I would not bet against you!).
What tool(s) are you using to simulate?

I am missing something: the general equations that link all the above factors. Hence my use of a simulator to demonstrate this in the specific, which I admit is not ideal.

Also, I have to reiterate, I’m not arguing that the degree of pre-ringing in this example is audible (in fact, based on temporal masking thresholds, I strongly suspect it is not).

What I’m saying is, firstly, that audibility thresholds for min-phase XO group delay are well-established (except in the bass), whereas thresholds for lin-phase XO pre-ringing are less well-established.

And secondly, that relying on on-axis cancellation to argue that lin-phase filters do not cause pre-ringing is problematic in terms of real-world speaker listening, where we (or at least I, lol) like to shift around a bit in the seat, sometimes sitting upright, sometimes slouching back, etc.

 

[dasdoing]

Well ok, but none of that changes the degree of pre-ringing introduced by the filter, nor the relationship between vertical axis and pre-ringing cancellation.

My purpose was to distil, not to mislead

Here's the same again, this time with a min-phase 2nd order roll-off at 40Hz.

On-axis (a small amount of pre-ringing is already visible here since the simulated woofer is no longer perfectly linear-phase at the XO point):

View attachment 102137

+15cm off-axis:

View attachment 102138

as I was just reading the following https://www.diyaudio.com/forums/mul...on-eq-fir-filtering-tool-186.html#post5042768

a perfect impulse gets ripples once you offset t=0....that doesn't mean it will ring

 

[dasdoing]

Perhaps an illustration will help:

View attachment 102132

Note the vertical scale... Edit, ah, @dc655321 already mentioned about the slope, thanks.

In this old article I use 2nd order Neville-Thiele crossovers if you want to dig in more as I show how the filters sum, etc.
The article summary is optimizing the XO design by measuring each driver, designing 3 way XO points well within the drivers published and measured specifications, linearizing each driver within that range, time alignment of the drivers, and then custom magnitude and excess phase correction to match the speaker system to the room. The article is 7 years old and technical improvements have been made since, but the principles still apply.

if this is the default in Acourate (which I would love to possess)?
what is that UB type?
I can't exactly reproduce that in Rephase but the reject filters come close

will read your article tomorrow. thanks

 

[andreasmaaan]

as I was just reading the following https://www.diyaudio.com/forums/mul...on-eq-fir-filtering-tool-186.html#post5042768

a perfect impulse gets ripples once you offset t=0....that doesn't mean it will ring

No, shifting a perfect impulse in time doesn’t cause it to ring, but a linear-phase high-pass or low-pass filter always rings (the one exception being when it is summed with a perfectly complementary filter and time-aligned, which is only the case on-axis).

 

[dasdoing]

No, shifting a perfect impulse in time doesn’t cause it to ring, but a linear-phase high-pass or low-pass filter always rings (the one exception being when it is summed with an equal opposite filter and perfectly time-aligned, which is only the case on-axis).

sure, wasn't denying that.
just saying that judging ringin looking at the impulse doesn't seam to work.
I always use group delay when creating FIR filters

 

[andreasmaaan]

sure, wasn't denying that.
just saying that judging ringin looking at the impulse doesn't seam to work.
I always use group delay when creating FIR filters

Yeh, but I’m using the step response merely to demonstrate that the ringing occurs.

Judgements concerning its audibility are another topic...

And group delay is yet another topic again

 

[dasdoing]

here is how it can be shown

brickwall HP (the crazy part below the wall is not important, but the negative exess group delay above the wall)

the LP looks similar, only mirrord. if you vector average the 2 you get the expected cancelation (perfect group delay)

then i moved the LP 1ms to the right and vector averaged again:

in my experience (with correction filters, not x-overs), in this frequency range 1ms of predelay is allready audible

 

[andreasmaaan]

here is how it can be shown

brickwall HP (the crazy part below the wall is not important, but the negative exess group delay above the wall)
View attachment 102146

the LP looks similar, only mirrord. if you vector average the 2 you get the expected cancelation (perfect group delay)

View attachment 102147
then i moved the LP 1ms to the right and vector averaged again:

View attachment 102148

in my experience (with correction filters, not x-overs), in this frequency range 1ms of predelay is allready audible

Is this modelling two speakers, with 1ms representing the interchannel delay caused by moving the LP to one side (ie closer to one speaker)?

 

[dasdoing]

Is this modelling two speakers, with 1ms representing the interchannel delay caused by moving the LP to one side (ie closer to one speaker)?

this is when the two filters are offset by 1ms = 343mm......meaning one must travel that 343mm longer then the other in the air. bad example I guess, because that would be meters(?) above, or below the speaker

 

[dc655321]

I am missing something: the general equations that link all the above factors. Hence my use of a simulator to demonstrate this in the specific, which I admit is not ideal.

Perhaps @mitchco could be persuaded to make an off-axis measurement of his system to evidence the effect?
Color me still skeptical this is problematic in a well-designed, mixed-phase system. Happy to learn tho!

 

[andreasmaaan]

Color me still skeptical this is problematic in a well-designed, mixed-phase system.

I totally agree (and would say the same about any well-designed, conventional, minimum-phase system, also).

Really what I’m warning against is the assumption that steep linear-phase filters don’t cause audible problems, which has not been demonstrated experimentally (to my knowledge).

 

[dc655321]

Perhaps @mitchco could be persuaded to make an off-axis measurement of his system to evidence the effect?

@andreasmaaan: I finally remembered that @mitchco had already provided such a measurement some time ago.

FYI: created my own sim in Python, but cannot simply share on this platform (it's a Jupyter notebook).
Result was (naturally) similar to your simulation.

 

[MRC01]

Does anyone know what the cheapest pair of full audible range 20hz - 20khz speakers there are? It seems like every pair of speakers I can find that can play the full 20hz - 20khz range cost tens of thousands of dollars.

A Neumann KH-420 or possibly a Genelec, will get you close for <€8000 a pair, including amplifiers.

Still pretty expensive. Also I already have amps so passive is better.

Whatever the cost, most of it comes from handling frequencies below 30 Hz. There's no such thing as "cheap" for any speaker that covers that range decently. If you want "cheap", consider speakers that are accurate above 30 Hz. Amir has reviewed a few here.

 

[andreasmaaan]

FYI: created my own sim in Python, but cannot simply share on this platform (it's a Jupyter notebook).
Result was (naturally) similar to your simulation.

Well done That would be beyond me (I used VituixCAD for the example btw, which I forget to mention earlier).

@andreasmaaan: I finally remembered that @mitchco had already provided such a measurement some time ago.

Thanks, that’s both interesting and helpful. @mitchco, did you happen to take vertical off-axis step response measurements? Although a little (absolutely benign) pre-ringing is visible in the horizontal off-axis measurements, you’d expect the verticals to be where it really showed up, as vertical displacement affects the relative distances from each driver’s acoustic centre to the mic much more.

Also, I’m curious to know roughly how steep the slopes were that you used there in the linear-phase crossovers?

 

[Helicopter]

MiniDSP and These.

 

[dasdoing]

if this is the default in Acourate (which I would love to possess)?
what is that UB type?
I can't exactly reproduce that in Rephase but the reject filters come close

View attachment 102143

will read your article tomorrow. thanks

I ended up using those. started with 300, went to 180....could hear pre-ringing while throwing a pulse at them. couldn't hear any with those 72 one, and since the FR didn't differ much, I stopped testing there and used those. compared against MP 48dB/oct filters this filled up half the hole I had at the crssover overlap:

 

[mitchco]

Thanks for posting @dc655321 I hope people notice in the link that was 3 feet off axis to the left and then 3 feet off axis to the right, covering a 6 ft horizontal span in a 9ft equilateral triangle and measuring both speakers in each location. The "only"reason there is preringing measured (not heard ;-) is because I am using 131,072 tap linear phase filter with 800ms of excess phase correction at 10 Hz. i.e. I have settled on a bit narrower sweet spot (like a 3ft window on the couch just for me That's why I get no low frequency reflections after the direct sound or at least the ideal "minimum phase" response.

Here is another system that I measured in 14 different locations across a 6ft x 2ft grid area of a 3 way system, again in a 9ft equilateral triangle. This system has digital XO points at 800 Hz and 5 kHz (and a large CTC gap between the midrange horn in the middle of cabinet and HF horn sitting on top - I should find a pic as it goes against conventional wisdom, but the measurements and listening say otherwise.

Virtually no preringing before t=0. This was using Neville Thiele 2nd order digital XO (@dasdoing - note it is the "type" of XO that makes the difference, not necessarily the slope) which is already posted in this thread. Additionally each driver was linearized and time aligned. Then room correction was applied, not as many filters taps nor as much low frequency excess phase correction as the previous example.

Note that there is (way) more variation in the reflected sound off the rear wall than anything relative to preringing...

When I sit on the couch and lean forward or back while listening, there is virtually no change in tonal response. Nor do I expect there to be. So I have not bothered to take any vertical measurements. Maybe the next time I get the mic out... Note that the measurements do cover a 2ft lean forward and back of MLP...

Wrt to preringing, I go into detail in my book to show various levels of preringing, how to get it and what to listen for. Again, based on my tests correlating measurements with listening, it takes a great deal to be audible. Orders of magnitude greater than anything presented in this thread. For me it is a non-issue.

 

[mitchco]

I ended up using those. started with 300, went to 180....could hear pre-ringing while throwing a pulse at them. couldn't hear any with those 72 one, and since the FR didn't differ much, I stopped testing there and used those. compared against MP 48dB/oct filters this filled up half the hole I had at the crssover overlap:

View attachment 102514

Try it with music too