World's Best Midranges - Shocking Results & Conclusions.

diyAudio Moderator
Joined 2008
Paid Member
Well, I don't remember which drivers JonBocani tested with, but it was two totally different drivers in size.

I tested if I could hear difference of a Seas 6.5 inch and Visaton AL170. Paper vs aluminium, similar driver size, same box, both crossed 1.8kHz, similar equalised, with tweeter and without. To be honest, I could not hear any difference
I was only talking about this the other day.. One of the issues here has to be that different drivers breakup differently. Multi-way is normally about using a driver arrangement only within a range where it can perform as needed, then crossing. Drivers tend to misbehave at breakup so unless it is dealt with properly, it is difficult to make fair comparisons.

In addition to breakup there is lobing due to cone size. Each driver requires different treatment. The best way to see what is happening for both issues is to look at polar plots. The peak itself isn't really the issue, and by that I mean it can be equalised, but there is more happening in that region. It is necessary to specify that some frequency is a problem, and make sure it is brought down to a low level. This means that the cross is started earlier.. how early can be traded off against how sharply it is crossed.

If the remaining band is working to specifications on all axes, then the response should be equalised to meet the design. Simple driver response resonances happen, but they can be fixed with EQ. If not done then there won't be a fair comparison as you will just be listening to the different responses. There is also no benefit to the system response being determined by chance due to the driver's own peaks and dips.

When a waveguided system is crossed, lobing is taken advantage of and this means crossing close to breakup. One option you'll often see used is a paper cone, which is known for its damping properties and its graceful breakup characteristics. Some feel that paper breaks up too readily, but perhaps if you are going to reach up into this region without needing this directivity then you are already choosing a compromise.

An example crossing in the breakup region: you choose to cross a woofer where the directivity is right, and it is just next to an early breakup mode. A suitable driver is chosen. The slope of the cross is selected to reduce the level of the next, more problematic breakup mode. You have already decided how low this frequency must be reduced in level. A notch filter might be used to help, but the goal is to have a smooth rolloff, so any filter may be used as long as this goal is achieved.

So when someone says that drivers sound different, one of the first things I'd look at is the design and how they're used. There are many reasons why you might find that the above conditions have been violated, and if they have then that becomes the new issue.
 
If I think about that a convolution (counter acting in time domain?) somehow the issues I don't think it can because the diffraction at baffle edge and the driver emits the sound at different time. I mean the driver would have to emit a "counter smear" for the diffraction to cancel out, but this only affects one location in space and everywhere else the response gets even weirder.

Hi tmuikku, here's my take...

Yes, that's what FIR can do, emit the "counter smear" later in time from the driver to offset the diffraction. So it is entirely possible to nearly eliminate the diffraction at a particular xyz location.

Admittedly, the diffraction and counter smear come from different physical locations; the diffraction coming from the baffle edges, the horn reflections, or even the edge of the driver itself, etc.
While the 'counter smear' comes from the driver itself.
But, it appears the triangulation distance errors between the xyz tuning spot, and other listening locations, are relatively minimal compared to the gains of having offset the diffraction in time.

This is the key point imo....that careful work can be done to reduce most of a speaker's direct diffractions/reflections, both on-ax and off-ax.
Again, because the benefits offsetting diffraction in time, dominate the less than perfect offset summations over area....

I've kinda come to consider diffraction as a "minimum phase byproduct output of the driver", because it directly follows what the driver is putting out.
But as we all well know, this 'direct byproduct output' is delayed in time.
So really, we need IIR EQs delayed in time to counter the min-phase byproduct output.

Which is exactly what FIR can provide.
I think we most often think of FIR's principal asset as the ability to adjust phase independently of frequency magnitude, ....and rightly so.
But it's good to remember it also can also handle larger time domain problems. In fact, if phase audibility is eventually proven to be less than significant, it might become FIR's main asset.

I think it's all kin to what Dave G is doing with Temporal EQ. Just fixing time domain problems to whatever extent they can be fixed fairly uniformly across polars.
 
Convolution's result and quality will be determined by the filter(s) quality and the experience of their's builder: It's your decision if you generate the filter(s) from a single measurement or a set of measurements which then you average. It's your decision if you measure at 5cm from the driver (excluding the baffle diffraction) or at 1m (including it). It's your decision wheter you window out the room response (by only using the part of the impulse response before any room artefact occurs, usually the floor reflection), or if you include part of the room response or all of it (which usually isn't such a good idea).



In a real-world multiway system you would filter/convolve all of the single ways to perform nicely in terms of theirs target SPL response and in terms of theirs target relative phase. And then you would apply another convolution over the whole system. And you would have to, well knowing the limits, compromise on all of these filters.



DSP/convolution is not the allround magic wand of speaker builders, but it is a very powerful tool. And as with every powerful tool, when not well applyied, you may distroy the whole world. So use it well and decently.


Very much agree with all said....

Other than I've found no need to apply another convolution over the whole system, after having convolved each single way separately.
In fact, i feel if a whole system convolution were needed, i failed somewhere in the single way process.

But that said, I'm speaking only about a speaker's quasi anechoic tuning.
I don't believe FIR is appropriate for room tuning, as I don't believe in special sweet spot tuning.
 
I was only talking about this the other day.. One of the issues here has to be that different drivers breakup differently. Multi-way is normally about using a driver arrangement only within a range where it can perform as needed, then crossing. Drivers tend to misbehave at breakup so unless it is dealt with properly, it is difficult to make fair comparisons.

The comparison was using DSP. With passive crossover the crossover operating points are not static. Even though the crossing is done away from resonance I always dealt with that frequency as the sound will be better that way.

I will point out 2 things that are not well accepted in blind listening:

1. Most people don't have the skill to perceive differences in an instant. Many people think that A is better than B but after a long time decided that B is better. This means that just because we think that A and B sound similar it doesn't mean it is okay to go with any of them (e.g. the cheaper one). Has to be critical/careful here.

2. GIGO, the weakest link concept. I have a speaker that I retain simply for testing amplifiers and I have amplifier that I retain simply for testing speakers. Some good characters in an amplifier cannot be heard through most speakers and some good characters in a speaker cannot be heard through most amplifiers. In this driver test, a DSP was used, which means we don't know or might not be able to hear what is "hidden" or masked by the electronics.
 
Wow - I never heard or read about this. Do you possible have som external references/research about this technique?

//


This paper talks about it for horn reflections. https://www.fulcrum-acoustic.com/wp...dspeaker-transient-response-with-dsp-2005.pdf

Fulcrum-acoustic's website has a bunch of less technical marketing info on it, termed Temporal EQ.

What i've come to realize/believe, is it doesn't matter whether we are talking horn relections, baffle edge diffraction, or whatever speaker-produced non-desired delayed output..........
........as long as they directly and consistently correlate to the original driving output....they are candidates for systematic correction in time.
 
I think in a conversation that transcends the types, we need to think in fair terms.. and besides, I obtain very similar results with both kinds. Necessity has always been the driver of that, ie the need to fix what can be heard.

I only know how the DSP works but have no experience with it. Sometimes a driver is so smooth that simple crossover will work just fine. But most of high-end midrange drivers are not paper cone and they are difficult to work with. Even if they are smooth they don't sound natural (to my ears).

I have an amplifier that is so transparent that it can portray details such as emotions or natural voice. I often had a goosebumps when listening to certain voice recordings (as if I'm not listening to a recording) but that can happen only with paper cones! Sometimes I have a thought, why not just live with the best paper cone available :)
 

TNT

Member
Joined 2003
Paid Member
This paper talks about it for horn reflections. https://www.fulcrum-acoustic.com/wp...dspeaker-transient-response-with-dsp-2005.pdf

Fulcrum-acoustic's website has a bunch of less technical marketing info on it, termed Temporal EQ.

What i've come to realize/believe, is it doesn't matter whether we are talking horn relections, baffle edge diffraction, or whatever speaker-produced non-desired delayed output..........
........as long as they directly and consistently correlate to the original driving output....they are candidates for systematic correction in time.

The whole Abstract for that article seem to contrast to your position... e.g.: "Some of these behaviors are time-variant, nonlinear, or spatially variable and are not good candidates for digital correction."

and in the conclusion:

"A transducer designer who expects his designs to be used in systems that make use of these preconditioning techniques may significantly change his focus in his development work. Rather than optimizing the uncorrected performance of a transducer, he might concentrate on maximizing the loudspeaker’s linearity, time invariance, and directional consistency...." (my bold)

So, no, diffraction - as it is a spatial phenomena, cant be corrected by DSP. I remain in this understanding until someone can come with better proof.

//
 
... Other than I've found no need to apply another convolution over the whole system, after having convolved each single way separately.
In fact, i feel if a whole system convolution were needed, i failed somewhere in the single way process.

But that said, I'm speaking only about a speaker's quasi anechoic tuning.
I don't believe FIR is appropriate for room tuning, as I don't believe in special sweet spot tuning.
You are right. For a productive, "mature" multiway system there should indeed be no need for an additional, overall convolution including the whole system. By the way my final multiway system had no overall filtering.

But overall convolution is a nice option: I wrote this stuff about the overall filtering/convolution from a reality as an ever-tweaking DIY-er: Whenever I changed someting with the baffle e.g. while tweaking for an improved polar behavior, it was more convenient to change one single overall filter than three filters for the distinct ways.

And by the way: I always null out the part of the measured impulse containing room information, in order to simulate an anechoic condition. I want to influence on the speaker, not on the speaker withing my test surrounding. The corresponding loss of information for the lower frequencies can then be synthesized by appropriate models. And neither for dayly listening I would include the room response within any DSP. Because it completely messes up the direct sound.
 
... So, no, diffraction - as it is a spatial phenomena, cant be corrected by DSP. I remain in this understanding until someone can come with better proof. ...

digitalthor said:
... maybe some kind of sound manipulation can do trikery with the diffraction too..... still need proof though ...

As mentionned quite a few times before, a near-perfect diffraction cancelling result and technique is only possible for the single point in space from where the measurement for the correction filter has been made. For this very special condition, it works, must work. It is not a matter of beleive or proofs, it's a matter of the basic principles of the underlying processes.You may eventually once again peep into this sequence with which I tried to shows how and why it works:

Four pictures ...

So much for the very special, singular point in space. Not shure if this until now is evident to you. I hope so. So leaving this special case, and now having a look at the whole space instead, where you refer your doubts into and where most of the audio takes place ...

For the much more important, and general situation of any other point in space, baffle diffraction cannot be completely cancelled out by one and the same filter. Nobody pretends this, and nobody will ever try to prove such a nonsense. Because at every other point in space the measured impulse from an always more or less directional speaker system is uniquely different. And so would be the derived filters.Therefore, you will never ever cancel a diffraction near-perfectly at more than one point in space at the same time.

You may instead very well improve on diffraction artefacts within a specific compartment of the space by DSP. E.g. by averaging some measures from out of the compartment where later some ears might be listening to audio material. And then build a filter from this average. Knowingly that then you encounter the holy mess of compromising, instead of reaching perfection. Hopefully compromise for the better, because you can also perfectly mess up things by DSP if you don't know what you are doing.

All this is not a flaw of convolution and DSP as such. That's audio in space ... Compromising ... Such is life ...
 
The whole Abstract for that article seem to contrast to your position... e.g.: "Some of these behaviors are time-variant, nonlinear, or spatially variable and are not good candidates for digital correction."

and in the conclusion:

"A transducer designer who expects his designs to be used in systems that make use of these preconditioning techniques may significantly change his focus in his development work. Rather than optimizing the uncorrected performance of a transducer, he might concentrate on maximizing the loudspeaker’s linearity, time invariance, and directional consistency...." (my bold)

So, no, diffraction - as it is a spatial phenomena, cant be corrected by DSP. I remain in this understanding until someone can come with better proof.

//

Is diffraction time-variant? I can't see how.
What are the "some of these behaviors" referred to in the Abstract, that are time-variant, nonlinear, or spatially variable?
Exactly what behaviors that can't be corrected?
They certainly exist; let's just pls try to make sure we are clear on what can and cant be corrected.

Also, please note your 'in the conclusion' quote refers to transducer designers.... which imho, is saying they do not fully understand the end game.
If anything, that quote supports the efficacy of correction capability, saying transducers designers need to see past the usual uncorrected frequency response measurements..

I do get diffraction is spatially variable. And that spatially variable is simply not even close to being completely correctable with DSP.

That said, what i continue to measure and hear, is that the portion of diffraction that directly hits a fairly wide listening window, is absolutely improved by correcting it.

The dispersed diffraction, the uncorrectable spatial diffraction, the diffraction heading off into the room to be reflected back into the direct listening window ......
......is low enough in magnitude and audibility, that i think it becomes a big relative 'so what'.