I am trying to get my speaker knowledge on a higher level. Until now, I usually implemented crossovers considering the electric slope only (plus maybe some delay). Is there any writeup on how to design the crossovers considering the acoustic slopes (e.g. the proper way)? I think I understand some of the basic principles, but I would definitely like to get some tips and tricks where to start. I can use REW and both single channel and dual channel measurements. The DSP I have now is limited to one global stereo EQ and 7 fully parametric EQs + HP and LP filters - Bessel, Butterworth, L-R and max 24 dB / oct slope.
I tried today the Harsch alignment (electric) for both MEH prototypes and it seems the transients are better than with LR24 I had before. Might be just expectation bias, I need to confirm that with measurements later. That was the motivation to get deeper into it. Any input will be appreciated.
I tried today the Harsch alignment (electric) for both MEH prototypes and it seems the transients are better than with LR24 I had before. Might be just expectation bias, I need to confirm that with measurements later. That was the motivation to get deeper into it. Any input will be appreciated.
The first thing is to calculate the slopes you need. These will be the same whether you use DSP or any other type of crossover. This is usually done with a crossover simulator and a collection of measurements.
Harsch is not a good modern alternative. It shouldn't be necessary on an MEH when done the Danley way, and it does not consider the whole of a multi-way speaker, only one axis.
Harsch is not a good modern alternative. It shouldn't be necessary on an MEH when done the Danley way, and it does not consider the whole of a multi-way speaker, only one axis.
I did some more measurements yesterday - AMT crossed over to 2x3FE22 in a MEH. I get a nice overlap with the center frequency around 890 Hz - and the overlap before rolloff is around 300 Hz to both sides. I will make a new set of measurements and upload it here, so that we are talking about the real thing and it will be easier for me to understand it as well.
Using the Harsch XO (with delay) got me the best step response so far. With LR4 I got always some strange mess in the step response (sounding OK nevertheless).
So to find the slopes I need I would have to do full set of polar measurements. That is not really possible at the moment due to weather. I know it is not the proper way of doing it, but can this be simplified to single axis measurements only? Keeping in mind that it is not 100% correct, but to learn the process?
This is actually getting close to another recent topic, that I follow. (https://www.diyaudio.com/community/...y-to-start-when-using-dsp.383188/post-6944108)
I think I need to start there...lots of useful information.
Using the Harsch XO (with delay) got me the best step response so far. With LR4 I got always some strange mess in the step response (sounding OK nevertheless).
So to find the slopes I need I would have to do full set of polar measurements. That is not really possible at the moment due to weather. I know it is not the proper way of doing it, but can this be simplified to single axis measurements only? Keeping in mind that it is not 100% correct, but to learn the process?
This is actually getting close to another recent topic, that I follow. (https://www.diyaudio.com/community/...y-to-start-when-using-dsp.383188/post-6944108)
I think I need to start there...lots of useful information.
That other thread is in a state of an identity crisis at the moment. It should be covering the way to get your acoustic slope, but it is talking about ways to play with DSP hardware.
You need to draw a line for yourself. For example if you want to take shortcuts, there are things you can do to make them even shorter. There are so many different kinds of shortcut that you could have many conversations running at once.
If on the other hand you wanted to do a crossover in a more complete way (yes, you say here you do not want that yet).....then you wouldn't even be at the question of dsp yet. However use that knowledge to help you not get swept up into these enthusiastic discussions and lose your way
You need to draw a line for yourself. For example if you want to take shortcuts, there are things you can do to make them even shorter. There are so many different kinds of shortcut that you could have many conversations running at once.
If on the other hand you wanted to do a crossover in a more complete way (yes, you say here you do not want that yet).....then you wouldn't even be at the question of dsp yet. However use that knowledge to help you not get swept up into these enthusiastic discussions and lose your way
I guess for an example of a normal procedure, you decide a crossover frequency and notice where the cone breakup is. This tells you how steep to make your filter. Or maybe the upper cutoff is where the MEH null happens.. same thing.
Some people will come up with ideas like making the driver flat and then applying fixed filters. I don't know but half the time it sounds like they aren't even thinking about the abovementioned things... but it could be done that way if you apply it right.
I prefer not to do that, but to decide what I want the filters to do, and do them straight. I create a target curve. Sometimes that involves combining/subtracting curves in a simulator...
However some make a further shortcut of this by measuring again until they find the response they want.
This may not work for a full crossover if you choose to follow a power curve rather than a response curve. But you can still create a filter target.
Some people will come up with ideas like making the driver flat and then applying fixed filters. I don't know but half the time it sounds like they aren't even thinking about the abovementioned things... but it could be done that way if you apply it right.
I prefer not to do that, but to decide what I want the filters to do, and do them straight. I create a target curve. Sometimes that involves combining/subtracting curves in a simulator...
However some make a further shortcut of this by measuring again until they find the response they want.
This may not work for a full crossover if you choose to follow a power curve rather than a response curve. But you can still create a filter target.
Hi penlanj, hi Allen,
First, yes i'm one of the (many) folk who advocates making the driver flat and then applying fixed filters. What is entirely implicit in that procedure is having already determined the useful range of the driver. IOW, cone breakup on the upper end, or over-excursion on the lower end, aren't even issues. Because we've already chosen drivers and the range they are suited for, before beginning the flatten and apply fixed filters process. It's simply part of the prerequisite acoustic design process.
I think most everyone advocating making a driver flat and then applying filters, has also determined driver range suitability before beginning.
As far as acoustic vs electrical xover slope....and how to use DSP...
Well, we all know acoustic xover is what matters, and that electrical xover is the tool to achieve acoustic xover.
A foundational belief for me, is that acoustical xovers must end up fully complementarity, other than for deliberate purposes of off-axis steering etc.
Electric is simply what it takes to get there.....
In the other thread, I recommended the use of real time transfer functions for this task.
Please note i did not mention FIR, as indeed FIR is not appropriate for real time tuning simply because who ever heard of a real time FIR processor. The closest I've seen is from Linea Research that has on the fly LR24 linear phase xovers, and on the fly FIR high shelves. But who ever heard of on the fly FIR PEQ's?
Besides, the in-band and out-of-band driver flattening needs to be IIR to begin with.
Ok, how to reach a desired acoustic xover that is complementary ....
Here's an easy real time method that just comes down to twisting knobs while you measure.
With a dual channel FFT, first dial in PEQs and shelving filters to make the driver flat as far away from planned xover frequency as necessary to give good summation. Like folks have said, and octave or so.
Use pink noise/periodic pink and simply add filters setting frequencies, bandwiths/Q's, and gains, till happy.
Easy peasy, what you see is what you get.
Then, make the dual channel FFT's reference channel include the acoustic xover you want to achieve.
IOW, let's say you want to achieve a 24dB/oct acoustic LR xover at say 300 Hz for low-mid driver, to cross over to a mid driver..
Run the pink noise thru a 24 dB/oct LR low-pass at 300Hz and make that the reference channel input.
Next, measure the flattened driver's transfer function against that reference.
It will be immediately obvious a low-pass has to be added to the driver. Select among fixed-type low passes until the transfer function flattens out thru the desired summation range. Its easy and very clear to do.
When you get to the flattest transfer you can get, which is usually more than sufficient for good complementary acoustic summation throughout the desired range, either stop or play with a few more filters added onto the low-pass for further flattening
Do the mid driver the same way, with a 300 Hz LR 24/dB oct high-pass as reference channel.
Bingo, ......a 4th order acoustic complementary achieved, and already verified via measurements.
Real time, and IIR
And pls note, the desired acoustic xover could be first (or second order) and still use fixed type xovers, in summation with flattening EQs.
It's the sum of a xovers (high or lowpass) and the flattening EQs that ultimately create the electrical xovers. And that sum of filters never has a name...it's unique.
imho, the distinction between fixed-type, or rather named and unnamed xovers, is simply semantics, when the whole electrical xover package is considered.. .
My 2c
First, yes i'm one of the (many) folk who advocates making the driver flat and then applying fixed filters. What is entirely implicit in that procedure is having already determined the useful range of the driver. IOW, cone breakup on the upper end, or over-excursion on the lower end, aren't even issues. Because we've already chosen drivers and the range they are suited for, before beginning the flatten and apply fixed filters process. It's simply part of the prerequisite acoustic design process.
I think most everyone advocating making a driver flat and then applying filters, has also determined driver range suitability before beginning.
As far as acoustic vs electrical xover slope....and how to use DSP...
Well, we all know acoustic xover is what matters, and that electrical xover is the tool to achieve acoustic xover.
A foundational belief for me, is that acoustical xovers must end up fully complementarity, other than for deliberate purposes of off-axis steering etc.
Electric is simply what it takes to get there.....
In the other thread, I recommended the use of real time transfer functions for this task.
Please note i did not mention FIR, as indeed FIR is not appropriate for real time tuning simply because who ever heard of a real time FIR processor. The closest I've seen is from Linea Research that has on the fly LR24 linear phase xovers, and on the fly FIR high shelves. But who ever heard of on the fly FIR PEQ's?
Besides, the in-band and out-of-band driver flattening needs to be IIR to begin with.
Ok, how to reach a desired acoustic xover that is complementary ....
Here's an easy real time method that just comes down to twisting knobs while you measure.
With a dual channel FFT, first dial in PEQs and shelving filters to make the driver flat as far away from planned xover frequency as necessary to give good summation. Like folks have said, and octave or so.
Use pink noise/periodic pink and simply add filters setting frequencies, bandwiths/Q's, and gains, till happy.
Easy peasy, what you see is what you get.
Then, make the dual channel FFT's reference channel include the acoustic xover you want to achieve.
IOW, let's say you want to achieve a 24dB/oct acoustic LR xover at say 300 Hz for low-mid driver, to cross over to a mid driver..
Run the pink noise thru a 24 dB/oct LR low-pass at 300Hz and make that the reference channel input.
Next, measure the flattened driver's transfer function against that reference.
It will be immediately obvious a low-pass has to be added to the driver. Select among fixed-type low passes until the transfer function flattens out thru the desired summation range. Its easy and very clear to do.
When you get to the flattest transfer you can get, which is usually more than sufficient for good complementary acoustic summation throughout the desired range, either stop or play with a few more filters added onto the low-pass for further flattening
Do the mid driver the same way, with a 300 Hz LR 24/dB oct high-pass as reference channel.
Bingo, ......a 4th order acoustic complementary achieved, and already verified via measurements.
Real time, and IIR
And pls note, the desired acoustic xover could be first (or second order) and still use fixed type xovers, in summation with flattening EQs.
It's the sum of a xovers (high or lowpass) and the flattening EQs that ultimately create the electrical xovers. And that sum of filters never has a name...it's unique.
imho, the distinction between fixed-type, or rather named and unnamed xovers, is simply semantics, when the whole electrical xover package is considered.. .
My 2c
You do know that high order crossover filters are not necessary in an MEH, right? And only enough phase delay is needed in the crossover filters to put the lower frequency drivers in phase with the high frequency ones. By definition...that's first order.
Polar lobing is not an issue in an MEH and so the root reason for using higher order filters is not an issue for MEHs.
Most people also do not understand that the Danley designs do not use high order IIR filters, because there is no need for them in a horn that achieves "unity summation" of the polar acoustic output internally.
For instance, the SH-50 uses a "phase link" design--with the midrange covering only <1.5 octaves, but is there to spread the acoustic load to support higher SPL operation, and to link the phase of the woofer to the HF compression driver. Significant overlap of driver output exists between the driver sections in the SH-50.
Chris
Last edited:
Only some basic tips how to make correct acoustic slopes with DSP, REW and RePhase (just IIR)
https://www.diyaudio.com/community/...8-bigoldradial-hypexfa123.378343/post-6855000https://www.diyaudio.com/community/...8-bigoldradial-hypexfa123.378343/post-6856194
https://www.diyaudio.com/community/...8-bigoldradial-hypexfa123.378343/post-6855000https://www.diyaudio.com/community/...8-bigoldradial-hypexfa123.378343/post-6856194
That is a nicely written description of the process. You covered everything from what I can tell... I wish I had this in 2019 when I started (!)
Today, I am more likely to make a full set of comprehensive driver measurements and simulate the DSP filters in VituixCad2, and then after optimization, load them into hardware. But your writeup is very useful for people who want to do it by iterative measure/modify/listen technique.
j.
Hi Mark. Nice writeup. The method itself can work and that's important.. as long (as you say) you don't use the breakup region.
It is a good fit with what pelanj is looking to do at the moment, and could be worth a try.
The benefit of your method is that it saves you producing a target function in advance.
This method is more about the filters and not about the acoustics. Many prized this method in the '90s achieving perfect measured response and phase, but don't do it any more.
Thanks Jim, glad it made sense.
I view the real-time IIR technique as a way to get started with DSP tuning, that also helps show the relationships between electrical xovers & EQ's, and final acoustic outputs.
I'm kinda convinced the "iterative measure/modify/listen technique" only requires iterations when the directivity curve is unruly.
And that the iterations are more about pleasing the ear in the particular room environment due to off-axis anomalies.
I personally don't use the technique anymore, because having become comfortable electrical to acoustic relationships, i feel confident with automated processes that build a complete electrical xover & EQs package, to produce the desired acoustic response.....(which has the IIR work for driver flattening, and linear phase xovers, embedded into a FIR file.)
I do think to that simulations like VituixCad2 are a great way to go, when they accurately predict both on-axis and off-axis responses, after xovers and EQs are added.
Right now, i just take polar measurements, and either average a set of them for the reference to tune to, or chose a single axis measurement, if one appears to be a good center of gravity to tune to.
Thanks Allen, kind words.
May i ask what you mean by the method is more about the filters and not about the acoustics. Seems to me the method is completely acoustic response. (I think all methods are about acoustic response....)
To the extent the method is applied to only a single tuning axis, i can see how you might say that. But i think that doesn't have anything to do with the viability of the method itsef, that that's just about the viability of tuning to 'on-axis' only and applies to any method.
The biggest part of my recommendation is the real-time tuning part. This has vastly accelerated my learning about xovers, EQs, their interactions, and how final acoustic output is effected...... compared to running sine sweeps after every filter change and reiterating. So much can be gleaned so quickly with real-time measurements...both on and off axis.
I guess in the '90s there weren't any dual channel FFTs available (other than in the proaudio world, like Meyer's SIMM).
Apart from that recognition, what do you think the folks that left the method are doing today, that's substantially different other than also taking in off-axis?
I'd like to provide a slightly different viewpoint on this topic.
Some people make much to do about which filter is best, what crossover topology results in better summation and integration, or whether one should try and fit the overall acoustic response to a well-known response type as a combination of the driver's rolloff and electrical filters (analog, passive, DSP, or whatever you choose). But I think these arguments are down in the weeds, so to speak, and are losing sight of the over-arching end-goal. In the end it all of these boil down to the pursuit of the same primary goal: getting drivers to sum in-phase at and around the crossover point. How you do that is really up to you and there really is not one superior approach IMHO. There are other points to consider (e.g. the time domain response, crossover complexity, noise and losses, etc.) but these are somewhat lesser in importance. So, start with careful measurements, pay attention to the relative phase angle and driver summation in your crossover modeler, and feel free to use whatever approach gets you over the finish line.
Some people make much to do about which filter is best, what crossover topology results in better summation and integration, or whether one should try and fit the overall acoustic response to a well-known response type as a combination of the driver's rolloff and electrical filters (analog, passive, DSP, or whatever you choose). But I think these arguments are down in the weeds, so to speak, and are losing sight of the over-arching end-goal. In the end it all of these boil down to the pursuit of the same primary goal: getting drivers to sum in-phase at and around the crossover point. How you do that is really up to you and there really is not one superior approach IMHO. There are other points to consider (e.g. the time domain response, crossover complexity, noise and losses, etc.) but these are somewhat lesser in importance. So, start with careful measurements, pay attention to the relative phase angle and driver summation in your crossover modeler, and feel free to use whatever approach gets you over the finish line.
OK, but at what point in space? (i.e. tweeter height, mid height, on-axis, off-axis, etc)
It's whatever point you (the designer) choose, and strictly speaking you can choose only one. But this is a specious argument. It's the same for any crossover implementation whether passive, active DSP IIR or FIR because of non-coincidence of sources.
The designer/builder usually chooses a "design axis", and that is often on-axis with the tweeter at some distance (that depends on the size of the loudspeaker, larger speakers need more distance to integrate properly). I usually aim for at least 2m distance for example.
Last edited:
This is a great discussion, I'm dealing with similar issues as my system evolves. I am much less hands on to date with the acoustic issues than any of you guys, I'm using a MiniDSP SHD to fix various speaker (driver physical time alignment is not possible in this system) and room problems. (Learning to use that effectively is a story in itself.) Following with not a lot to add to the conversation at this point.
May I ask you to look at this measurement? It is in room with a crossover somewhere around 1 kHz (I do not remember which version this was). Could you please comment on this measurement pointing out what is wrong? Or if this kind of measurement is even usable for anything?
I am also getting some really strange measurements it Windows 10 using UMIK. In the IR display, I get impulses like 30 ms away from the main one, that does not make any sense. It is almost like a Windows 10 thing doing some strange compensations to the sound...
I am also getting some really strange measurements it Windows 10 using UMIK. In the IR display, I get impulses like 30 ms away from the main one, that does not make any sense. It is almost like a Windows 10 thing doing some strange compensations to the sound...