I am considering a software project that would allow the careful optimization of crossovers that would use multiple placed microphone measurements from the actual drivers in situ of the assembled loudspeaker. The system would therefore not have the customary variables to deal with such as approximating diffraction, actual driver performance, etc.
The user may use measured values of components or combinations of components or input multi-element approximations for elements - ie a capacitor and a high value resistor in parallel to approximate a capacitor.
The idea would be to test a crossover through multiple measurements where the affect of the crossover can be seen on multiple measurements easily and quickly before it is actually assembled. The number of measurements would be practically unlimited.
In the case of a two way speaker:
The user would place a microphone in a specific location relative to the assembled speaker, take a measurement of the woofer, then the tweeter and have a curve pair for this mic location, then the user would move the microphone and repeat. This process could be repeated a practically unlimited number of times, giving a wide range of mic locations from which to use in the Xover design.
These measurements would be saved.
A crossover emulator would then be applied to the measurements where real elements could be measured or approximations could be used such as the mentioned above with the capacitor and resistor.
Small, medium and large signal measurements of both the Xover elements and the drivers could be made.
It would take a few hours to take, maybe 90 measurements of a woofer and 90 measurements of a tweeter (30 large signal, 30 medium signal, 30 small signal) and save them into a file. After these measurements were taken, the user could use approximated devices in an Xover simulator that would allow practically unlimited complexity.
After an Xover is designed, it could then be tested by measuring the impedance of the real crossover elements (possibly using small, medium and large signal excitations for low cost elements)
The resulting system would allow the testing of an xover system with nearly 100 % accuracy over a large number of locations relative to a speaker.
This would be a pre-build tool rather than a design tool. It would be the final step before assembly of an actual system.
The system would be written in VC6, and not require the latest gadgets/spyware from MS, and still run in later versions of their OS'.
Any ideas ? general thoughts ?
The user may use measured values of components or combinations of components or input multi-element approximations for elements - ie a capacitor and a high value resistor in parallel to approximate a capacitor.
The idea would be to test a crossover through multiple measurements where the affect of the crossover can be seen on multiple measurements easily and quickly before it is actually assembled. The number of measurements would be practically unlimited.
In the case of a two way speaker:
The user would place a microphone in a specific location relative to the assembled speaker, take a measurement of the woofer, then the tweeter and have a curve pair for this mic location, then the user would move the microphone and repeat. This process could be repeated a practically unlimited number of times, giving a wide range of mic locations from which to use in the Xover design.
These measurements would be saved.
A crossover emulator would then be applied to the measurements where real elements could be measured or approximations could be used such as the mentioned above with the capacitor and resistor.
Small, medium and large signal measurements of both the Xover elements and the drivers could be made.
It would take a few hours to take, maybe 90 measurements of a woofer and 90 measurements of a tweeter (30 large signal, 30 medium signal, 30 small signal) and save them into a file. After these measurements were taken, the user could use approximated devices in an Xover simulator that would allow practically unlimited complexity.
After an Xover is designed, it could then be tested by measuring the impedance of the real crossover elements (possibly using small, medium and large signal excitations for low cost elements)
The resulting system would allow the testing of an xover system with nearly 100 % accuracy over a large number of locations relative to a speaker.
This would be a pre-build tool rather than a design tool. It would be the final step before assembly of an actual system.
The system would be written in VC6, and not require the latest gadgets/spyware from MS, and still run in later versions of their OS'.
Any ideas ? general thoughts ?
I am not sure what data you are collecting from the microphone. Is it amplitude's and phase's frequency responses?
I'm not a speaker or crossover expert, at all. I think I "know" that crossovers primarily are used to route the correct frequency range to each speaker, hopefully resulting in the correct amplitude at each frequency. And I think that the phase relationships between the signals after the crossover might affect the sound (as might the phase relationships caused by relative driver placements). But once you get past the drivers, there are also lots (and lots) of other things that can affect the amplitude and phase of everything, maybe way more than the crossover can.
There are so many variables. Maybe I'm confused, but, how do you decide on the location(s) for which to optimize? Or do you have some criterion or "optimization metric" to use that is then automatically optimized (by setting crossover parameters) based on the data for ALL locations tested? (And would that then really be "optimal"? I guess it would all depend on the optimization criterion, and maybe some weighting scheme for the locations.)
And I wonder how sensitive the results might be to other factors. For example, if a person is then added to the room, would the optimized result then be changed (by "too much")? Similarly for drapes open/closed, doors open/closed, furniture moved, etc, and, speakers moved around. I guess I'm wondering if optimizing _anything_ based on listening LOCATION is worth doing, or is even practical to attempt (because I have no idea what the relative magnitudes of all of the effects of all of the many things involved might be).
And if you're going to that much trouble, wouldn't you want to also at least optimize the drivers' relative placements? (and maybe the speaker enclosure, and the amplifier, and the room, and...)
I'm not necessarily trying to discourage you. Your system would probably be a good tool to have, to measure, and to try to discover good ways to try to optimize, the sound. It might also be a useful tool for evaluating speaker designs and placements, at the least.
You might also need to worry about how to make the microphone locations accurate and repeatable. Would you not need the exact-same location tested for all drivers at all power levels? But I guess that could be handled by fixing the mic location and then measuring all drivers at all power levels before moving the mic again.
For lots of measuirement locations, just setting the microphone locations, and then getting them all into the software, sounds like maybe it should be automated, with some kind of automatic (x,y,z) coordinates' position sensing. Might be a can of worms, there. If you do that, maybe also consider positioning it automatically, too. Similar to a CNC table, maybe you could use a large set of fixed, far-apart rails, on the floor (or the ceiling, so you wouldn't have to remove all of the furniture), carrying the other horizontal axis' movable perpendicular rail, which would carry the movable vertical z axis positioner, with servos or steppers running them all, under control of your computer. Making it "room-sized" might be an "interesting" project. But I'm sure there must other more-viable alternatives.
Sorry for blathering-on about all of that. I can see wanting to get the amplitude vs frequency optimized, and wanting to get the phase-alignment (time-alignment) optimized. I guess maybe I just don't understand how, or how well, using multiple locations, in a real room, is going to work, or what it will buy for you. I can imagine that it MIGHT give great results, and can easily believe that it should be better than doing nothing similar at all. But I don't know, for example, whether or not it would be almost as good to just measure at one location (OK, maybe "several" locations), far-enough from the speakers, either outside or in an anechoic chamber (or maybe even just in the room in which the speakers will be used).
Tom
I'm not a speaker or crossover expert, at all. I think I "know" that crossovers primarily are used to route the correct frequency range to each speaker, hopefully resulting in the correct amplitude at each frequency. And I think that the phase relationships between the signals after the crossover might affect the sound (as might the phase relationships caused by relative driver placements). But once you get past the drivers, there are also lots (and lots) of other things that can affect the amplitude and phase of everything, maybe way more than the crossover can.
There are so many variables. Maybe I'm confused, but, how do you decide on the location(s) for which to optimize? Or do you have some criterion or "optimization metric" to use that is then automatically optimized (by setting crossover parameters) based on the data for ALL locations tested? (And would that then really be "optimal"? I guess it would all depend on the optimization criterion, and maybe some weighting scheme for the locations.)
And I wonder how sensitive the results might be to other factors. For example, if a person is then added to the room, would the optimized result then be changed (by "too much")? Similarly for drapes open/closed, doors open/closed, furniture moved, etc, and, speakers moved around. I guess I'm wondering if optimizing _anything_ based on listening LOCATION is worth doing, or is even practical to attempt (because I have no idea what the relative magnitudes of all of the effects of all of the many things involved might be).
And if you're going to that much trouble, wouldn't you want to also at least optimize the drivers' relative placements? (and maybe the speaker enclosure, and the amplifier, and the room, and...)
I'm not necessarily trying to discourage you. Your system would probably be a good tool to have, to measure, and to try to discover good ways to try to optimize, the sound. It might also be a useful tool for evaluating speaker designs and placements, at the least.
You might also need to worry about how to make the microphone locations accurate and repeatable. Would you not need the exact-same location tested for all drivers at all power levels? But I guess that could be handled by fixing the mic location and then measuring all drivers at all power levels before moving the mic again.
For lots of measuirement locations, just setting the microphone locations, and then getting them all into the software, sounds like maybe it should be automated, with some kind of automatic (x,y,z) coordinates' position sensing. Might be a can of worms, there. If you do that, maybe also consider positioning it automatically, too. Similar to a CNC table, maybe you could use a large set of fixed, far-apart rails, on the floor (or the ceiling, so you wouldn't have to remove all of the furniture), carrying the other horizontal axis' movable perpendicular rail, which would carry the movable vertical z axis positioner, with servos or steppers running them all, under control of your computer. Making it "room-sized" might be an "interesting" project. But I'm sure there must other more-viable alternatives.
Sorry for blathering-on about all of that. I can see wanting to get the amplitude vs frequency optimized, and wanting to get the phase-alignment (time-alignment) optimized. I guess maybe I just don't understand how, or how well, using multiple locations, in a real room, is going to work, or what it will buy for you. I can imagine that it MIGHT give great results, and can easily believe that it should be better than doing nothing similar at all. But I don't know, for example, whether or not it would be almost as good to just measure at one location (OK, maybe "several" locations), far-enough from the speakers, either outside or in an anechoic chamber (or maybe even just in the room in which the speakers will be used).
Tom
Last edited:
The measurements would include both amplitude and phase.
The idea here is to simulate only the crossover network when everything else is done.
In some cases a careful mic arrangement may be required. If you are building a small two way speaker, then a large number of mic locations at and around the axis of the speaker +/-10 degrees is a good experiment - its very repeatable if you start with good drivers. I've done this and tested the repeatability of the experiment. Any experiment must of course be tested for repeatability as well as being relevant. The experiment repeatability depends on the speaker configuration as well as the mic array locations and number. Three locations in front of a planar speaker will never be repeatable.
A large number of mic measurements will also filter out room affects, giving the time period of the impulse (gating) necessary to determine response with sufficient resolution to a low enough frequency. In a typical 20 X 20 X 8 foot room, you can work with a 6.5 inch/ two way. You could of course work with the upper Xover frequency of a 3 way. For the lower Xover, close mic locations and the knowledge that the drivers are omni-directional should suffice.
A set of measurements with mic locations at the listening area for each driver may also be made.
After the measurements are done a precise simulation of the crossover is possible, I believe. The impedance of real elements may be measured and used.
As a hobby builder, it makes sense to build enclosures that will predictably have very low diffraction. For commercial builders this is not viable, the square corners are all that is viable. In this case more mic locations would be used.
The idea here is to simulate only the crossover network when everything else is done.
In some cases a careful mic arrangement may be required. If you are building a small two way speaker, then a large number of mic locations at and around the axis of the speaker +/-10 degrees is a good experiment - its very repeatable if you start with good drivers. I've done this and tested the repeatability of the experiment. Any experiment must of course be tested for repeatability as well as being relevant. The experiment repeatability depends on the speaker configuration as well as the mic array locations and number. Three locations in front of a planar speaker will never be repeatable.
A large number of mic measurements will also filter out room affects, giving the time period of the impulse (gating) necessary to determine response with sufficient resolution to a low enough frequency. In a typical 20 X 20 X 8 foot room, you can work with a 6.5 inch/ two way. You could of course work with the upper Xover frequency of a 3 way. For the lower Xover, close mic locations and the knowledge that the drivers are omni-directional should suffice.
A set of measurements with mic locations at the listening area for each driver may also be made.
After the measurements are done a precise simulation of the crossover is possible, I believe. The impedance of real elements may be measured and used.
As a hobby builder, it makes sense to build enclosures that will predictably have very low diffraction. For commercial builders this is not viable, the square corners are all that is viable. In this case more mic locations would be used.
Hi Doug,
I appreciate you taking the time to try to educate me.
Am I correct in assuming that the measurements will be able to be used to generate simulation models of the drivers, "in situ", i.e. as the drivers will be installed and used? To get simulation models, (I am just guessing) you would be fitting parameters into the equations for the drivers' sound-propogations and their propogated sounds' interactions, based on the dataset, or something like that. Am I even close?
Once you have a way to use the data, or simulation models generated from the data, with simulation models of crossovers, you could then hope to be able to find crossover parameters that are optimal in some sense, as defined by some function of the simulated responses of the drivers. Correct?
Finding the optimal crossover architecture or topology and then finding the optimal component parameters seems like a difficult problem, at least to someone like me with so little knowledge. Are you planning to try to automate the optimization process (at least once the basic crossover architecture/topology is chosen)? OR, have I missed the point and there is some way to use the driver/propogation models to directly calculate a desirable set of transfer functions for the paths through the crossover?
Here is something that I just remembered: LTSpice (free from linear.com) can use WAV files as both inputs and outputs! So it would be possible to simulate a crossover with actual sound as the input and get the separate sound outputs for each driver. And component parasitics, etc, can be modeled pretty well in LTspice (even with ESR as a function of frequency and temperature for electrolytics, for example, with a little work). I wonder if that might be useful. It could even be used for some sort of iterative optimization process, where LTspice automatically steps through component values, creates WAV files for each driver with the current component value (which is not acheived in real time, unfortunately), your computer plays the WAV files through the drivers and records mic measurements, then it steps to the next component value, and so on. You could probably even use optimization software or techniques to "manually" pick the subsequent component values, to try to be converging (toward optimality) based on some optimization criteria.
But I realize that you will probably have a much quicker and better way to do the crossover optimization, "off line", since you will have driver/propogation models. But maybe the LTspice WAV outputs from "candidate" crossovers could at least be used for testing them, so you didn't have to build each one that you wanted to actually try. Just a thought!
Thanks again for taking the time to try to educate me.
Tom
I appreciate you taking the time to try to educate me.
Am I correct in assuming that the measurements will be able to be used to generate simulation models of the drivers, "in situ", i.e. as the drivers will be installed and used? To get simulation models, (I am just guessing) you would be fitting parameters into the equations for the drivers' sound-propogations and their propogated sounds' interactions, based on the dataset, or something like that. Am I even close?
Once you have a way to use the data, or simulation models generated from the data, with simulation models of crossovers, you could then hope to be able to find crossover parameters that are optimal in some sense, as defined by some function of the simulated responses of the drivers. Correct?
Finding the optimal crossover architecture or topology and then finding the optimal component parameters seems like a difficult problem, at least to someone like me with so little knowledge. Are you planning to try to automate the optimization process (at least once the basic crossover architecture/topology is chosen)? OR, have I missed the point and there is some way to use the driver/propogation models to directly calculate a desirable set of transfer functions for the paths through the crossover?
Here is something that I just remembered: LTSpice (free from linear.com) can use WAV files as both inputs and outputs! So it would be possible to simulate a crossover with actual sound as the input and get the separate sound outputs for each driver. And component parasitics, etc, can be modeled pretty well in LTspice (even with ESR as a function of frequency and temperature for electrolytics, for example, with a little work). I wonder if that might be useful. It could even be used for some sort of iterative optimization process, where LTspice automatically steps through component values, creates WAV files for each driver with the current component value (which is not acheived in real time, unfortunately), your computer plays the WAV files through the drivers and records mic measurements, then it steps to the next component value, and so on. You could probably even use optimization software or techniques to "manually" pick the subsequent component values, to try to be converging (toward optimality) based on some optimization criteria.
But I realize that you will probably have a much quicker and better way to do the crossover optimization, "off line", since you will have driver/propogation models. But maybe the LTspice WAV outputs from "candidate" crossovers could at least be used for testing them, so you didn't have to build each one that you wanted to actually try. Just a thought!
Thanks again for taking the time to try to educate me.
Tom
Last edited:
The program, if it is a good idea, could have features added to it as time progresses such as optimizers or radiation pattern modeling.
I pit this post up because its really just a seed idea, I haven't really researched the other stuff out there that does this.
As far as the *wav file idea goes, the program would have an impedance measuring device that would measure the real impedance of each component to be put in the crossover if desired. Good components can be modeled with a resistor + capacitor or resistor + inductor for audio frequency applications. Of course, for much higher frequencies complexity needed approaches that of distributed systems.
I pit this post up because its really just a seed idea, I haven't really researched the other stuff out there that does this.
As far as the *wav file idea goes, the program would have an impedance measuring device that would measure the real impedance of each component to be put in the crossover if desired. Good components can be modeled with a resistor + capacitor or resistor + inductor for audio frequency applications. Of course, for much higher frequencies complexity needed approaches that of distributed systems.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.