Hi,
I don't have an electronics background and pretty much everything I've learned about audio has been picked up here recently.
Tbh just over a week ago I thought XOs just "handed" a signal over to another driver at a certain frequency, I had no idea about the specifics of XO.
I now understand it more, you basically have to constantly destroy the signal above or below the frequency range you want for that driver?
As I think more about DSP, I'm starting to think that FIR filters are basically frequency "cancellers" or "boosters", you use the actual FR to map out the frequencies you have to constantly target to get the sound to your target dB for that frequency?
The science of it is beyond me and it helps me to think of complex systems in ways like this?
Does that make sense, or am I barking up the wrong tree?
Thanks.
I don't have an electronics background and pretty much everything I've learned about audio has been picked up here recently.
Tbh just over a week ago I thought XOs just "handed" a signal over to another driver at a certain frequency, I had no idea about the specifics of XO.
I now understand it more, you basically have to constantly destroy the signal above or below the frequency range you want for that driver?
As I think more about DSP, I'm starting to think that FIR filters are basically frequency "cancellers" or "boosters", you use the actual FR to map out the frequencies you have to constantly target to get the sound to your target dB for that frequency?
The science of it is beyond me and it helps me to think of complex systems in ways like this?
Does that make sense, or am I barking up the wrong tree?
Thanks.
Last edited:
i am a great fan of fullrange drivers corrected with EQ.
But I have to admit I also came across well made multi ways which did everything right.
Especially coax designs can sound as good as my favourite design.
I like fullranges because of their good behaviour in the time domain.
With lucky matching of these with current driven amps amplitude errors become astonishingly less important.
But I have to admit I also came across well made multi ways which did everything right.
Especially coax designs can sound as good as my favourite design.
I like fullranges because of their good behaviour in the time domain.
With lucky matching of these with current driven amps amplitude errors become astonishingly less important.
Thanks, the link states:
"These electrical components divide the signal from the amplifier and distribute the frequencies to the correct drivers"
I think this is the sort of oversimplification that was holding me back because now I don't think it actually divides/distributes the signal.
If I use a simple XO for a two-way, doesn't it take the entire signal and mirror it into two paths, the LF path and the HF path. The LF path has the HF energy in its signal targeted and destroyed. The HF path has the LF energy in its signal targeted and destroyed?
A crossover does not destroy the signal (unless there's a resistor, like for attenuating the tweeter).
Inductors and capacitors (to a good approximation) do not dissipate any energy, just move it around.
The driver does dissipate power, some of which creates the sound.
The crossover is a filter which works as a voltage divider that is frequency dependent.
The L and C can store energy, which, in the steady state, acts to drop voltage without dissipation.
Within the driver's pass band, the crossover attenuation is less. Outside the pass band, the attenuation is more.
Crossovers are tricky to design because of the low impedance of the driver, which requires the driver to be part of the filter.
Inductors and capacitors (to a good approximation) do not dissipate any energy, just move it around.
The driver does dissipate power, some of which creates the sound.
The crossover is a filter which works as a voltage divider that is frequency dependent.
The L and C can store energy, which, in the steady state, acts to drop voltage without dissipation.
Within the driver's pass band, the crossover attenuation is less. Outside the pass band, the attenuation is more.
Crossovers are tricky to design because of the low impedance of the driver, which requires the driver to be part of the filter.
You'll learn more about filter theory not looking for help with crossovers. Many speaker builders do not come from a background of electronics and signal processing even though they may be good at what they do.
The first part of a crossover should be the acoustic design, though you've made it clear this is not what you're asking in this case.
FIR is useful for the same reasons as other filter methods, it's a different way of doing it and the goals are the same in any case. One of the reasons people like to use it is so they can set phase independently in an arbitrary way. Others may not find this necessary.
The first part of a crossover should be the acoustic design, though you've made it clear this is not what you're asking in this case.
FIR is useful for the same reasons as other filter methods, it's a different way of doing it and the goals are the same in any case. One of the reasons people like to use it is so they can set phase independently in an arbitrary way. Others may not find this necessary.
They are guilty of oversimplification. The L and C are not like resistors, rather impedances, with voltage and current not "in phase".
This is why an L or C does not dissipate power, whereas an R does.
In a resistor, V = I x R where current and voltage are "in phase" with each other.
In a capacitor, I = C x dV/dt where current leads voltage by 90 degrees.
In an inductor, V = L x dI/dt where voltage leads current by 90 degrees.
The impedance of an L or C does vary with frequency. Specifically, Z(L) = jwL and Z(C) = 1/jwC where w = 2 x Pi x f
So for an L, the impedance is proportional to f. But for a C, the impedance is inversely proportional to f.
What I don't know about crossovers would fill a book, but designers may use crossovers to 'shape' the drivers' responses and sound.
This might be something as simple as an "L-Pad" to reduce the tweeter output, or selecting a combination of parts to correct impedance issues or driver breakup. Or it could be that designer wants to 'voice' a speaker in a certain way. For example, Paul Carmody - whose designs are widely built and well regarded - originally published two versions of his "Classix II" bookshelf: same drivers, but they had different crossover parts so that one was voiced for classical music and the other for pop/rock.
Geoff
This might be something as simple as an "L-Pad" to reduce the tweeter output, or selecting a combination of parts to correct impedance issues or driver breakup. Or it could be that designer wants to 'voice' a speaker in a certain way. For example, Paul Carmody - whose designs are widely built and well regarded - originally published two versions of his "Classix II" bookshelf: same drivers, but they had different crossover parts so that one was voiced for classical music and the other for pop/rock.
Geoff
Thanks, how did Paul Carmody do that, was it just by tweaking the crossover frequency and slope?
I wouldn't use this term. It's all about frequency dependent attenuation - be it caps, indictors or IIR as well as for FIR. Filters are frequency dependent attenuation. Full stop 🙂
//
https://sites.google.com/site/undefinition/bookshelf-speakers/diy-classix
My mistake, Paul called them 'hi fi' and 'laid back', I think.
Other designers do similar things with components to 'shape' the sound but don't always go into detail in their write ups.
I've done projects with the same drivers but different crossovers, the difference can be remarkable. I did test builds of Michael Chua's "Lark" and later "Lark SM" speakers - same drivers but quite different crossovers. The first sounded very good; the second, excellent. The crossover schematics need to be purchased from Mr Chua, but his write-ups (Google 'Ampslab speakers') indicate the differences and improvements.
Geoff
My mistake, Paul called them 'hi fi' and 'laid back', I think.
Other designers do similar things with components to 'shape' the sound but don't always go into detail in their write ups.
I've done projects with the same drivers but different crossovers, the difference can be remarkable. I did test builds of Michael Chua's "Lark" and later "Lark SM" speakers - same drivers but quite different crossovers. The first sounded very good; the second, excellent. The crossover schematics need to be purchased from Mr Chua, but his write-ups (Google 'Ampslab speakers') indicate the differences and improvements.
Geoff
I will take a stab at this topic.
- A crossover is a pair of filters that are designed to operate in such a way to make the sum of their outputs equal to the input.
- A filter changes the amplitude and phase of a signal that passes through it. One would like to have a "perfect filter" that perfectly passes some signals and completely attenuates other signals, where the difference between passed and stopped signals occurs at some frequency. This is sometimes called a "brick wall" filter.
- It's not possible to realize (make one in the real world) a brick wall filter, so engineers have resorted to approximating it using various techniques in the analog and digital realm. There is a term called the filter "order" that is related to how well the actual real world filter works like the perfect brick wall filter. The higher the order the better the real world filter approximates the brick wall filter. The filter order is also proportional to the complexity of the filter, for example the number of physical parts that are needed to construct it for a passive or analog active filter. If you have a mathematics background, the filter order is the highest power in the last term retained in a rational power series approximation to the ideal filter's brick wall response.
- Due to the appoximated response, the difference between passed and stopped signals for practical real world filters has a sort of "grey area" in which the level of attenuation (of the signal you do not want passed) slowly builds up. In a crossover, one filter should pass high frequencies and the other filter should pass the low frequencies. But around the crossover frequency the "grey area" from both filters is overlapping and their outputs interact by adding together. You would like the sum to everywhere equal the input signal, and this is where the phase of the output from each filter becomes important. Certain filter functions, typically named after the person or people who developed them, have been found to sum "nicely" when used in this way (e.g. Butterworth, Linkwitz-Riley, etc.). Each type of filter is available in various orders, and the higher the order the faster the filter will attenuate the undesired signal in the "grey area" at the beginning of the stop-band.
- We would like to have the behavior of the ideal brick wall filter but adding complexity represents increased cost, and as more and more components are used the error between what you want the filter to do and what the filter will actually do grows. This typically puts a limit on how high of an order filter is used for passive and analog active filters. Also, the more components and gain stages an active filter includes the more noise it produces. These sorts of things limit the practical complexity of analog domain filters.
What about filters that are implemented in the digital domain?
Digital filters operate on sampled signals, that is the audio signal is captured at regular intervals in time via some representation and a stream of these captures are used as the filter input. These are afterwards converted back to the analog domain, into a continuous signal. You may have heard of things like the sample rate, and the bit depth, which are related to how the real-world analog domain signal is represented in a digital form. Processing/filtering of digital signals is called "digital signal processing" or DSP. DSP can be performed in hardware (e.g. using integrated circuits that are specifically designed for that task) or in software (on a computer, using a program that gets/puts the audio from the operating system).
Digital filters come in two types: IIR and FIR:
- IIR filters have responses that are like analog filters, that is they have an order and their outputs have the same sorts of responses that you could get with an analog filter (one you make with physical components). The advantage of IIR filters over their analog counterparts is that they produce no noise and precisely match the desired filter function (there is zero error). The exception comes in the highest frequencies, where there is often some differences between the IIR response and the desired analog one. Typically this sort of error is not important for audio signal processing but it does exist.
- FIR filters are able to generate new responses that are not possible with IIR or analog domain filters. This is because the phase response and the amplitude response of the filter are independent of each other and you can prescribe filter responses that have (for example) a linear phase characteristic or even a phase characteristic that is opposite what you can get with IIR and analog filters. Any phase and frequency response characteristic can be implemented with FIR filtering.
- The FIR filter uses a "kernel" that should include all the information needed to implement the filter response. The FIR filter kernel is an approximation to the desired time domain response of the filter and therefore various errors in the amplitude and phase are produced by any FIR filter. This depends on how accurately the desired tiem domain response is represented by the FIR "kernel", which is often related to how sophisticated (how many taps) the FIR kernel has.
- When performing filtering via IIR filters, typically many of them are used in series to achive the overall filtering action that is needed or desired.
- When performing filtering via FIR filters, there is always one filter per output (e.g. per driver) and all of the filtering action is included in one large and complex filter.
- The FIR filter must be designed to match the desired target response while minimizing the errors that are produced as well as the complexity (length) of the kernel and the delay it produces. This is done using sophisticated software, and requires some knowledge and skill and the process is repeated any time you wish to change what the filter is doing overall.
- IIR filters are "predesigned" and using them is a matter of picking and choosing which to use and then stringing them into a filter chain that produces the overall effect. When you wish to change one part of the overall response, you can just change one filter in the chain without having to redesign anything.
Hi,
wealth of information already, I'll expand all the technical stuff with small snippet of philosophy 🙂
As already mentioned by others, filtering is specific to what you have built including the transducers, the physical construct that holds them all together, size of everything, distances of transducers and physical features in general, which all together define how sound radiates acoustically from any of the transducers and how the electrically split and manipulated sound combines back acoustically and is eventually processed by auditory system into a perception of it. Crossover optimal to one construct is not optimal for another different construct, and filtering in general could vary some per room and person for example, vary per context.
Philosophically you could think there is always an ideal crossover and filtering for any given construct, and that it's possible to find it just by using effort. There is free tools and good knowledge to help with the process should you choose to utilize them, and nothing prevents to reach ideal filtering within limitations you might have, like budget and whether it's only analog.
So, the crossover could be thought as "mere trivial task" at the end of speaker design process, that needs to be done after everything is built. The actual performance and most of the effort is actually in the physical construct, in what you have built. In other words, a crossover can be fit to set of measurements without much problem, so the actual limit of performance is the measurements. This means how appropriately the system radiates sound relative to what you'd want to perceive from the system. And to get good measurements you have to figure out what kind of a construct gives those? and root for it all is to figure out what is good measurement? how does perception of sound relate to measurements, what you wanna perceive, and how to make auditory system provide that.
wealth of information already, I'll expand all the technical stuff with small snippet of philosophy 🙂
As already mentioned by others, filtering is specific to what you have built including the transducers, the physical construct that holds them all together, size of everything, distances of transducers and physical features in general, which all together define how sound radiates acoustically from any of the transducers and how the electrically split and manipulated sound combines back acoustically and is eventually processed by auditory system into a perception of it. Crossover optimal to one construct is not optimal for another different construct, and filtering in general could vary some per room and person for example, vary per context.
Philosophically you could think there is always an ideal crossover and filtering for any given construct, and that it's possible to find it just by using effort. There is free tools and good knowledge to help with the process should you choose to utilize them, and nothing prevents to reach ideal filtering within limitations you might have, like budget and whether it's only analog.
So, the crossover could be thought as "mere trivial task" at the end of speaker design process, that needs to be done after everything is built. The actual performance and most of the effort is actually in the physical construct, in what you have built. In other words, a crossover can be fit to set of measurements without much problem, so the actual limit of performance is the measurements. This means how appropriately the system radiates sound relative to what you'd want to perceive from the system. And to get good measurements you have to figure out what kind of a construct gives those? and root for it all is to figure out what is good measurement? how does perception of sound relate to measurements, what you wanna perceive, and how to make auditory system provide that.
Last edited:
Start off with a first order crossover, a capacitor in series with the tweeter, and an inductor in series with the woofer. Both circuits are connected in parallel.
The capacitor provides a high impedance to low frequencies, so the low frequencies will go to the low frequency (woofer) circuit with the inductor preferentially. The inductor provides a high impedance to high frequencies, so the high frequencies will preferentially be diverted to the low impedance (tweeter) circuit with the capacitor. I think you can get an intuitive grasp by thinking about where high and low frequencies go whenever you see a component.
You also need to learn about a resistive voltage divider because that is really the basis by which these circuits (and L-pads) work. Find a wikipedia article on voltage dividers.
The capacitor provides a high impedance to low frequencies, so the low frequencies will go to the low frequency (woofer) circuit with the inductor preferentially. The inductor provides a high impedance to high frequencies, so the high frequencies will preferentially be diverted to the low impedance (tweeter) circuit with the capacitor. I think you can get an intuitive grasp by thinking about where high and low frequencies go whenever you see a component.
You also need to learn about a resistive voltage divider because that is really the basis by which these circuits (and L-pads) work. Find a wikipedia article on voltage dividers.
That's my way of looking at it. Combining an inductor and capacitor in series forms a voltage divider.
In a voltage divider, the component with the larger opposition to the flow of current will have the larger voltage across it.
Consider supplying a mixture of low and high frequency currents to this series inductor and capacitor combination:
Since the inductor has the larger opposition to the flow of high frequency currents, the high frequency voltages are found across the inductor.
Since the capacitor has the larger opposition to the flow of low frequency currents, the low frequency voltages are found across the capacitor.
The inductor can therefore be used as the high frequency voltage source for a tweeter and/or the capacitor can be used as the low frequency voltage source for a woofer.
Totally agree with this line of reasoning.
YES, crossovers are easy. Measurements aren't.
tmuikku, I split your paragraph into three separate quotes because i think it was so powerfully spot-on.
Very, well said & well reasoned ...thx 😀
I think the premise of this thread is pretty interesting, compared to the typical "here is my xo, what do you think" or "how to I do such and such".
@tmuikku , @mark100 - I don't necessarily disagree with you, but I'm going to take a contrarian position because I think it would be informative to expand on the posts above. Especially for a beginner who may not realize why you have stated certain positions.
1) Why would it be more effort to physically construct a speaker? Is this because using a saw, router, wood, glue, etc. takes more effort than simulations on a computer and prototyping with some small, well understood xo components? Or because, physically constructing a state of the art cabinet, in terms of resonances, directivity, etc. is hard?
I think it is pretty common for people to come to DIY speakers with reasonably good woodworking skills but not much knowledge of electronics and the xo is the most challenging thing. Building a rectangular prism is trivial. I think the crossover is a "trivial task" for someone who knows how to create a crossover and it is the hardest task for someone new to the hobby.
2) Why are measurements harder than developing a crossover? "Yes, crossovers are easy, measurements aren't". Why? To measure don't you just stick a speaker on a measurement rig, spin it around, and measure it a bunch of times? Then do some stuff with nearfield measurements and merge it all together nicely. On the other hand, crossovers require all kinds of knowledge on acoustic slopes, phase, group delay, impedance, impedance phase, etc.
3) What is good measurement and how does perception of sound relate to measurements? Is there more to this than pointing a reader to some of Toole/Harmon/Sean Olive's work (or other similar resource) and aren't these things well reflected in simulations like VituixCAD which focuses on such relations - On-axis response, listening window, estimated in-room response, directivity index, etc.?
@tmuikku , @mark100 - I don't necessarily disagree with you, but I'm going to take a contrarian position because I think it would be informative to expand on the posts above. Especially for a beginner who may not realize why you have stated certain positions.
1) Why would it be more effort to physically construct a speaker? Is this because using a saw, router, wood, glue, etc. takes more effort than simulations on a computer and prototyping with some small, well understood xo components? Or because, physically constructing a state of the art cabinet, in terms of resonances, directivity, etc. is hard?
I think it is pretty common for people to come to DIY speakers with reasonably good woodworking skills but not much knowledge of electronics and the xo is the most challenging thing. Building a rectangular prism is trivial. I think the crossover is a "trivial task" for someone who knows how to create a crossover and it is the hardest task for someone new to the hobby.
2) Why are measurements harder than developing a crossover? "Yes, crossovers are easy, measurements aren't". Why? To measure don't you just stick a speaker on a measurement rig, spin it around, and measure it a bunch of times? Then do some stuff with nearfield measurements and merge it all together nicely. On the other hand, crossovers require all kinds of knowledge on acoustic slopes, phase, group delay, impedance, impedance phase, etc.
3) What is good measurement and how does perception of sound relate to measurements? Is there more to this than pointing a reader to some of Toole/Harmon/Sean Olive's work (or other similar resource) and aren't these things well reflected in simulations like VituixCAD which focuses on such relations - On-axis response, listening window, estimated in-room response, directivity index, etc.?
OP:
Get XSim or VituixCAD to help you learn. You can simulate crossovers for free, and if you go to Dayton Audo's website you can even get driver files you can use.
I wrote an article on crossover design here I think you might find useful.
Get XSim or VituixCAD to help you learn. You can simulate crossovers for free, and if you go to Dayton Audo's website you can even get driver files you can use.
I wrote an article on crossover design here I think you might find useful.
Hi,
well, my post was bit of a broader framework and not directly related to crossovers but to speakers and playback in general, to give perspective about crossovers. And you are right, it's not something for beginners so kept it brief and clearly labelled to philosophy. I think it's always healthy to see broader picture, what ever one is doing, because it's very easy to get lost in details, which might not have any meaning in the end, and seeing the big picture might help reason through it.
Measurements aren't too hard, although if they contain error the crossover gets the same error so it's critical to get technically good measurements and it might take some effort to get good enough. My reference to measurements was specifically how to build physical construct so that it measures like you want. And even more work is know what you want it to measure like. Main point relating to the thread was that, crossover is made for particular speaker, particular set of measurements, and that's it, the end quality is tied to how closely it measured like you wanted it to and whether you wanted the right thing 😉 One could look for Toole/Harmon/Olive work, but there might be more specifics to a particular situation, your own room and your own practical stuff like positioning and what you like about a sound, what's the budget and aesthetics would be the main things that steer what to build. But, these get quite far from beginner advice and is something one might end up at some point, if perceived audio quality is high in the list of requirements for a speaker project. If you get the structure wrong, the crossover for that one is meaningless, except what it is worth in experience, as you need to build another construct that suits better and make another crossover for that one. Measurements and simulations are relatively easy and low cost to do, and equipped with knowledge available for example on this forum, so it's not that hard to come up with a crossover for any particular set of measurements.
Last edited: