This is the second installment of my investigation into the audibility of crossovers. The first part is here, where I asked people what crossovers they considered least audible. The premise is that I may rebuild my current dipole speakers as a 4-way, which would then introduce a crossover at 700Hz.
The primary concern is that introducing a crossover at 700Hz would put it right in the middle of the human voice range, where we are very sensitive to sound. But there are several reasons for doing so: it would allow me much greater flexibility in choosing midrange drivers; it would allow me to move my other two crossover points out of the midrange (from 275Hz and ~2kHz out to about 140Hz and 2.8kHz); it would allow greater regularity of the off axis response; and finally it would allow much greater output SPL and lower lower nonlinear distortion. Those are some pretty good reasons.
The results of the previous poll were a little surprising - almost nobody used higher order crossovers (above 4th order) and many people don't go for any particular alignment.
So what I have done now it to try and develop a test to help determine just how audible a crossover is, particularly at 700Hz.
What I've done is take two 6" drivers, specifically Eminence Alpha6a. I EQ them to be approximately flat between 250Hz and 10kHz, and as similar to each other as possible. Here are the two on axis response curves:
As you can see, they are very close - note the 3dB per vertical division. Here is one driver's polar response, just so people know what we are dealing with. The overall sound is okay - good for a single driver system, but poor compared to full range speaker systems.
My intention was to be able to listen to a single driver, and then using A/B type switching in my active crossover software (Reaper), introduce a crossover by using both drivers. The crossovers of interest were LR2, LR4 and LR8, all at 700Hz.
Here is an example of the LR4 crossover - one driver gets the lower half, the other gets the upper half.
Here is a comparison of both drivers individual frequency responses, overlaid with the LR2 response.
Here are the three FR's of the LR2, LR4 and LR8 overlaid. As you can see, they are exceedingly close:
The primary concern is that introducing a crossover at 700Hz would put it right in the middle of the human voice range, where we are very sensitive to sound. But there are several reasons for doing so: it would allow me much greater flexibility in choosing midrange drivers; it would allow me to move my other two crossover points out of the midrange (from 275Hz and ~2kHz out to about 140Hz and 2.8kHz); it would allow greater regularity of the off axis response; and finally it would allow much greater output SPL and lower lower nonlinear distortion. Those are some pretty good reasons.
The results of the previous poll were a little surprising - almost nobody used higher order crossovers (above 4th order) and many people don't go for any particular alignment.
So what I have done now it to try and develop a test to help determine just how audible a crossover is, particularly at 700Hz.
An externally hosted image should be here but it was not working when we last tested it.
What I've done is take two 6" drivers, specifically Eminence Alpha6a. I EQ them to be approximately flat between 250Hz and 10kHz, and as similar to each other as possible. Here are the two on axis response curves:
An externally hosted image should be here but it was not working when we last tested it.
As you can see, they are very close - note the 3dB per vertical division. Here is one driver's polar response, just so people know what we are dealing with. The overall sound is okay - good for a single driver system, but poor compared to full range speaker systems.
An externally hosted image should be here but it was not working when we last tested it.
My intention was to be able to listen to a single driver, and then using A/B type switching in my active crossover software (Reaper), introduce a crossover by using both drivers. The crossovers of interest were LR2, LR4 and LR8, all at 700Hz.
Here is an example of the LR4 crossover - one driver gets the lower half, the other gets the upper half.
An externally hosted image should be here but it was not working when we last tested it.
Here is a comparison of both drivers individual frequency responses, overlaid with the LR2 response.
An externally hosted image should be here but it was not working when we last tested it.
Here are the three FR's of the LR2, LR4 and LR8 overlaid. As you can see, they are exceedingly close:
An externally hosted image should be here but it was not working when we last tested it.
The results? There were some audible changes compared to a single full range driver.
There are two issues with crossovers as I see it - phase and power response. The phase (group delay) nonlinearity of LR type crossovers is fairly well known. Linkwitz discusses it here. His conclusion is that the phase errors introduced by LR2 and LR4 type crossovers are inaudible.
Perhaps less know is John Kreskovsky's analysis of the power response dip due to crossovers. In summary, his calculations suggest LR type crossovers introduce a dip of at least -3dB around the crossover region in the speakers total acoustic output (power response), with the width of that dip narrowing for higher order crossovers. He postulates that higher order crossovers are more audible than lower order crossovers due to the more abrupt change in power response.
So what I did was take a listen to the various alignments, with various types of music. I would select two alignments, listen for some time, and then do a quick switch between them - one alignment was always the full range driver, and the other was one of the three crossover variations. I did this over several days, both sighted and blind. Interestingly, classical music, which is usually very revealing, was quite unrevealed to me. But human voice was very revealing - both male and female. Also interesting, sighted and blind tests gave the same conclusions, but sighted tests took less time to make a determination.
So what were the results???
The basic trend was that the higher orders were more audible. LR2 was indistinguishable from the single full range driver, while LR8 was easily identifiable. LR4 fell in the middle - it was slightly audible, only on very close examination.
How did they sound different? Higher orders made human voices sound more forward and direct, over a very small part of their spectrum. The full range driver (or LR2) sounded smoother, with better ambience. I should add here that this may be a feature of the 700Hz crossover frequency, and I would expect different results much lower or higher (like 150Hz or 3000Hz).
I want to emphasize that the differences ranged from very to extremely subtle. A poorly done crossover is many time more audible than the differences between these crossover alignments. The sense of forwardness was hard to pinpoint - the basic frequency response of the music didn't change - but it was like a small little section of the voice stood out more compared to the full range driver. It seemed apparent that this was due to a reduction in the ambient sounds associated with the regions of change - basically a reduction in the voices' reverberant field. These differences would probably be impossible to judge without quick A/B switching.
But I will add something that I felt was important - LR8 always had a sense of unpleasantness to it. The degree that part of the voice jutted forward was always got on my nerves. Because of this, I've decided to avoid higher order crossovers (above LR4), at least in this application. Like I said earlier, LR8 was also fairly easy to discern while switching.
I feel this test is evidence supporting John K's conclusions - that the crossover's main issue is the reduction in the total sound output at the crossover region, with higher order crossovers being more intrusive. Initially, I had thought the opposite would be true, but I'm glad to know just how little audible difference the LR4 crossover at 700Hz made. I'm guessing Linkwitz's test didn't show these results because it was a test of electrical phase changes, not acoustic summing of multiple drivers.
There are two issues with crossovers as I see it - phase and power response. The phase (group delay) nonlinearity of LR type crossovers is fairly well known. Linkwitz discusses it here. His conclusion is that the phase errors introduced by LR2 and LR4 type crossovers are inaudible.
Perhaps less know is John Kreskovsky's analysis of the power response dip due to crossovers. In summary, his calculations suggest LR type crossovers introduce a dip of at least -3dB around the crossover region in the speakers total acoustic output (power response), with the width of that dip narrowing for higher order crossovers. He postulates that higher order crossovers are more audible than lower order crossovers due to the more abrupt change in power response.
So what I did was take a listen to the various alignments, with various types of music. I would select two alignments, listen for some time, and then do a quick switch between them - one alignment was always the full range driver, and the other was one of the three crossover variations. I did this over several days, both sighted and blind. Interestingly, classical music, which is usually very revealing, was quite unrevealed to me. But human voice was very revealing - both male and female. Also interesting, sighted and blind tests gave the same conclusions, but sighted tests took less time to make a determination.
So what were the results???
The basic trend was that the higher orders were more audible. LR2 was indistinguishable from the single full range driver, while LR8 was easily identifiable. LR4 fell in the middle - it was slightly audible, only on very close examination.
How did they sound different? Higher orders made human voices sound more forward and direct, over a very small part of their spectrum. The full range driver (or LR2) sounded smoother, with better ambience. I should add here that this may be a feature of the 700Hz crossover frequency, and I would expect different results much lower or higher (like 150Hz or 3000Hz).
I want to emphasize that the differences ranged from very to extremely subtle. A poorly done crossover is many time more audible than the differences between these crossover alignments. The sense of forwardness was hard to pinpoint - the basic frequency response of the music didn't change - but it was like a small little section of the voice stood out more compared to the full range driver. It seemed apparent that this was due to a reduction in the ambient sounds associated with the regions of change - basically a reduction in the voices' reverberant field. These differences would probably be impossible to judge without quick A/B switching.
But I will add something that I felt was important - LR8 always had a sense of unpleasantness to it. The degree that part of the voice jutted forward was always got on my nerves. Because of this, I've decided to avoid higher order crossovers (above LR4), at least in this application. Like I said earlier, LR8 was also fairly easy to discern while switching.
I feel this test is evidence supporting John K's conclusions - that the crossover's main issue is the reduction in the total sound output at the crossover region, with higher order crossovers being more intrusive. Initially, I had thought the opposite would be true, but I'm glad to know just how little audible difference the LR4 crossover at 700Hz made. I'm guessing Linkwitz's test didn't show these results because it was a test of electrical phase changes, not acoustic summing of multiple drivers.
Last edited:
Very nice test - thanks! I certainly have never seen it done that way before.
Yesterday I was experimenting with crossover slopes, too. My crossover is at about 600Hz. Changing types had a similar effect - moving things closer and farther away. It seemed to me that the better the phase, the farther back things moved.
Similar to MP3 vs uncompressed. MP3 often sounds more forward to me.
Thanks for posting that.
Yesterday I was experimenting with crossover slopes, too. My crossover is at about 600Hz. Changing types had a similar effect - moving things closer and farther away. It seemed to me that the better the phase, the farther back things moved.
Similar to MP3 vs uncompressed. MP3 often sounds more forward to me.
Thanks for posting that.
This is true when the crossover is at a frequency where off axis nulls appear. If the separation of the drivers is significantly less than a wave length at the crossover frequency then the dip for the LR crossover will be reduced and ultimately disappear as the wave length gets longer. At 700 Hz the wave length is 1.6 feet so for drivers as you have them positioned I would not expect to see much of a dip in the power resposne. For a dipole set up you will have to compare the vertical and horrizontal response and see if the vertical deviates for the horrizontal. If it does, it is an indication of a null resulting from the crossover and a dip in power response.
I've considered it, but haven't felt it high enough priority - as far as I can see from just data, the only differences between LR and BW type crossovers are BW have a small on axis bump up, less power dip, and a different orientation of the vertical off axis nulls. While LR2 was inaudible (to me), and LR4 only slightly, I would guess BW3 would be in between - but the differences are so small, it probably doesn't matter (again, at least to me).
I'd love to see someone else try this experiment, to see if they get similar or different results than I did. If you have a computer and some measurement capability, it was actually pretty easy to put together...
Horizontal or vertical? My experience with measuring off axis responses would indicate that the final acoustic response would be a combination of the electrical filter, and the above posted polar response - so nothing surprising. What would you be looking for?
Thanks for chiming in John, you have a very good point - the vertical separation is only about half a wavelength, so there shouldn't be much vertical lobing - perhaps that was why things seemed so subtle. In the future I was thinking of testing higher and lower crossover frequencies - thanks for giving me hints as to which might be more audible. I think the audibility of the LR8 filter was significant though - hearing it definitely turned me off to it.
I'd be looking for something I didn't expect.
Great series of posts BTW.
I've often wondered if, similar to the way some people are 'color blind', some people are 'phase deaf'. Some people seem to prize flat frequency response above all else, other are looking for coherent transients. Like yourself, I don't like high-order filters. Even if they "can't be heard" I get this nagging sense of wrongness.
I'm getting the sense these are all warped phase crossovers. Have you tried linear phase? You may also want to try A/Bing both drivers allpass against crossed; I'd be cautious of A/Bing where one driver is on in one case and off in another as speakers don't normally turn drivers on and off. Maybe I missed it, but what exactly are you looking to get out of a four way that's missing in the three way version of the Violet DSP?
Since you want data points, my current system is a 6.5 in woofer/25mm tweeter two way with a linear phase LR6 at 2kHz (and correction of the port resonance phase wrap at 40Hz). Originally shipped with a second order passive crossover at 2.7kHz and I ran it warped phase LR4 at 1.9kHz for several years with an Ashly XR-1001 before switching to Allocator and auditioning warped and linear phase LR2 up through B7 (doing LR8 and above in Allocator is a bit tedious). As you'd expect in a two way I found the dominant effects were managing tweeter excursion and woofer cone breakup---LR4's not steep enough, B5's better, B7 is OK, and I liked LR6 the best. But I've no problem hearing group delays when A/Bing linear and warped phase and find the audible difference between linear and warped decreases at higher orders due to the narrower transition region.
I've since implemented and measured a few three and four way candidate crossovers for various projects in Arbitrator. Haven't listened to any of them yet as I'm still designing the necessary power amp channels and selecting drivers. But so far my preference remains LR6 as I'd like to use the additional ways to push drivers less hard rather than easing the crossover. Personally I'm not real keen on LR2 due to the broad transition range and prefer LR4 and up.
Since you want data points, my current system is a 6.5 in woofer/25mm tweeter two way with a linear phase LR6 at 2kHz (and correction of the port resonance phase wrap at 40Hz). Originally shipped with a second order passive crossover at 2.7kHz and I ran it warped phase LR4 at 1.9kHz for several years with an Ashly XR-1001 before switching to Allocator and auditioning warped and linear phase LR2 up through B7 (doing LR8 and above in Allocator is a bit tedious). As you'd expect in a two way I found the dominant effects were managing tweeter excursion and woofer cone breakup---LR4's not steep enough, B5's better, B7 is OK, and I liked LR6 the best. But I've no problem hearing group delays when A/Bing linear and warped phase and find the audible difference between linear and warped decreases at higher orders due to the narrower transition region.
I've since implemented and measured a few three and four way candidate crossovers for various projects in Arbitrator. Haven't listened to any of them yet as I'm still designing the necessary power amp channels and selecting drivers. But so far my preference remains LR6 as I'd like to use the additional ways to push drivers less hard rather than easing the crossover. Personally I'm not real keen on LR2 due to the broad transition range and prefer LR4 and up.
Doesn't all this depend on each individual project?
There is no general opinion that "First order is better period" because there is no general optimal speaker design period. Full range speaker absolute suck for any of my applications so for me no crossover is impossible. I have used LR4, etc with ribbon tweeters with great success in an active setup, of course that important 600Hz to 2000Hz range is covered by a mid range woofer anyways. I really wish there was a great woofer that ran 50Hz to 2KHz without issues but I understand enough to say its almost impossible to have one do that and still have decent distortion, SPL, directivity
The first thread had statistical insignificance in terms of not enough votes to matter, now we are on to the subjective side. Im learning about the issues that could exist but I still believe each build has its own requirements.
There is no general opinion that "First order is better period" because there is no general optimal speaker design period. Full range speaker absolute suck for any of my applications so for me no crossover is impossible. I have used LR4, etc with ribbon tweeters with great success in an active setup, of course that important 600Hz to 2000Hz range is covered by a mid range woofer anyways. I really wish there was a great woofer that ran 50Hz to 2KHz without issues but I understand enough to say its almost impossible to have one do that and still have decent distortion, SPL, directivity
The first thread had statistical insignificance in terms of not enough votes to matter, now we are on to the subjective side. Im learning about the issues that could exist but I still believe each build has its own requirements.
Last edited:
Great Job "cuibono" and many thanks to share your findings
I'm agree with this. I use LR2 where the speaker are very close in dimension like a 8" and a 5" at 500Hz or a 2" dome and a 1" dome at 3000Hz.
My last three 3 ways is a 8"+3"+1" 500Hz and 3000Hz LR2. The sound is not as clean as a LR4 but the sound stage is better than a LR4.
I find the microdynamics ( impact of percussions, strings attacks, shocks are more dry for example ) is better with LR4 and the sound is cleaner.
It depends on taste and/or demand ?
I'm agree with this. I use LR2 where the speaker are very close in dimension like a 8" and a 5" at 500Hz or a 2" dome and a 1" dome at 3000Hz.
My last three 3 ways is a 8"+3"+1" 500Hz and 3000Hz LR2. The sound is not as clean as a LR4 but the sound stage is better than a LR4.
I find the microdynamics ( impact of percussions, strings attacks, shocks are more dry for example ) is better with LR4 and the sound is cleaner.
It depends on taste and/or demand ?
Good point - I am using regular, nonlinear phase crossovers. I have the capability of using linear phase, I just haven't tried it yet. I definitely should though. But I'm not sure linear phase crossover software like Allocator (which I used extensively), PLParEQ, or Waves (which is what I use now) are good enough - the software addresses the electrical aspect of the filter, but what needs to be compensated is the final acoustic response, both of the individual drivers and of the total system. Arbitrator does address this, but I was never impressed with it, and moved on to other software. I will address phase nonlinearity soon though - but like my previous work, I value measurements much more than predicted responses, and measuring absolute phase is not easy. Look at paragraph 2 of the first post for why I want to add the crossover.
Yes, what I have presented is particular to my situation, but I would expect it to be generalizable somewhat - OB crossovers in the 400-1200Hz range? Just guessing.
You bring up a good point - 'learning about the issues that could exist'. As far as I can tell, there are almost an infinite number of issues people have thought up - what I wanted to do here was some actual listening to see how high to prioritize the issue of losses due to crossovers. For me, the lesson is that it isn't too big a deal, which is good. I feel that 'potential issues' is the bane of current audio design - there are many many potential issues, but not too many people actually testing to see whether they matter or not.
Great Job "cuibono" and many thanks to share your findings
I'm agree with this. I use LR2 where the speaker are very close in dimension like a 8" and a 5" at 500Hz or a 2" dome and a 1" dome at 3000Hz.
My last three 3 ways is a 8"+3"+1" 500Hz and 3000Hz LR2. The sound is not as clean as a LR4 but the sound stage is better than a LR4.
I find the microdynamics ( impact of percussions, strings attacks, shocks are more dry for example ) is better with LR4 and the sound is cleaner.
It depends on taste and/or demand ?
You're welcome! It sounds like you've had the same experience as I have. Comparing LR2 and LR4, it is hard to know if the change in sound is due to changes in nonlinear distortion, off axis response or phase distortion. The main thing, at least for me, is to listen and see which is preferable, and possibly closer to what is recorded. Taste is probably the final arbiter.
By the way, has anyone noticed the new severed-head-in-a-bag-smiley?
What does it mean? "I can't in breathe here"? "Don't hold my/your breath"?I think all this is great and its simple enough for even a disfunctional crossover guy like myself to understand!
My experience is the dominant sources of phase rotation are the crossover and mechanical resonance (driver Fs or ports). Room interactions will rotate phase. From the Violet DSP thread it seems like your DRC style is more like mine and Linkwitz's (minimalistic) than StigErik's (equalize everything in sight), so I wouldn't expect that to be a big issue for you.
Can't really comment on Waves as they don't seem to sell LinEQ separately and all the bundles which include it are out of my price range, but five bands is kind of a hassle for implementing a crossover. Best I can tell Allocator gets the wavefront launch off my ported box speakers flat within 10 degrees. Room interactions then shift it to plus or minus 20 degrees. I'm just starting with dipoles but so far I expect similar results. Unfortunately for dipole work Allocator's equalization is not linear phase and, unfortunately for you, probably the only low latency linear phase software or hardware solution is FIR. Latency aside, PLParEQX3 measures out flat phase in my trials. The missing piece is a plugin which performs frequency dependent phase adjustment.
I'd be happy to hand all this off to a Motu 828 or the like. But I've a hard requirement for linear phase and the hardware's not there yet. My next project will likely be in that direction, but I've dipoles to hang and at least eight channels of power amps to build first.
You know, I have very little experience trying with trying to adjust phase only - could you describe how you measure phase? I could use some guidance here. We don't use the same software, but one of the basic issues is removing excess phase, so that one is looking at only the actual acoustic/absolute phase response. Another issue is whether the software calculates phase, or if it reports the actual measured phase. There are also issues with phase rotation due to the mic, mic correction file, and preamp. This is about the limit of my experience, so I hope people who've done rigorous phase measurements will chime in. John K?
The other thing that I'm guessing is important is that any phase correction would be most successful if it corrected for the drivers and systems acoustic response. But all the linear phase software I've seen only address the electrical component. But you've gotten nearly flat phase responses? Could you describe more about how you do it?
I found this piece of software I will start playing with soon - voxengo pha-979. I haven't tried it yet, but it looks like it might do frequency dependent phase shifting. I'm sure there is some way to do it, I just haven't figured it out yet.
PS - I built a multichannel amp recently for my project - I HIGHLY recommend using GC type amps - I went with something slightly more complicated (I had to develop the boards and prototype the amp), and it ended up being a major resource sucker. I would definitely use 8 channels of LM3875 if I had to do it again - it would be sooo many times easier. I seriously wish I had used someone elses already proven PCBs.
The other thing that I'm guessing is important is that any phase correction would be most successful if it corrected for the drivers and systems acoustic response. But all the linear phase software I've seen only address the electrical component. But you've gotten nearly flat phase responses? Could you describe more about how you do it?
I found this piece of software I will start playing with soon - voxengo pha-979. I haven't tried it yet, but it looks like it might do frequency dependent phase shifting. I'm sure there is some way to do it, I just haven't figured it out yet.
PS - I built a multichannel amp recently for my project - I HIGHLY recommend using GC type amps - I went with something slightly more complicated (I had to develop the boards and prototype the amp), and it ended up being a major resource sucker. I would definitely use 8 channels of LM3875 if I had to do it again - it would be sooo many times easier. I seriously wish I had used someone elses already proven PCBs.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.