So, one of the things I've done within the last few days is move the mid-tweeter XO point up to 2600Hz from 1700Hz. I knew for sure that this would deepen the off axis polar dip, but I did it mainly because several people were saying 1700Hz is too low for the Neo3pdr tweeter.
Here are the before and after polar measurements. As you can see, at 60deg off axis, the dip deepens by about 2dB.
Before:
After:
I assumed this would be audible, but, at regular listening levels, switching between the two settings showed the difference to be inaudible. That is good. I guess some amount of off axis irregularity is going to be inaudible - but I don't know where the transition is from audible to inaudible.
Similarly, at regular listening levels, I detected no difference in the sound at all - so I would surmise that nonlinear distortion is not an issue at modest SPLs. But that shouldn't be a surprise. I then listened at very loud levels. I didn't do A/B comparison switching, but I thought things might have been better maintained - there were fewer incidence of thinking 'whoa, thats too loud', which is where I would guess is where NLD is becoming an issue.
Good news all in all. I'll probably keep the XO point at 2600Hz because it is probably better for the tweeter, and no cost to the acoustics.
Here are the before and after polar measurements. As you can see, at 60deg off axis, the dip deepens by about 2dB.
Before:
An externally hosted image should be here but it was not working when we last tested it.
After:
An externally hosted image should be here but it was not working when we last tested it.
I assumed this would be audible, but, at regular listening levels, switching between the two settings showed the difference to be inaudible. That is good. I guess some amount of off axis irregularity is going to be inaudible - but I don't know where the transition is from audible to inaudible.
Similarly, at regular listening levels, I detected no difference in the sound at all - so I would surmise that nonlinear distortion is not an issue at modest SPLs. But that shouldn't be a surprise. I then listened at very loud levels. I didn't do A/B comparison switching, but I thought things might have been better maintained - there were fewer incidence of thinking 'whoa, thats too loud', which is where I would guess is where NLD is becoming an issue.
Good news all in all. I'll probably keep the XO point at 2600Hz because it is probably better for the tweeter, and no cost to the acoustics.
Here is a little more interesting rearrangement I've made recently too.
I suppose what got me onto this was thinking about one of the basic tenants of LR type crossovers - the two drivers working around the crossover must be equidistant from the listener to sum so that the vertical nulls are 'where they are supposed to be'. I then realized that my design, and all the other OB designs I could think of, have the same issue - the acoustic centers of each driver are different distances to the ear. The bass is particularly effected, being about 8% farther than the midrange from my ear.
Similarly, I kept having to adjust the EQ by ear, no matter how well I measured it - particularly, setting the woofer and tweeter level relative to the midrange. I usually had to boost the bass by several dB, and lower the treble by a couple dB. I didn't know why, but I could tell that even if my measurements were flat anechoicly, something wasn't right - what I had previously refered to as a need do final in room adjustments for the 'room curve'.
Well, then I realized that not only was the path lengths different, but the angle of radiation that I was listening on unequal - which leads to fairly significant in room radiated power imbalances.
The woofer is a good example - I have already had to boost its level by .7dB to account for the path length difference - so while the listening position may measure flat, the total woofer output has been boosted by that amount relative to the midrange. Additionally, I listen at 30deg off axis horizontally (because I sit in an equilateral triangle), and about 25deg vertically (because the woofer's acoustic center (AC) is 12" off the ground, while my ear is 44"), relative to the woofer's axis of radiation - this lead to at least 3dB, if not more, of level difference between the listening position direct radiation, and the total power response of the woofer, relative to the midrange. I think this is what made adjusting levels difficult - the direct radiation level and in room power were not scaling together. Additionally, the time of arrival is different between drivers, so their phase is not summing perfectly. Not an idea situation.
Drawings might make this more understandable (please excuse my drawings!). The first arrangement is where I had the AC's of the drivers arranged vertically:
The second is my more preferable configuration, where things are arranged radially from the ear:
There are two major differences between the two arrangements: first, each driver's AC is equidistant to the ear; second, each driver's axis is pointed directly at the ear. I am fairly certain that both of these conditions are necessary to ensure proper acoustic summing and power balance between all drivers.
After doing this, there were some pretty good improvements - the was a substantial increase in coherency through the whole frequency range; things sounded more punchy; there was both an increase in clarity and low level ambience. Things sound smoother and easier to listen to - more natural.
Also, the system required almost no adjustment after measurement - the only adjustment I made was to shelf the treble -1dB, which is probably due to my room being all hard surfaces, and listening relatively closely. I would conclude that this made a substantial improvement in the overall quality of the presentation, and highly recommend it. I would go so far as to say that any OB speaker that doesn't do this cannot both sum flat on axis and have a flat power response, say nothing of the phase response.
I should add one note though - previously, I had been measuring (outdoors) with each driver set so that the mic replicated the individual distances and angles of the listening position in room. I no longer do that - for this trick to work, each driver must be measured directly on the driver's axis (preferably at about 2m - at 1m or less, the driver's dipole cancellation is not fully seen). But this actually makes measuring easier.
Anyways, this is highly recommended.
I suppose what got me onto this was thinking about one of the basic tenants of LR type crossovers - the two drivers working around the crossover must be equidistant from the listener to sum so that the vertical nulls are 'where they are supposed to be'. I then realized that my design, and all the other OB designs I could think of, have the same issue - the acoustic centers of each driver are different distances to the ear. The bass is particularly effected, being about 8% farther than the midrange from my ear.
Similarly, I kept having to adjust the EQ by ear, no matter how well I measured it - particularly, setting the woofer and tweeter level relative to the midrange. I usually had to boost the bass by several dB, and lower the treble by a couple dB. I didn't know why, but I could tell that even if my measurements were flat anechoicly, something wasn't right - what I had previously refered to as a need do final in room adjustments for the 'room curve'.
Well, then I realized that not only was the path lengths different, but the angle of radiation that I was listening on unequal - which leads to fairly significant in room radiated power imbalances.
The woofer is a good example - I have already had to boost its level by .7dB to account for the path length difference - so while the listening position may measure flat, the total woofer output has been boosted by that amount relative to the midrange. Additionally, I listen at 30deg off axis horizontally (because I sit in an equilateral triangle), and about 25deg vertically (because the woofer's acoustic center (AC) is 12" off the ground, while my ear is 44"), relative to the woofer's axis of radiation - this lead to at least 3dB, if not more, of level difference between the listening position direct radiation, and the total power response of the woofer, relative to the midrange. I think this is what made adjusting levels difficult - the direct radiation level and in room power were not scaling together. Additionally, the time of arrival is different between drivers, so their phase is not summing perfectly. Not an idea situation.
Drawings might make this more understandable (please excuse my drawings!). The first arrangement is where I had the AC's of the drivers arranged vertically:
An externally hosted image should be here but it was not working when we last tested it.
The second is my more preferable configuration, where things are arranged radially from the ear:
An externally hosted image should be here but it was not working when we last tested it.
There are two major differences between the two arrangements: first, each driver's AC is equidistant to the ear; second, each driver's axis is pointed directly at the ear. I am fairly certain that both of these conditions are necessary to ensure proper acoustic summing and power balance between all drivers.
After doing this, there were some pretty good improvements - the was a substantial increase in coherency through the whole frequency range; things sounded more punchy; there was both an increase in clarity and low level ambience. Things sound smoother and easier to listen to - more natural.
Also, the system required almost no adjustment after measurement - the only adjustment I made was to shelf the treble -1dB, which is probably due to my room being all hard surfaces, and listening relatively closely. I would conclude that this made a substantial improvement in the overall quality of the presentation, and highly recommend it. I would go so far as to say that any OB speaker that doesn't do this cannot both sum flat on axis and have a flat power response, say nothing of the phase response.
I should add one note though - previously, I had been measuring (outdoors) with each driver set so that the mic replicated the individual distances and angles of the listening position in room. I no longer do that - for this trick to work, each driver must be measured directly on the driver's axis (preferably at about 2m - at 1m or less, the driver's dipole cancellation is not fully seen). But this actually makes measuring easier.
Anyways, this is highly recommended.
Last edited:
One of the big benefits to OB speakers is there are much much wider 'sweet spots' compared to boxed speakers - in fact, I hear very little to no change in the acoustics anywhere in the room, which is another thing that makes them very enjoyable to listen to. The only thing that changes by moving is the perspective into the soundstage - so, no, I don't find this new arrangement to be more limiting of the sweet spot. It seems to be the opposite, where the sound is more balanced everywhere in the room.
Imaging hasn't changed much - but there has been an increase in clarity and reverberant space, which helps reinforce the illusion of the soundstage.
I'm not sure what your trying to say about the backwave, but in my experience, the distances between the walls, speakers and listener(s) only changes the size of the perceived soundstage - similar to a 30" television versus a 70" projection. I haven't found any of the arrangements I've tried change the perceived frequency response.
I have measured reflection related peaks and nulls in the room - and I've tried EQing them. I've had varying results, but my most recent efforts suggest that it is best to leave them alone, presuming that the anechoic measurements were done well. I'll post some of the in room response measurements I've done, sooner or later, but still the bottom line for me was that reflections (either backwave or frontwave) sounded best left alone.
nor am I.
it's just that until now i thought of dipole cancellation as a planar 2d shape, orienting driver cone will make it a more complex 3d shape, but perhaps it probably is always like that just me not considering, and actually the mid-highs will just have the very same cancellation pattern just rotated a little, and at the crossover frequency with the woofers it will not be relevant... ok i don't know what i'm saying again thanks for a very instructive thread!
Your right - PLParEQ is very CPU heavy, and the latency is loooooong. Price.. well, it depends. The full version is 1000$, not cheap, but I think its worth the money. More value than expensive cables at least.....
I have a 1.8 GHz CoreDuo machine, and running 4-way XO with EQ on this is difficult, I'm balancing on the edge so to speak, with 97% or so CPU util.....
Are you kidding? $1000 for an incredibly simple piece of software (I use the term lightly, a plugin for a host program is hardly a functional piece of software)? And its not even efficient?!
You people seriously need to consider linux... I run 3 way mains plus a horn sub with long (65536 tap, read: detailed) filters on an ancient athlon system and I dont even approach half CPU capacity. This includes crossover, eq and phase correction for every driver... thats 21 filters. The idea that you cant even run crossovers and eq on a modern core2 system with this awful "plugin" baffles me....
I assume you have managed to fudge around the windows kmixer? Your setup is crippled if you haven’t.
You spoke about driver issues, adding random delay to channels because it assumes you have 5.1... another example of why using windows for audio isn’t a good idea... Alsa will give you direct access to the hardware on your card, no layers of idiotic "value adding" software between you and the outputs.
You run the stream in through an A/D converter before you apply your filters? Why? I can only assume your using a CD "transport" with all of the errors that introduces. Why would any sane person decode digital data from an optical media with their expensive transport "correcting" the read errors on the fly to a D/A converter in it, only to then pass it through a A/D, apply filters and then out through another D/A...? I cant gleam from your posts, but you may actually be putting out SPDIF from your player and taking it into the PC that way... but even this is inherently inferior to direct playback from the PC. Which is another thing my ancient athlon system (I'm talking socket A K7, not 64) can do at the same as all the detailed FIR filters... decode lossles compression formats into PCM streams.
As for the OPs system... still seems like a work in progress. You talk about flat off axis response and then claim large improvements from tilting the drivers so your listening on axis... something that could only narrow the sweet spot and require different eq. If its a matter of time alignment then this can be done on the PC. Seems like your still chasing your tail.
A linux system with MPD (music player daemon) piped to brutefir. MPD is controllable with a plethora of other software, directly or over a network..
Brutefir is an incredibly efficient way of applying FIR filters, and has great flexibilty in the structure of them you create. You can cascade as many filters as you like, or combine them all into one. Generation of these filters can be done in many ways, matlab/octave, numerous freeware programs, holm impulse....
Eq and phase correction filters, I feel, can be best created using holm impulse, the many ways to export measured data mean it can be manipulated in any way you like to create detailed accurate filters.
How about rather than whining, starting your own thread detailing what you do. Recognize that not everyone wants to spend their time programing computers and operating systems.
WRT my system, did you see the title? Doesn't "evolution" imply 'work in progress'? Doesn't diyAudio imply "work in progress"? Changing the axis of radiation is not a matter of time alignment - if you don't understand what we are doing here, you shouldn't be so quick to criticize. I think the civil thing would be to ask why we are doing what we are doing, rather than cast off a bunch of unfounded criticism.
Sounds like time alignment to me....
Apologies for the long post that was slightly off topic (you do have DSP in your title though)... I was more reacting to the idea that anyone would spend $1000 on something so pointless... at least cables are physical.
Of course all of our systems here at diyaudio are works in progress... with my last paragraph I may have been suggesting you may need to re-examine your eq after changing the position of the drivers.
First, time physical time alignment has different results than time delayed alignment. The sort of alignment I do is not possible using a time delay. You can read up on this elsewhere.
Second, I did remeasure my EQ - "for this trick to work, each driver must be measured directly on the driver's axis".
Third, tilting the drivers is a mater of balancing the direct response versus total power radiation, and is not related to time alignment.
Fourth, don't assume, without trying it yourself, that the adjustments I've made "could only narrow the sweet spot". Assumptions are a lot more crippling than a slow computer or a long latency.
I'm still interested in other solutions to PCXO, as I'm sure others are. If you started a detailed thread describing what your system comprises of, I'm sure people would appreciate it. Personally, I've always avoided linux just because I have enough to deal with already - if you could make it simpler for people like me, that would be great.
Second, I did remeasure my EQ - "for this trick to work, each driver must be measured directly on the driver's axis".
Third, tilting the drivers is a mater of balancing the direct response versus total power radiation, and is not related to time alignment.
Fourth, don't assume, without trying it yourself, that the adjustments I've made "could only narrow the sweet spot". Assumptions are a lot more crippling than a slow computer or a long latency.
I'm still interested in other solutions to PCXO, as I'm sure others are. If you started a detailed thread describing what your system comprises of, I'm sure people would appreciate it. Personally, I've always avoided linux just because I have enough to deal with already - if you could make it simpler for people like me, that would be great.
Do you believe your modification gives rise to a different dispersion pattern? I see no mention of this in previous posts and it seems the only positive result possible from your change that couldnt be achieved with delay.
I disagree with this statement. If you flatten fr on axis you are not correcting the power response. On axis only represents a fraction of the power response.
It distresses me slightly that this is the opinion of many people in the DIY scene.... you can spend countless hours reading up on theory and design, but because computers are 'newfangled' anything but the obvious (windows) is too much effort, yet going out of your comfort zone with acoustic ideas and principles is common place. This is especially annoying when it seems obvious to me that ALL audio reproduction will be computer based in the not too distant future.
It would be great to get some advice from the experts about the following things I'm hoping to do. This seems like a good thread to ask these questions.
I have a semi-omni directional loudspeaker that I'm intending to take into an anechoic chamber, make some measurements and use FIR filters to adjust the frequency response of the speakers.
As I understand it, if I put the microphone into the sweet spot (1m away, directly in front, on axis), measure the frequency response of the speakers, and then make that response flat in the frequency domain using FIR filters then there should be the following result: the on axis response will be flatter and because it's in an anechoic environment (minimum phase) the impulse response will be closer to an ideal response, with less ringing in the time domain.
That's the good news. However I have read that the bad news is what happens to the off axis response. That is, improving the on axis response necessarily degrades the off axis response.
I was thinking about this and wondering whether it's actually likely to be true. For instance, let's say that the speaker’s off axis response is quite close to the on axis response. That's the sort of response I associate with the term “constant directivity” speakers: as the angle of measurement moves from 0° in front of the loudspeaker to wider and wider angles, the frequency response measurement tracks fairly well with the 0° response.
If that's so then wouldn't it be the case that as we improve the frequency response of the 0° measurement we also by direct association improve the frequency response of the off axis response?
The second question is: so you have a loudspeaker where the on axis response is a certain shape and the off axis response is actually quite different from that. If you improve the on axis response and therefore cause a change to the off axis response, given that the off axis response is probably pretty bad, and probably un-correlated with the on axis response, is it plausible that the off axis response could in fact be improved instead of degraded, and that there is no way of knowing in advance until you actually do a listening test or off axis measurement?
Hope this all makes sense. If any knowledgeable people can shed light on these questions it would be great to hear from them. Apologies in advance if these seem naive questions to be asking.
I have a semi-omni directional loudspeaker that I'm intending to take into an anechoic chamber, make some measurements and use FIR filters to adjust the frequency response of the speakers.
As I understand it, if I put the microphone into the sweet spot (1m away, directly in front, on axis), measure the frequency response of the speakers, and then make that response flat in the frequency domain using FIR filters then there should be the following result: the on axis response will be flatter and because it's in an anechoic environment (minimum phase) the impulse response will be closer to an ideal response, with less ringing in the time domain.
That's the good news. However I have read that the bad news is what happens to the off axis response. That is, improving the on axis response necessarily degrades the off axis response.
I was thinking about this and wondering whether it's actually likely to be true. For instance, let's say that the speaker’s off axis response is quite close to the on axis response. That's the sort of response I associate with the term “constant directivity” speakers: as the angle of measurement moves from 0° in front of the loudspeaker to wider and wider angles, the frequency response measurement tracks fairly well with the 0° response.
If that's so then wouldn't it be the case that as we improve the frequency response of the 0° measurement we also by direct association improve the frequency response of the off axis response?
The second question is: so you have a loudspeaker where the on axis response is a certain shape and the off axis response is actually quite different from that. If you improve the on axis response and therefore cause a change to the off axis response, given that the off axis response is probably pretty bad, and probably un-correlated with the on axis response, is it plausible that the off axis response could in fact be improved instead of degraded, and that there is no way of knowing in advance until you actually do a listening test or off axis measurement?
Hope this all makes sense. If any knowledgeable people can shed light on these questions it would be great to hear from them. Apologies in advance if these seem naive questions to be asking.
Hi-resolution Phase-Linear EQ will never be efficient, but I don't care as long as it works for me. If the computer runs at 15% or 95% CPU load really doesnt matter I think. Of course - convolver type of filters are much more efficient. The same goes for (phase-warping) IIR digital EQ plugins.
There are spesific reasons that PLParEQ uses a lot of CPU. One of them is that it upsamples the data before processing.
I can only answer for myself here: I use a soundcard (RME Fireface 800) that bypasses the kmixer. I also use a pure digital source (Squeezebox), so there is no A/D conversion involved either.
That is why Constant Directivity is important. Frequency response should be the same both on-axis and off-axis. It is possible to do that over a large frequency range. It will require that all drivers are operated below the dipole peak.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.