This is where adding a second W4 can make a difference. You will reach ~91dB @300Hz, and all you have lost is tiny bit of vertical coherence and a couple of bucks !
Better than going to 4 way trouble, IMO.
Is this where I get to say I told you so?
Yes, you did!
Bratislav - I'm still on the fence about whether I should change the mid setup - as far as I can tell, the mid's aren't actually having any issues. Adding another W4-1320 is one possibility - its all about cost versus complexity.
I'm leaning towards a 4-way, and the primary reason is because of the tweeter - it may need a higher XO point. It seems to be the limiting factor when it comes to max SPL output. I may make that a whole new design though. As it is now, the three drivers give plenty of SPL output (well, I guess it's 10 actually...).
Yes, you did!
The Linux+BruteFIR+Octave solution discussed a page or two back is free in dollar terms. For Windows Allocator (USD 150), PLParEQX3 (USD 50), and some sort of host (USD 50-75) is the cheapest option I know. Don't know of any Mac solutions which I'd call low cost.
I think its the card that costs the most. I haven't found one below $400 worthy of an active solution. Something with the quality of LynxTwo-B is going to be $900
Maybe the M-audio Delta 410 is good enough?
There's an M-Audio Firewire 410 lots of folks use and a Delta 44. Not sure about a Delta 410. Generally speaking I'd suggest an interface with balanced outputs. But you can cost reduce a bit if you're fine with unbalanced. My experience is USB interfaces are best avoided, but your mileage will vary.
I use a Focusrite Saffire Pro 40 I eBayed new for USD 350 shipped. As a DAC it edges out my Wolfson 8740 based Cambridge Azur 640C, so I'd say it's possible to get good results under $400. If you only need three way I've seen the Saffire Pro 24 routinely eBay new for USD 225 shipped. In comparision my software stack costs USD 275. Besides, you know you want a good interface for measurements, right?
If you really want to shave bucks and don't mind a bit of a wait, Thuneau tends to run Black Friday deals. If you have a bit more budget the Motu Ultralite has a box spec that suggests all warped phase operations can be offloaded to its DSP. Interesting option for reducing CPU load or running standalone without the PC if need be.
I use a Focusrite Saffire Pro 40 I eBayed new for USD 350 shipped. As a DAC it edges out my Wolfson 8740 based Cambridge Azur 640C, so I'd say it's possible to get good results under $400. If you only need three way I've seen the Saffire Pro 24 routinely eBay new for USD 225 shipped. In comparision my software stack costs USD 275. Besides, you know you want a good interface for measurements, right?
If you really want to shave bucks and don't mind a bit of a wait, Thuneau tends to run Black Friday deals. If you have a bit more budget the Motu Ultralite has a box spec that suggests all warped phase operations can be offloaded to its DSP. Interesting option for reducing CPU load or running standalone without the PC if need be.
Last edited:
I do things pretty cheap - but there is always a little sacrifice when it comes to usability.
I use the Asus Xonar D2 - 2 in, 8 out. Its about $125. The only thing is its quirky, and not quite meant to be used like other 'prosumer' cards.
For software, I use Reaper. Its basically free, and comes with the basic plugins needed to do XO and EQ. The only thing is the 'linear phase' EQ - I haven't found a free version, but I haven't looked. I'm guessing a thorough search might turn something up. But then again, using linear phase EQ might not be necessary to doing phase correction.
I use the Asus Xonar D2 - 2 in, 8 out. Its about $125. The only thing is its quirky, and not quite meant to be used like other 'prosumer' cards.
For software, I use Reaper. Its basically free, and comes with the basic plugins needed to do XO and EQ. The only thing is the 'linear phase' EQ - I haven't found a free version, but I haven't looked. I'm guessing a thorough search might turn something up. But then again, using linear phase EQ might not be necessary to doing phase correction.
twest - they seem to. What makes you think otherwise?
The last few days, I've been working on a tangent. I got a bug in my bonnet - my memory of the Orions were as a little on the warm side, with a pretty smooth character - almost too smooth in the treble. Listening to my Plutos, I get the same impression. But listening to the Violets, I feel like they sound relatively analytical, with some emphasis or hardness in the treble, and I was wondering why (if it is actually true).
In the past, I had postulated that the smooth/warm character of the Orions and Plutos was due to some aspect of the ASP (analog signal processor) - lots of opamps? Simple power supply? So what I've done the past few days was hack into one of the Pluto ASP boards, install a DPDT switch, and then recreate the ASP's filter function in my DSP (digital signal processor) computer. I can then switch between the ASP and DSP, and instantly compare the analog and digital versions.
After a bit of fussing, good news - no difference! I was concerned, but apparently there is no need to. I'm pretty glad - if there were differences, that would require a bit of reevaluation. I think one of the important points here is that, while both routes are doing a lot of signal manipulation, they are doing it fairly different ways - but the final results end up audibly the same. One less thing to worry about!
Of course the necessary qualifier - these results only apply to my listening under the specified circumstances - generalize at your own risk!
I did some measurements I could post, showing that the filter functions are virtually the same. I did learn one thing that is of concern - both types of filters have relatively limited S/N ratios - like 50dB. I'll probably try to address that at some point.
The last few days, I've been working on a tangent. I got a bug in my bonnet - my memory of the Orions were as a little on the warm side, with a pretty smooth character - almost too smooth in the treble. Listening to my Plutos, I get the same impression. But listening to the Violets, I feel like they sound relatively analytical, with some emphasis or hardness in the treble, and I was wondering why (if it is actually true).
In the past, I had postulated that the smooth/warm character of the Orions and Plutos was due to some aspect of the ASP (analog signal processor) - lots of opamps? Simple power supply? So what I've done the past few days was hack into one of the Pluto ASP boards, install a DPDT switch, and then recreate the ASP's filter function in my DSP (digital signal processor) computer. I can then switch between the ASP and DSP, and instantly compare the analog and digital versions.
After a bit of fussing, good news - no difference! I was concerned, but apparently there is no need to. I'm pretty glad - if there were differences, that would require a bit of reevaluation. I think one of the important points here is that, while both routes are doing a lot of signal manipulation, they are doing it fairly different ways - but the final results end up audibly the same. One less thing to worry about!
Of course the necessary qualifier - these results only apply to my listening under the specified circumstances - generalize at your own risk!
I did some measurements I could post, showing that the filter functions are virtually the same. I did learn one thing that is of concern - both types of filters have relatively limited S/N ratios - like 50dB. I'll probably try to address that at some point.
Reaper 'supports' Linear Phase eq via VST plugins, but no such linear phase eq is supplied with Reaper itself.
I'm planning to try setting up Convolver VST to do a 3 way xover in Reaper as soon as I get time to try it (which may not be as soon as I'd like). In theory this should work, but I'm not entirely sure how multi-channel vsts work in Reaper. Hopefully it won't require 3 separate 2-channel instances.
ASP vs DSP
Wow. If it were me I would be very disappointed that the pc crossover didn't sound much better.
Wow. If it were me I would be very disappointed that the pc crossover didn't sound much better.
Why should it?
Cuibono, I think the lack of warmth is a matter of applying the right EQ to the bass. The mid-treble region of your speakers is as good as it gets. What I do to get the right timbre and natural warmth, is EQ the range below about 500 hz to get a smoothly descending curve at the listening postion (spatial average of multiple measurements), of about 1 dB/octave. The mid to high range already has this downward slope without any EQ. At this point I think the speaker is no longer the weakest part of my system - it is the room. My next project will be unobtrusive room treatment.
sendler - why should the digital version sound better than the analog version. Just for the record, here is Linkwitz's ASP. Fairly sophisticated, IMO.
keyser - thanks for the suggestion, I may try it. I guess it really comes down to preference. Some other weak points worth mentioning are the recording (bigger weakness compared to the room, IME), and one's state of mind.
In another thread, someone suggested listening in complete darkness (as in, its totally black, you can't see anything). I tried it, and it's really interesting! My sense of hearing becomes much more acute, and my perception of the recording really changes. Pretty enjoyable, actually.
keyser - thanks for the suggestion, I may try it. I guess it really comes down to preference. Some other weak points worth mentioning are the recording (bigger weakness compared to the room, IME), and one's state of mind.
In another thread, someone suggested listening in complete darkness (as in, its totally black, you can't see anything). I tried it, and it's really interesting! My sense of hearing becomes much more acute, and my perception of the recording really changes. Pretty enjoyable, actually.
What dwk123 said:
dwk123 said:
If the measurements are acoustic that sounds roughly normal depending on frequency. If the measurements are electric you should be seeing better than that unless your measurement gear has noise issues.
Assuming both are warped phase (and using unbalanced interconnects) I wouldn't expect much difference so long as the digital version is a good replication the ASP's magnitude and phase response. The PC crossover can, in principle, outperform the ASP. Depends on the audio interface ADC (if applicable), op amps, DAC, and layout. I'd need details of Cuibono's setup---preferably including internal photos of the interface and part numbers of the chips used---to speculate. A well executed digital crossover will also have lower noise and nonlinear distortion floors than the ASP. How much that matters depends on typical signal levels through the crossover and ambient noise sources like fridges, furnaces, busy roads, children, small dogs...
IMO the main win for a PC crossover---as opposed to other digital crossover solutions---is linear phase.
Multiple dacs, no actives, no passives
Multiple dacs sharing the workload instead of one, connected directly to the amps eliminating multiple stages of opamps and countless passive components. And beyond that the most important attribute of pc cross that wasn't part of this comparison, the complete flexibility to optimize the drivers to your specific room. You know. All of the usual pc cross stuff.
Cuibono,
Excellent work! Those polars are very impressive. I think it was your thread on trying to maintain off axis response for dipoles that made me give up on starting a dipole project. I didn't realize you've had so much success in the end. I may have to try a dipole for my next project.
I'm wondering about your current filter setup. I went through the opposite progression as you did it seems. I started out running convolver with Matlab generated brick wall filters. Then played around a bit with the Sonitus plug ins that were bundled with Sonar. Finally, I broke down and got Allocator and have been very happy with it. Your conversion to linear phase filters makes we want to revisit that. At this point, how do you design your filters? You mentioned a few posts back that you first tried to recreate the transfer function from your min phase filters. How exactly did you do this? Does your filter design take into account the phase of the drivers or is it just the electrical portion that is linear phase? Did polar response change when you went to linear phase filtering?
Dan
Excellent work! Those polars are very impressive. I think it was your thread on trying to maintain off axis response for dipoles that made me give up on starting a dipole project. I didn't realize you've had so much success in the end. I may have to try a dipole for my next project.
I'm wondering about your current filter setup. I went through the opposite progression as you did it seems. I started out running convolver with Matlab generated brick wall filters. Then played around a bit with the Sonitus plug ins that were bundled with Sonar. Finally, I broke down and got Allocator and have been very happy with it. Your conversion to linear phase filters makes we want to revisit that. At this point, how do you design your filters? You mentioned a few posts back that you first tried to recreate the transfer function from your min phase filters. How exactly did you do this? Does your filter design take into account the phase of the drivers or is it just the electrical portion that is linear phase? Did polar response change when you went to linear phase filtering?
Dan
Hi Dan, all good questions.
I design my filters by measuring their on axis response, and EQing/XOing it till it hits the bandpass I want (like LR4 at 275 and 2600Hz). I'm relatively strict about not using drivers beyond the point where the off axis radiation differs from their on axis, which makes things a little more difficult, in terms of driver selection, but easier because then you don't need to worry about off axis response differences. Measurements are done at about 2m distance, about 3m in the air, so the results are anechoic and fairly high resolution. What I do then is point the axis of each driver at the ear - this assures equal level balance and phase summation between each section (woofers, mid, high).
When I switched to 'linear phase' software, I didn't have to remeasure things acoustically - but I did electrically. The way the two pieces of software use the parametric values (Freq, Q, dB) varies, so there was something of a translation process. I will in the future recheck the final acoustic response, including the differences in phase. So far, I've only been playing with the electrical portion of things - I haven't yet done the necessary phase measurements, or found software that allows me to manipulate phase (easily). I haven't measured the polar response since changing filter types, but I don't see why it would change the polar response (at least much) - polar response seems to be only a function of the driver's physical geometry.
If you (or anyone else) could post a simple/concise how to on generating linear phase filters for use in a convolver, I'd really appreciate it. I've read up on it some, but it is pretty dense! Nothing like a simple overview, or a concise hands-on step by step process. I think I was last looking at BruteFIR.
I design my filters by measuring their on axis response, and EQing/XOing it till it hits the bandpass I want (like LR4 at 275 and 2600Hz). I'm relatively strict about not using drivers beyond the point where the off axis radiation differs from their on axis, which makes things a little more difficult, in terms of driver selection, but easier because then you don't need to worry about off axis response differences. Measurements are done at about 2m distance, about 3m in the air, so the results are anechoic and fairly high resolution. What I do then is point the axis of each driver at the ear - this assures equal level balance and phase summation between each section (woofers, mid, high).
When I switched to 'linear phase' software, I didn't have to remeasure things acoustically - but I did electrically. The way the two pieces of software use the parametric values (Freq, Q, dB) varies, so there was something of a translation process. I will in the future recheck the final acoustic response, including the differences in phase. So far, I've only been playing with the electrical portion of things - I haven't yet done the necessary phase measurements, or found software that allows me to manipulate phase (easily). I haven't measured the polar response since changing filter types, but I don't see why it would change the polar response (at least much) - polar response seems to be only a function of the driver's physical geometry.
If you (or anyone else) could post a simple/concise how to on generating linear phase filters for use in a convolver, I'd really appreciate it. I've read up on it some, but it is pretty dense! Nothing like a simple overview, or a concise hands-on step by step process. I think I was last looking at BruteFIR.
Polar response is a function of the acoustic size of the source and its radiation pattern. Within the transition band of a crossover the source consists of both drivers and the radiation pattern's a function of the drivers' relative signal strengths and phases. If the crossover slope and phase tracking didn't change in converting from warped to linear phase (e.g. you switched from a warped phase even order LR or odd order Butterworth alignment to the linear phase version of the same) then the polar response won't change.
DDMF's LP10 has the ability to vary an equalizer band's phase response from warped to linear to inverse warped. Interesting, although not the pure phase adjuster we've been looking for. I tried the demo last night and, unfortunately, found the sound quality to be sufficiently poor detailed investigation didn't seem worthwhile.
Doc entry point for Matlab's FIR synthesis functions is here (Octave is a clone of Matlab). Basic flow is to call the functions and load the returned coefficients into Convolver/SigmaStudio/whatever; sounds like you may have that much already, in which case following the MathWorks examples is probably a good starting point. I've not done audio oriented FIR design so I can't comment on the best choices for managing preringing and such.
DDMF's LP10 has the ability to vary an equalizer band's phase response from warped to linear to inverse warped. Interesting, although not the pure phase adjuster we've been looking for. I tried the demo last night and, unfortunately, found the sound quality to be sufficiently poor detailed investigation didn't seem worthwhile.
Doc entry point for Matlab's FIR synthesis functions is here (Octave is a clone of Matlab). Basic flow is to call the functions and load the returned coefficients into Convolver/SigmaStudio/whatever; sounds like you may have that much already, in which case following the MathWorks examples is probably a good starting point. I've not done audio oriented FIR design so I can't comment on the best choices for managing preringing and such.
cuibono and twest820,
Here is a thread I started a couple of years ago when I was playing around with convolver and fir filters generated by Matlab before I bought Allocator. It includes the script I ran in Matlab to generate filters. I think I found this script somewhere on this forum and it might work as is in octave as well.
http://www.diyaudio.com/forums/digital-line-level/129199-generating-running-fir-filters-active-loudspeakers.html
The basic workflow was to generate the filter coefficients in Matlab and read them into convolver. I had convolver running as a vst plugin in Sonar. However, everything didn't work out so well. The stop band attenuation was only 44 db. I found that out by running sine waves into the filter and reading the output in my soundcard's mixer panel (patchmix for emu 1820m). I saw that at 20 Hz, the tweeter's filter had only attenuated the signal by 44 dB rather than the 100+ predicted when running the filter in Matlab. I suspected this was due to convolver's lower numerical accuracy compared to Matlab. Some of the feedback I got from the forum is that brick wall filters have tradeoffs associated with stop band attenuation. My plan at the time was to try to generate shallower filters and hopefully get better stopband performance. However, I tried out Allocator and it sounded so good and was much easier to tune. I haven't spent the time to set up more complicated filter scripts in Matlab yet. This thread has sparked my interest though.
That's the thing that prompted my question. It seems to me like it's not straightforward to go from a min. phase filter to a linear phase one. Let's take a simple example. Let's say we have a woofer and a tweeter that are coincident. Let's also say the tweeter just happens to have an LR4 rolloff at 2000 Hz already. First, let's apply minimum phase equalization to the woofer, either through a passive crossover or digitally with IIR filters for example, so that the acoustic amplitude response of the woofer is LR4 low pass at 2000 Hz. The acoustic response of the woofer+filter will still be minimum phase and the woofer and tweeter will sum to give us flat response.
If we replace the filter we used to get our LR4 reponse from the woofer with a linear phase filter that has the same amplitude response as our original filter, we will get the same acoustic amplitude from the woofer. However, woofer+filter won't be minimum phase anymore, but it probably won't be linear phase either. The tweeter will be minimum phase though. It seems to me that when we combine the woofer and tweeter we'll probably get a new amplitude response, at least in the crossover region because we've changed the phase tracking between the two. Does this make any sense? To get a truly linear phase response, we'd need to take into account the amplitude and phase response of the woofer in our filter design to get linear phase acoustically. With the tweeter, even though it already hits our LR4 slope, we still need to design an FIR filter to correct the phase to linear and match the woofer. Thus, neither the woofer nor tweeter filters would be linear phase filters, but if we designed the filters correctly the final acoustic response of both would be linear phase.
This is my impression of what Acourate can do and is what I was toying around with trying to set up in Matlab. Maybe I'm way off base with all this and just confused? Cuibono, I noticed your measurement plots were taken with Soundeasy. I think Soundeasy can generate FIR filter coefficients can't it? Maybe you can do something along the lines of what I've described in Soundeasy? I'd be interested to see your results if you remeasure your setup with the linear phase filters. I hope this all doesn't come off as sounding negative as I'm really impressed with all the work you've done on this speaker. You've opened my eyes to what can be done with dipoles and I'm thinking something similar to what you've done here will probably be my next project.
Dan
Here is a thread I started a couple of years ago when I was playing around with convolver and fir filters generated by Matlab before I bought Allocator. It includes the script I ran in Matlab to generate filters. I think I found this script somewhere on this forum and it might work as is in octave as well.
http://www.diyaudio.com/forums/digital-line-level/129199-generating-running-fir-filters-active-loudspeakers.html
The basic workflow was to generate the filter coefficients in Matlab and read them into convolver. I had convolver running as a vst plugin in Sonar. However, everything didn't work out so well. The stop band attenuation was only 44 db. I found that out by running sine waves into the filter and reading the output in my soundcard's mixer panel (patchmix for emu 1820m). I saw that at 20 Hz, the tweeter's filter had only attenuated the signal by 44 dB rather than the 100+ predicted when running the filter in Matlab. I suspected this was due to convolver's lower numerical accuracy compared to Matlab. Some of the feedback I got from the forum is that brick wall filters have tradeoffs associated with stop band attenuation. My plan at the time was to try to generate shallower filters and hopefully get better stopband performance. However, I tried out Allocator and it sounded so good and was much easier to tune. I haven't spent the time to set up more complicated filter scripts in Matlab yet. This thread has sparked my interest though.
That's the thing that prompted my question. It seems to me like it's not straightforward to go from a min. phase filter to a linear phase one. Let's take a simple example. Let's say we have a woofer and a tweeter that are coincident. Let's also say the tweeter just happens to have an LR4 rolloff at 2000 Hz already. First, let's apply minimum phase equalization to the woofer, either through a passive crossover or digitally with IIR filters for example, so that the acoustic amplitude response of the woofer is LR4 low pass at 2000 Hz. The acoustic response of the woofer+filter will still be minimum phase and the woofer and tweeter will sum to give us flat response.
If we replace the filter we used to get our LR4 reponse from the woofer with a linear phase filter that has the same amplitude response as our original filter, we will get the same acoustic amplitude from the woofer. However, woofer+filter won't be minimum phase anymore, but it probably won't be linear phase either. The tweeter will be minimum phase though. It seems to me that when we combine the woofer and tweeter we'll probably get a new amplitude response, at least in the crossover region because we've changed the phase tracking between the two. Does this make any sense? To get a truly linear phase response, we'd need to take into account the amplitude and phase response of the woofer in our filter design to get linear phase acoustically. With the tweeter, even though it already hits our LR4 slope, we still need to design an FIR filter to correct the phase to linear and match the woofer. Thus, neither the woofer nor tweeter filters would be linear phase filters, but if we designed the filters correctly the final acoustic response of both would be linear phase.
This is my impression of what Acourate can do and is what I was toying around with trying to set up in Matlab. Maybe I'm way off base with all this and just confused? Cuibono, I noticed your measurement plots were taken with Soundeasy. I think Soundeasy can generate FIR filter coefficients can't it? Maybe you can do something along the lines of what I've described in Soundeasy? I'd be interested to see your results if you remeasure your setup with the linear phase filters. I hope this all doesn't come off as sounding negative as I'm really impressed with all the work you've done on this speaker. You've opened my eyes to what can be done with dipoles and I'm thinking something similar to what you've done here will probably be my next project.
Dan
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.