That seems a reasonable explanation. But comparing like this is like apples and oranges. Picking the right Xover frequency is about more things though, like radiation pattern and actual distortion profile. Plus: breakup distortion is completely different from harmonic distortion.
You are making good progress very quickly, just don't burn yourself out. Supposed flat-summation XO is hardly a guarantee of low-distortion transcient response. I think, this audio science is both much simpler in the essentials and much more complicated (i.e. distorted) than the commercial audio companies (and many diyers) would admit. I'll borrow again what @tmuikku showed me: simulated impulse responses from bandwidth-poor single driver and ideal 2-way; I judged one to be very bad and the other still worse (much). The input-output "difference measure" that I came up with and eyeballed, was a very simple one: area in common under both input and output curves, versus their symmetrc difference i.e. area bounded by one curve but not the other. No one would pay good money for a pair of speakers labelled "impulse response fidelity 20%" would they? THD 0.5% so it became. The gross distortions that you (and most people) could hear, merely got Totalled away to near-zero; the industry was saved whew.
Last edited:
Fortunately human hearing is not sensitive to phase and impulse "anomalies". Most likely bats and whales who use sonar detection would not be happy...
FIR dsp gives possibility correct impulse and some people say that they can detect the difference.
Regarding the topic, woofer/mid cone+surround resonances are bad and in general it is wise to rule them out of passband, and sometimes a notch is needed. Old school wisdom. This way HD gets much better.
Lars Risbo of Purifi explains more here (specifically the 2nd link)
https://purifi-audio.com/blog/tech-...o-whats-the-point-of-low-distortion-drivers-4
https://purifi-audio.com/blog/app-notes-2/low-distortion-filter-for-ptt6-5x04-naa-11
With a simple 2nd order low pass filter and serial notches, the peaks are virtually removed, and the frequency response approaches accurately a chosen 3rd order lowpass target response. Moreover, the THD is driven down into the noise floor of the measurement setup. There is no need to resort to extremely sharp filters or a low cross over frequency to handle the peaks using the shown methodology.
A filter for a final speaker-box application needs to have baffle step correction which was left out in the shown examples to allow high precision measurements in the 2pi anechoic chamber.
Finally, it was shown how different filter configurations resulting in approximately the same SPL response can have significantly different impact on the distortion of the system.
ps. So, serial pasive notch is better than parallel, because of impedance effect
FIR dsp gives possibility correct impulse and some people say that they can detect the difference.
Regarding the topic, woofer/mid cone+surround resonances are bad and in general it is wise to rule them out of passband, and sometimes a notch is needed. Old school wisdom. This way HD gets much better.
Lars Risbo of Purifi explains more here (specifically the 2nd link)
https://purifi-audio.com/blog/tech-...o-whats-the-point-of-low-distortion-drivers-4
https://purifi-audio.com/blog/app-notes-2/low-distortion-filter-for-ptt6-5x04-naa-11
Conclusions
The PTT6.5X04-NAA-08 aluminum cone driver has a very well-behaved frequency response with two break-up peaks at 5kHz and 10kHz.With a simple 2nd order low pass filter and serial notches, the peaks are virtually removed, and the frequency response approaches accurately a chosen 3rd order lowpass target response. Moreover, the THD is driven down into the noise floor of the measurement setup. There is no need to resort to extremely sharp filters or a low cross over frequency to handle the peaks using the shown methodology.
A filter for a final speaker-box application needs to have baffle step correction which was left out in the shown examples to allow high precision measurements in the 2pi anechoic chamber.
Finally, it was shown how different filter configurations resulting in approximately the same SPL response can have significantly different impact on the distortion of the system.
ps. So, serial pasive notch is better than parallel, because of impedance effect
Last edited:
I disagree with this based on my experience. I'd be happy to discuss, but I'm not willing to sit through the video to find it.
It’s an useless claim. You can’t substitute the woofer with a tweeter. So forget that video.
@AllenB it is 2-3 minutes before the middle of the video. This video has a lot on cone breakup and I would recommend it to anyone coming here exactly for this topic.
And I would like to point to my personal Tang Band W6 driver with a pic on topic of this thread (cone breakup):
This is FR taken on axis and at various angles up to 45+ degrees. I want to stress that I always measure not in the middle of the room at stand but at the computer desk 1 ft from the wall with table, walls and monitor reflections. This is the real listening position result. 600 Hz is the baffle step, not the driver. 250 Hz is the backwall.
My understanding is that as long as the driver movements are pistonic it radiates remarcably identical on wide ranges. And then the cone vibrations are becoming surface wave moving differently at different areas of the cone, which manifests with random fluctuations at different angles. It starts at 2,200 Hz, and that's where I need to cross. This woofer has the cone breakup frequency of 2,200 Hz.
I want to stress that those peaks at 3 and 4.5 are creating major, major audible distortions, this is not about measurement, I have 100s of hours of listening on this driver because I actually like it a lot.
The W6 tweeter has the resonance frequency of 1 kHz, so I need a very sharp tweeter rolloff to make it -30 Db from 2,500 to 1,000. And I want the woofer to be like -10 Db at 3,000.
Also, this woofer has a great punch and very clean pleasant bass, to keep it I want to keep only one single element at the signal path, and it will be 14 gauge 2 mF inductor, and do the rest with serial notch filters across the signal path, this doesn't affect the sound.
And I would like to point to my personal Tang Band W6 driver with a pic on topic of this thread (cone breakup):
This is FR taken on axis and at various angles up to 45+ degrees. I want to stress that I always measure not in the middle of the room at stand but at the computer desk 1 ft from the wall with table, walls and monitor reflections. This is the real listening position result. 600 Hz is the baffle step, not the driver. 250 Hz is the backwall.
My understanding is that as long as the driver movements are pistonic it radiates remarcably identical on wide ranges. And then the cone vibrations are becoming surface wave moving differently at different areas of the cone, which manifests with random fluctuations at different angles. It starts at 2,200 Hz, and that's where I need to cross. This woofer has the cone breakup frequency of 2,200 Hz.
I want to stress that those peaks at 3 and 4.5 are creating major, major audible distortions, this is not about measurement, I have 100s of hours of listening on this driver because I actually like it a lot.
The W6 tweeter has the resonance frequency of 1 kHz, so I need a very sharp tweeter rolloff to make it -30 Db from 2,500 to 1,000. And I want the woofer to be like -10 Db at 3,000.
Also, this woofer has a great punch and very clean pleasant bass, to keep it I want to keep only one single element at the signal path, and it will be 14 gauge 2 mF inductor, and do the rest with serial notch filters across the signal path, this doesn't affect the sound.
Linear vs. nonlinear distortion
https://www.css-audio.com/single-po...3qy4uex0tU5JgPSdiHOrsoLj4XAIwi7XqlcCDCRN8WdKk
https://www.klippel.de/fileadmin/kl...on caused by design_part 1_Klippel_Werner.pdf
Linear = deviation from flat response/phase
Nonlinear (harmonic and IMD) = added extra frequencies to signal
A peak in (on-axis) response does not introduce harmonic distortion, but it adds some relative distortion to it's dividers. Notching spl peaks reduces or eliminates that effect
https://www.css-audio.com/single-po...3qy4uex0tU5JgPSdiHOrsoLj4XAIwi7XqlcCDCRN8WdKk
https://www.klippel.de/fileadmin/kl...on caused by design_part 1_Klippel_Werner.pdf
Linear = deviation from flat response/phase
Nonlinear (harmonic and IMD) = added extra frequencies to signal
A peak in (on-axis) response does not introduce harmonic distortion, but it adds some relative distortion to it's dividers. Notching spl peaks reduces or eliminates that effect
True to a certain extent. THD only applies to a single tone. When one wants to find out what happens to a real life complex tone, things get much more complicated and THD is of no help. To analyze a complex tone one would need to know the underlying nonlinearity in all its orders. This is hidden in any THD measurement. In a complex tone case some harmonics may be lowered and others enhanced, it is not at all simple.
I agree that the discussion is going nowhere. But "You don’t really believe you actually can notice 0,1% of 1% distortion of any live musical instrument, do you?" is not actually true. I can create a nonlinearity with a THD of .1% that will sound terribble and another at THD of 10% that will be inaudible. The point here is that THD as a measure of sound quality is useless.
I've made playlist "Most challenging tracks to stress the HiFi system" specifically designed to expose a system defects not capable of reproducing complex music. It has 3 hours of good music, I was not putting there anything boring just to torture your ears. Please give it time, listen all the 3 hours without skipping on the best system you have access to. Later, you will be able to easily find the defects on any substandard system with the same recordings, and that will answer the question what distortion level you can hear. You will be surprised.
Also, I found it peculiar that it's easier to detect distortion with HD music than with CD quality music. I presume it's because of the harmonics count. A key played on grand piano may (I suspect) send the chain of 30 harmonics down to the amplifier, CD will record only 20 harmonics and MP3 only 10. Hence, as discussed above, the HD music will generate a lot more of distortion. It's also a lot more pleasant when playing simpler instruments that do not generate distortion. I can say that I can perfectly play Led Zeppelin, Dire Straits and Billie Eilish on any of my XO variants without immediately audible distortion because this kind of music doesn't require much resolution from the system. So you won't find this kind music in my playlist.
Also, I found it peculiar that it's easier to detect distortion with HD music than with CD quality music. I presume it's because of the harmonics count. A key played on grand piano may (I suspect) send the chain of 30 harmonics down to the amplifier, CD will record only 20 harmonics and MP3 only 10. Hence, as discussed above, the HD music will generate a lot more of distortion. It's also a lot more pleasant when playing simpler instruments that do not generate distortion. I can say that I can perfectly play Led Zeppelin, Dire Straits and Billie Eilish on any of my XO variants without immediately audible distortion because this kind of music doesn't require much resolution from the system. So you won't find this kind music in my playlist.
What woofer is it again? Too-close-together passive notch filters would interact and a single notch filter would leave a deep(er) valley between 3khz and 4.5khz. So a cocktail fix might be required. You might try applying a coating or (on dustcap and non-paper cone) automotive stretchy wire-bundling tape (variously-branded-plast), a miracle low-mass material that is removable if it didn't work. Also try a wool disk ~8cm to physically absorb much of the peak 4.5khz. My many minimalist LXs (some XO-less) use a combination of absorber (between the drivers) and natural off-axis fall-off to attenuate woofer cone breakup.
@wchang tb w6-2313 coaxial. I accept some 3 Db wiggles in the response as an inevitable price for having the perfect imaging and drilling only one hole in the small desktop setup. I don't feel any difference in sound when I move my chair or decide to sit deeper. With 60 cm distance from my ear to the cone it's a nice feature. You saw above 0 and 45 degrees are identical. I understand it's impossible to make it flat. I will play with wool, good idea, thanks, will take some time though. I want to put wool over perfect XO, and not making XO over wool 🙂
I use small synthetic selfadhesive felt pads to tame cone resonances of SEAS MR18 woofer. I didn't save .mdat files.
Attachments
Last edited:
The thread from 2018 is here https://www.diyaudio.com/community/threads/seas-mr18-3-way.322839/
Overlays of two very different xos for M/T (without felt pads). Most saved measurements are with too low spl to see distortion.
Overlays of two very different xos for M/T (without felt pads). Most saved measurements are with too low spl to see distortion.
I merely addressed the discussion about distortion generated by distortion signals, causing a nnlinearty of 0,001%. Ten times as much won’t cause trouble. By far the most would be even order. And these things show up in spectral analysis or Farina sweeps, be it not separated. Academic: yes. Practical: no.
About using harmonic distortion to assess the Intermodulation distortion, which is indeed hard to measure because it requires to play multiple frequencies. I'll post again the same measurement as above #189 but with 3rd HD:
We are talking about the same 3 kHz cone breakup from #189. Here it creates the peak at 1 kHz, as 1,000 * 3H = 3,000.
This peak is at 42 Db, with main signal at 84 Db, so it's 42 Db = 0.01% distortion measured as a harmonic one.
So, can I hear a 0.01% harmonic distortion? Probably not, I won't bother trying to set up a listening test trying to detect it.
Can I hear the IMD that can be measured with HD test that pinpoints the distortion area with 0.01% peak at HD3 graph?
Absolutely. I don't just hear it, it hits me with pain in the ears with dozens of test tracks listed in #193, I grade this distortion well above acceptable and I am working for weeks trying to find the XO to fix it. I refuse to accept my speakers in this state until it's gone.
So the conclusion is that the 0.01% Intermodulation distortion measured with H3 is way louder than the audibility threshold. I suspect that threshold will be at the end somewhere in the 0.003% area but I will report the exact number when I have it.
We are talking about the same 3 kHz cone breakup from #189. Here it creates the peak at 1 kHz, as 1,000 * 3H = 3,000.
This peak is at 42 Db, with main signal at 84 Db, so it's 42 Db = 0.01% distortion measured as a harmonic one.
So, can I hear a 0.01% harmonic distortion? Probably not, I won't bother trying to set up a listening test trying to detect it.
Can I hear the IMD that can be measured with HD test that pinpoints the distortion area with 0.01% peak at HD3 graph?
Absolutely. I don't just hear it, it hits me with pain in the ears with dozens of test tracks listed in #193, I grade this distortion well above acceptable and I am working for weeks trying to find the XO to fix it. I refuse to accept my speakers in this state until it's gone.
So the conclusion is that the 0.01% Intermodulation distortion measured with H3 is way louder than the audibility threshold. I suspect that threshold will be at the end somewhere in the 0.003% area but I will report the exact number when I have it.
I want to thank @Juhazi for posting several links that explained me my mistake with XO and understanding the graphs. For weeks, I thought that the cone breakup is the 4,300 peak at #189 and #199. After following all the threads I've realized that this was the 3,000 peak that I neglected that has to be suppressed. I have yet to assemble the 21st version of XO but I'm hopeful that's the one. At least I finally understand the problem. I am a noob, that's my first DIY speaker.