To reduce the third order distortion just by x-over, the x-over point needs to be quite low. It does not help with steep filter at 2-3 kHz if the raised distortion occurs at 1.5 kHz. You need the x-over at 1kHz or lower.Nice testing.
My take on notch filtering, when I worked a couple of years with my HYBRID loudspeaker project, this was an active dsp loudspeaker with some passive components between the tweeter-power amp ( a capacitor equalizing the waveguide + a resistor to lower harmonic tweeter distortion, and trying to notch out the Seas er18rnx resonances with a passive filter, all set in two boxes outside the speaker ) .
Measurement microphone was line audio om1 , Audient id14 and Audiotools with an ipad.
1. Unfortunaltely my two Seas er18rnx had different resonances at different frequencies - one had a peak at 4,3 kHz , and the other had its peak at 4,8 kHz . I had to use different components for the passive notch for L and R speaker .
A pain to do passive, but easy to do with the dsp crossover ( dbx pa2 ) .
2. Even then - I thought that the sound was better without the passive notch , and instead using a dsp notch .
With a dsp-notch its easy to get the exact Q value of the notch.
3. I tried many options, but the best sound was without any notch filtering at all, using steeper crossovers instead . A dsp 12 dB/oct with a passive or active notch filtering sounded worse than just using a dsp 24 dB/oct or 48 dB/oct crossover - without notch filtering.
4. My conclusion of this after some years of comparing, is that its probably much better to use drivers that dont need any notch filtering in the beginning, these will sound better in the midrange . This is the way Genelec does it, as you can see with the M040 and 8030 monitor .
5. For an even better, clearer sound - a dsp crossover without an A/D is probably the way to go, keeping the signal digital as long as you can.
.View attachment 239519View attachment 239520View attachment 239522
With a small coil in series with the midrangedriver, one can lower the harmonic distortion somewhat , but the sound with real music will be unnafected or worse, because this trick doesnt lower the IMD . IMD is much more important to keep low than a lower harmonic distortion .Active speakers can include impedance corrections at the output stage. But this is another level of speaker than just using DSP x-overs and conventional amplifiers.
How do you know, do you have measurements of their drivers without crossover?4. My conclusion of this after some years of comparing, is that its probably much better to use drivers that dont need any notch filtering in the beginning, these will sound better in the midrange . This is the way Genelec does it, as you can see with the M040 and 8030 monitor .
Well, I havent had that much experience with metal cone drivers , except that I often prefer the sound from metal tweeter domes , so maybe I have drawn the wrong conclusions . Some newer metal drivers like SB NBAC 15 have its resonanses higher in frequency then my seas er18rnx , - this SB Acoustics driver seems to be really interesting as a midrange in a big threeway speaker .How do you know, do you have measurements of their drivers without crossover?
By the way my personal (also like yours non scientifically proven) experience is different, I prefer hard membrane drivers as long as the resonance is far away and suppressed from their used frequency region, soft and overdamped membranes just smear their modes to a lower and broader region which can often be their used frequency band which I think makes a part of their audible difference (plus of course their different radiation due to the decreasing active emitting surface).
They are generally Peerless HDS drivers (for the 8xxx series). They met be modified version though. But the stock ones are very well regarded and you can find plenty of measurements of them. They are excellent.How do you know, do you have measurements of their drivers without crossover?
I have also heard that they are Peerless and am sure they are excellent, but would like to see measurements of them as a driver can be also excellent with a resonance peak above its usage band like for example the Purify.They are generally Peerless HDS drivers (for the 8xxx series). They met be modified version though. But the stock ones are very well regarded and you can find plenty of measurements of them. They are excellent.
They are made by Thympany , probably with spec from Genelec. The numbers on the drivers are not ordinary HDS DIY units.I have also heard that they are Peerless and am sure they are excellent, but would like to see measurements of them as a driver can be also excellent with a resonance peak above its usage band like for example the Purify.
Thank you.Nice testing.
I recall you mentioned this before and I agree, it is unfortunate. One (or both) of the drivers appears to have an issue by your description. Without seeing the data, I would suspect the one with the 4.3kHz peak is the issue since the driver should have peak at 4.8kHz according to Seas' specs. For example, I have four of the W18E001 v1 and they all measure the same in every way including the surface breakup mode at 4.3kHz even after 20 years of use in multiple cabinets. And I showed they are matched to Seas' datasheet too (see above!). And, I have two newer W18E001 v2 which also have resonance matched to each other and to Seas's spec even though one of them has a delaminated surround. I needed to look at the distortion measurement to see that issue! And the issue was bad enough to hear the distortion due to the surround flapping around (again, data posted earlier in this thread!). Regarding your situation, one of your drivers is likely broken.My take on notch filtering, when I worked a couple of years with my HYBRID loudspeaker project, this was an active dsp loudspeaker with some passive components between the tweeter-power amp ( a capacitor equalizing the waveguide + a resistor to lower harmonic tweeter distortion, and trying to notch out the Seas er18rnx resonances with a passive filter, all set in two boxes outside the speaker ) .
Measurement microphone was line audio om1 , Audient id14 and Audiotools with an ipad.
1. Unfortunaltely my two Seas er18rnx had different resonances at different frequencies - one had a peak at 4,3 kHz , and the other had its peak at 4,8 kHz .
I agree that that passive is harder than DSP for a DIY, hobbyist, or casual builder. Right now, my knees are hurting from crawling around underneath the test baffle swapping out passive filters. I built that baffle so I didn't have to swap out drivers in the FFLOTSAM cabinets to keep them looking OK with all of this fiddling! Even though I built a fixture for the LCR notch filter, I still take a few minutes to swap out components to change the filter. So yeah, passive takes minutes (or more) and effort to implement a change that takes a just a moment and a few keystrokes with DSP. If you are doing DIY, or fiddling around with measurements like I am, then yes passive can be a pain. It would be a different type of problem if you were a speaker manufacturer and had to mitigate large variability like you describe in the production line. But, the thing is, by your description you got a broken driver. The mitigation a large manufacturer would take in your case isn't custom filters, it's remove from production and return to vendor for replacement. I agree with you strongly about how easy DSP is for the average user, but the point you make using the defective driver is not a valid argument. This is an example of where you depart from objective and reasonable. Lastly, a speaker manufacturer will have a much different perception of how much of a 'pain' passives are than you or me. Especially, passives actually require you have a basic understanding of circuits, components, and some prototyping/fabricating skills. Speaker manufacturers have these competences, we less so.I had to use different components for the passive notch for L and R speaker .
A pain to do passive, but easy to do with the dsp crossover ( dbx pa2 ) .
I don't know how you can make a sweeping generalization like 'thought that the sound was better' even if you had two drivers that were working the same. With one of drivers having such a defective response compared to the other, your argument doesn't make sense. On the flip-side, some months ago, you were making arguments for hyper-matched capacitors and inductors having really big impacts on (for instance) Left/Right balance of speakers. I recall showing you the actual impact of components that are within tolerance in real-world passive crossover filters was very small. It is odd that you are using an obviously out of tolerance L/R woofer mismatch as the basis of this argument today, it's really upside down...2. Even then - I thought that the sound was better without the passive notch , and instead using a dsp notch .
No, there are equations. Requires some basic circuit skills and speaker manufacturers don't struggle with that, at least the good ones. I don't usually either. DSP just turns the equations into sliders in a GUI so the lay person can easily operate in moments. And I do love those sliders!! Yes, DSP filter flexibility is unmatched by passives, but that's not what I'm testing here. I am trying to find out more about the distortion reduction observed in the PuriFi whitepaper. What PuriFi showed was a very tangible reduction in distortion with a passive notch, and they discussed the physical mechanism behind that distortion reduction. And, given their paper, it seemed worth exploring. And, I am finding that regarding distortion, PuriFi is right. And that the distortion reduction isn't just at the fundamental frequency, but throughout the audible band with interesting harmonics. I also think it matters now especially since PuriFi, Seas, Scan Speak, etc. have so many ultra-low distortion drivers. If you look at the published frequency responses, many of them they have resonant breakup modes. Here are a few Scan Speak:With a dsp-notch its easy to get the exact Q value of the notch.
You were chasing ghosts with that bad woofer. But yes, a steeper slope is going to help with the resonance. A lower crossover frequency also helps. With your bad driver's resonance pushed down by another 500Hz, these would help even more.3. I tried many options, but the best sound was without any notch filtering at all, using steeper crossovers instead . A dsp 12 dB/oct with a passive or active notch filtering sounded worse than just using a dsp 24 dB/oct or 48 dB/oct crossover - without notch filtering.
I don't know that the drivers used in Genelec are intrinsically flat. Your conclusions about driver resonances and associated difficulties seem to be deeply convoluted with having a driver that measured way out of tolerance. As I reported above, I had an issue with one of my Seas woofers. I just got the replacement in the mail today, I recommend you contact your seller and/or Seas, you should have the measurements ready for them. As you can see above, Seas/Madisound replaced my bad driver as soon as I contacted them.4. My conclusion of this after some years of comparing, is that its probably much better to use drivers that dont need any notch filtering in the beginning, these will sound better in the midrange . This is the way Genelec does it, as you can see with the M040 and 8030 monitor .
This is a different topic, not relevant to the passive notch and reduction in distortion that I am investigating here...5. For an even better, clearer sound - a dsp crossover without an A/D is probably the way to go, keeping the signal digital as long as you can.
I like the look of the finished speaker. Can I ask, what is the network you show?
The ones on the picture is for the tweeter, one capacitor used for eq ( and the same time protection because of the active approach ) of the waveguide combined with a resistor to get a slightly higher driving impedance for less dynamic compression. The actual crossover slopes was done in dbx pa2, non textbook because of the combined eq ( passive) and filtering ( active ) . The pa2 is very flexible where you can have different slopes for HP and LP filtering, at different frequencies. A lot of measurements where done to mirror my reference, at first my monitor jbl 305 ( easily surpassed ) and later a Genelec 8330 that I couldnt beat soundwise.Thank you.
I recall you mentioned this before and I agree, it is unfortunate. One (or both) of the drivers appears to have an issue by your description. Without seeing the data, I would suspect the one with the 4.3kHz peak is the issue since the driver should have peak at 4.8kHz according to Seas' specs. For example, I have four of the W18E001 v1 and they all measure the same in every way including the surface breakup mode at 4.3kHz even after 20 years of use in multiple cabinets. And I showed they are matched to Seas' datasheet too (see above!). And, I have two newer W18E001 v2 which also have resonance matched to each other and to Seas's spec even though one of them has a delaminated surround. I needed to look at the distortion measurement to see that issue! And the issue was bad enough to hear the distortion due to the surround flapping around (again, data posted earlier in this thread!). Regarding your situation, one of your drivers is likely broken.
I agree that that passive is harder than DSP for a DIY, hobbyist, or casual builder. Right now, my knees are hurting from crawling around underneath the test baffle swapping out passive filters. I built that baffle so I didn't have to swap out drivers in the FFLOTSAM cabinets to keep them looking OK with all of this fiddling! Even though I built a fixture for the LCR notch filter, I still take a few minutes to swap out components to change the filter. So yeah, passive takes minutes (or more) and effort to implement a change that takes a just a moment and a few keystrokes with DSP. If you are doing DIY, or fiddling around with measurements like I am, then yes passive can be a pain. It would be a different type of problem if you were a speaker manufacturer and had to mitigate large variability like you describe in the production line. But, the thing is, by your description you got a broken driver. The mitigation a large manufacturer would take in your case isn't custom filters, it's remove from production and return to vendor for replacement. I agree with you strongly about how easy DSP is for the average user, but the point you make using the defective driver is not a valid argument. This is an example of where you depart from objective and reasonable. Lastly, a speaker manufacturer will have a much different perception of how much of a 'pain' passives are than you or me. Especially, passives actually require you have a basic understanding of circuits, components, and some prototyping/fabricating skills. Speaker manufacturers have these competences, we less so.
I don't know how you can make a sweeping generalization like 'thought that the sound was better' even if you had two drivers that were working the same. With one of drivers having such a defective response compared to the other, your argument doesn't make sense. On the flip-side, some months ago, you were making arguments for hyper-matched capacitors and inductors having really big impacts on (for instance) Left/Right balance of speakers. I recall showing you the actual impact of components that are within tolerance in real-world passive crossover filters was very small. It is odd that you are using an obviously out of tolerance L/R woofer mismatch as the basis of this argument today, it's really upside down...
No, there are equations. Requires some basic circuit skills and speaker manufacturers don't struggle with that, at least the good ones. I don't usually either. DSP just turns the equations into sliders in a GUI so the lay person can easily operate in moments. And I do love those sliders!! Yes, DSP filter flexibility is unmatched by passives, but that's not what I'm testing here. I am trying to find out more about the distortion reduction observed in the PuriFi whitepaper. What PuriFi showed was a very tangible reduction in distortion with a passive notch, and they discussed the physical mechanism behind that distortion reduction. And, given their paper, it seemed worth exploring. And, I am finding that regarding distortion, PuriFi is right. And that the distortion reduction isn't just at the fundamental frequency, but throughout the audible band with interesting harmonics. I also think it matters now especially since PuriFi, Seas, Scan Speak, etc. have so many ultra-low distortion drivers. If you look at the published frequency responses, many of them they have resonant breakup modes. Here are a few Scan Speak:
View attachment 239651
I tried to get the x-axes to somewhat align... But you can see a theme in all of these drivers. Resonant peaks at 3-5kHz, odd uneven zone extending down to 1kHz or so.
Browsing datasheets, Scan Speak is on the low side for resonant peaks. Seas Excel line is perhaps has more dramatic, but moved to higher frequencies:
Or maybe an Accuton C168 7"ceramic:
I do find occasional driver with ruler flat response and no evidence of resonance, but those aren't the norm. It's all part of a tradeoff. These drivers all seem to use ultra-rigid cones with minimal mechanical damping these days Things like standing wave resonances on the cone's surface, non-linearities in the surround, etc. become observable among all of the other non-linearities of the driver if you don't coat it with goo, or fuzzy stuff, etc...
You were chasing ghosts with that bad woofer. But yes, a steeper slope is going to help with the resonance. A lower crossover frequency also helps. With your bad driver's resonance pushed down by another 500Hz, these would help even more.
I don't know that the drivers used in Genelec are intrinsically flat. Your conclusions about driver resonances and associated difficulties seem to be deeply convoluted with having a driver that measured way out of tolerance. As I reported above, I had an issue with one of my Seas woofers. I just got the replacement in the mail today, I recommend you contact your seller and/or Seas, you should have the measurements ready for them. As you can see above, Seas/Madisound replaced my bad driver as soon as I contacted them.
This is a different topic, not relevant to the passive notch and reduction in distortion that I am investigating here...
I like the look of the finished speaker. Can I ask, what is the network you show?
Yes, 20 or 30 years ago I was really looking hard at Marsh's MTM speakers. I built the Thors instead, which were part of that craze. When I first made the Thor active, I kit-bashed a pair of Linkwitz' Orion boards as the filters, first using SL's active notch. Later I moved to the series notch, but did the rest of the filtering active. I kept the passive notch when I later converted the Thors to DSP-active. I will say I thought it better to electro-mechanically damp the resonance at the motor rather than just try to feed it less energy at the resonance peak, but until I recently read the PuriFi paper I didn't fully appreciate the benefit of the notch in series with the driver. In any case, thanks for the kind words and interest in this!Hey, thanks so much for your experiments and analysis. I don't think this was fully appreciated(that the passive notch could electrically damp the bell resonance by interacting with the back emf) long ago when SL was designing his Orion and LX521, and there was a bit of a craze in 8" metal cones matched to dome tweeters low. (jonmarsh's m8ta and my RS22528A).
Yeah, I know, we're going way back... I've taken sort of a 10 year hiatus from speakerbuilding. But, as life changes, I have a bit more time to spend in this direction. Anyway, while in some respects, very little has changed in speaker design over the last 10-15 years, some things have.
And thanks to the Purifi link. Everything Bruno Putzeys touches seems to turn to gold...
So yeah, there are advantages, it seems, to passively notching cone resonances even if you go active!
markk
Impressive and very interesting, very well done!Putting together a speaker:
I thought I would integrate the woofer and tweeter, and make some spinoramas as descried by @napilopez in this fantastic tutorial.
I am using the W18W001 v1 drivers. I might try the v2 later, since they are different.
First, I applied crossovers at 2.4kHz, 24 dB/octave.
I made a passive filter that had nearly identical performance as the DSP notch:
View attachment 239845
This filter is on the lower side for L, as noted earlier when L is increased there is incrementally lower distortion, but with the penalty of higher DCR and the need to re-equalize the response with a DSP shelf.
I made a turntable out of a camera tripod. Angle is read off of graduations on the pan-head. Vertical measurements are done with the speaker sideways on the head. The corner of my basement is not ideal, but I am looking for relative differences between two speaker configurations, so should be OK.
View attachment 239844
I did a full set of measurements in 10 degree increments. Looks like mismatched DI between tweeter and woofer leads to jagged in-room response at the crossover frequency. The important thing for this investigation, not much difference between passive and DSP notch.
View attachment 239873
Comparing in-room responses of the two filters directly shows almost perfect matching. I think you will not be able to tell the difference between these two speaker's responses, or flaws!
View attachment 239875
Horizontal and vertical plots for the two notch-filter types:
View attachment 239878
Looking at these for differences, I started trippin' pretty hard. They are super close to identical. I need to take a walk and clear my head. What differences there are likely approach my ability to make repeatable measurements.
We observed a distortion reduction with the passive notch when we measured the woofer both at the 4.3kHz peak, and at 1.7kHz here a peak in 3rd order distortion existed. How about the speaker's distortion?
View attachment 239880
At the 4.3kHz resonant peak of the v1 driver where the passive outperformed the notch for distortion on the bare woofer, no difference can be observed in the speaker's performance. Perhaps if a higher crossover frequency or shallower slope is used, that difference in distortion would become measurable. It is not measurable with 24dB/octave filter slopes at 2.4kHz. However, the reduction in distortion at 1.7kHz is measurable on the speaker since it is well below the woofer's crossover point. I'm not sure this reduction in distortion would be audible under most conditions, but it is there.
Summing up, the two speaker configurations are close to identical. The benefits of the passive notch's distortion reduction are small, but perhaps are the largest of all the differences. Higher values of inductance for the filter might further improve the distortion. Either way, the speaker is listenable, and certainly plays very loud without distorting with really awesome dynamics. Combined with the JETSAM subwoofers and all of those amps, it is thrilling. The MiniDSP Flex Eight made all of the filters so easy to implement and simplified swapping between the passive and DSP notches. Next, I think I need to understand directivity better.
Thanks to @thewas for getting this conversation started. And thanks to @napilopez for the tremendous tutorial that got me started with VituixCAD (donate!)
MAB - thanks for testing.Putting together a speaker:
I thought I would integrate the woofer and tweeter, and make some spinoramas as descried by @napilopez in this fantastic tutorial.
I am using the W18W001 v1 drivers. I might try the v2 later, since they are different.
First, I applied crossovers at 2.4kHz, 24 dB/octave.
I made a passive filter that had nearly identical performance as the DSP notch:
View attachment 239845
This filter is on the lower side for L, as noted earlier when L is increased there is incrementally lower distortion, but with the penalty of higher DCR and the need to re-equalize the response with a DSP shelf.
I made a turntable out of a camera tripod. Angle is read off of graduations on the pan-head. Vertical measurements are done with the speaker sideways on the head. The corner of my basement is not ideal, but I am looking for relative differences between two speaker configurations, so should be OK.
View attachment 239844
I did a full set of measurements in 10 degree increments. Looks like mismatched DI between tweeter and woofer leads to jagged in-room response at the crossover frequency. The important thing for this investigation, not much difference between passive and DSP notch.
View attachment 239873
Comparing in-room responses of the two filters directly shows almost perfect matching. I think you will not be able to tell the difference between these two speaker's responses, or flaws!
View attachment 239875
Horizontal and vertical plots for the two notch-filter types:
View attachment 239878
Looking at these for differences, I started trippin' pretty hard. They are super close to identical. I need to take a walk and clear my head. What differences there are likely approach my ability to make repeatable measurements.
We observed a distortion reduction with the passive notch when we measured the woofer both at the 4.3kHz peak, and at 1.7kHz here a peak in 3rd order distortion existed. How about the speaker's distortion?
View attachment 239880
At the 4.3kHz resonant peak of the v1 driver where the passive outperformed the notch for distortion on the bare woofer, no difference can be observed in the speaker's performance. Perhaps if a higher crossover frequency or shallower slope is used, that difference in distortion would become measurable. It is not measurable with 24dB/octave filter slopes at 2.4kHz. However, the reduction in distortion at 1.7kHz is measurable on the speaker since it is well below the woofer's crossover point. I'm not sure this reduction in distortion would be audible under most conditions, but it is there.
Summing up, the two speaker configurations are close to identical. The benefits of the passive notch's distortion reduction are small, but perhaps are the largest of all the differences. Higher values of inductance for the filter might further improve the distortion. Either way, the speaker is listenable, and certainly plays very loud without distorting with really awesome dynamics. Combined with the JETSAM subwoofers and all of those amps, it is thrilling. The MiniDSP Flex Eight made all of the filters so easy to implement and simplified swapping between the passive and DSP notches. Next, I think I need to understand directivity better.
Thanks to @thewas for getting this conversation started. And thanks to @napilopez for the tremendous tutorial that got me started with VituixCAD (donate!)
Thank you very much!very good tread true quality content thankyou
Regarding THD i saw an experiment done in some hifi magasine 30 years ago with driving woofers with an amp with very high output impedance practically infinite.
And it changed the distortion profile and the level of thd was lower , obviusly the fr was completely off and had to be compensated.
I don't have the skills to assess this, but it could be interesting
This is different than what I am doing. You applied an inductor as a first order crossover element with a cutoff frequency of 2.4kHz with your ER18 woofer, I am using an RLC network which is a notch, with resonance at 4.3kHz. Your language is vague so I am not clear on what you mean by "compensated". But, I already know where you are going with this:MAB - thanks for testing.
You can get the same amount of slightly, tiny, less distortion with just a small coil in series with the driver. But after testing this, and compensated the two active dsp hybrid crossovers ( one with a 0,4 mH coil and one without ) to be the same,
What you thought you heard has no bearing especially here without proper experiment, this is ASR not Cable-Lifter-Magic.com. And, (once again) you provide no measurements, just words so you get no belief from me. Assuming you compensated in DSP for the first-order crossover formed by the inductor, you are saying to us that you can hear the difference between a first order crossover implemented with DSP vs. one implemented with passives. I reject this unless you provide a scientific study.I thought that the sound from the speaker was better without the coil .
Static signals...Theres probably more to the sound than only measuring harmonic distortion with static signals.
Neumann uses different output impedances for different drivers with their more expensive models, it seems.I wonder, with the increasing number of active speakers, how many brands that use more non-conventional amplification, e.g. transconductance, ace-bass etc?